Daily Research Digest
arXiv Papers
2026-09-17
385
Papers
9
Categories
66
Translated
收藏清单 0
精选 · Favorites
66
cs.AI / 1 / 2609.18598
Hypothesis-Driven Autonomous Materials Synthesis with Multimodal LLM Agents
假设驱动的多模态大语言模型智能体自主材料合成
large language model
大语言模型相关
Abstract
Self-driving laboratories can explore synthesis conditions autonomously, but their decision-making layer is typically a black-box optimizer, and the output is a set of optimized samples, with the measurements reduced to predefined scalar objectives and the reasons behind success left unarticulated. Here we present SynAgent, a framework in which large language model agents operate an automated experimental system and maintain an explicit, revisable understanding of the synthesis process as the campaign's primary output. Starting with no predefined analysis pipeline, SynAgent adaptively generates analysis skills for newly acquired data and evolves this understanding through multimodal reasoning over experimental data such as X-ray diffraction patterns and electron micrographs. The evolution is guided by a verify-falsify scheme, in which the agent deliberately challenges its own hypotheses by testing conditions predicted to fail as well as those predicted to succeed. In a single campaign of 18 autonomous experiments using LiCoO2 (001) thin-film deposition as a testbed, SynAgent synthesized highly crystalline films and evolved an understanding of how the substrate temperature governs crystallization, discovering an abrupt threshold and a narrow optimal growth window at 650-690 °C. These results extend autonomous experimentation beyond optimized samples to testable, human-readable understanding.
Chinese Translation
自驱动实验室能够自主探索合成条件，但其决策层通常是黑箱优化器，输出是一组优化后的样品，测量结果被简化为预先定义的标量目标，而成功背后的原因则未被阐明。在此，我们提出了 SynAgent，这是一个大语言模型智能体操作自动化实验系统、并将对合成过程的显式且可修订的理解作为实验活动主要输出的框架。SynAgent 在没有预定义分析流程的情况下起步，针对新获取的数据自适应地生成分析技能，并通过对 X 射线衍射图谱和电子显微图像等实验数据的多模态推理来演化这一理解。这一演化由验证—证伪方案引导，在该方案中，智能体通过测试被预测为失败的条件以及被预测为成功的条件，有意挑战其自身的假设。在以 LiCoO2 (001) 薄膜沉积为试验平台的单次包含 18 个自主实验的活动中，SynAgent 合成了高结晶度的薄膜，并演化出关于衬底温度如何调控结晶的理解，发现了一个突变的阈值以及位于 650-690 °C 的狭窄最优生长窗口。这些结果将自主实验拓展到了优化样品之外，迈向可检验、人类可读的理解。
cs.AI / 2 / 2609.17695
GraphEcho: Structural Redundancy and Evidence Provenance in LLM Graph Agents
GraphEcho：LLM 图智能体中的结构冗余与证据来源
large language model
大语言模型相关
Abstract
A large language model (LLM) agent can follow more graph paths without acquiring more independent evidence. GraphEcho tests whether agents mistake these repeated encounters for additional corroboration. The benchmark varies path counts and evidential origins while holding evidence content fixed, and evaluates both judgments and active exploration. Controlled synthetic experiments reveal model-dependent judgment shifts, but redundant supporting paths increase the share of repeated walks across all evaluated frozen agents. Provenance-aware post-training (PAPT) reduces revisits and improves synthetic accuracy, yet covers fewer distinct sources. On scientific claims, it continues to reduce repetition while accuracy declines. These findings expose a gap between efficient exploration and effective evidence use: an agent can learn to stop repeating itself while overlooking information it needs. GraphEcho provides a controlled way to evaluate both what graph agents conclude and whether their exploration reaches distinct evidential sources.
Chinese Translation
大型语言模型（LLM）智能体可以沿着更多的图路径行进，却并未获得更多独立证据。GraphEcho 检验智能体是否会把这些重复遭遇误认为是额外的佐证。该基准在保持证据内容固定的同时改变路径数量和证据来源，并同时评估判断与主动探索。受控的合成实验揭示了依赖于模型的判断偏移，但冗余的支持路径会提高所有被评估的冻结智能体中重复行走所占的比例。来源感知的后训练（PAPT）减少了重复访问并提升了合成任务上的准确率，但覆盖的不同来源更少。在科学主张上，它继续减少重复，而准确率却下降。这些发现揭示了高效探索与有效利用证据之间的差距：智能体可以学会不再重复自身，却忽略了它所需的信息。GraphEcho 提供了一种受控的方式，用以评估图智能体得出什么结论，以及它们的探索是否抵达不同的证据来源。
cs.AI / 3 / 2609.17696
GVD: Governed Versioning and Deduplication for Document Repositories
GVD：面向文档库的受治理版本管理与去重
large language model
大语言模型相关
Abstract
Document repositories evolve continuously. Guidelines and policies are revised, superseded, and re-uploaded, so the same content recurs in different wording and newer versions refine or contradict earlier ones. These inconsistencies belong to the growing collection rather than to any single document, yet existing work treats versioning, duplicate detection, and contradiction detection as isolated pairwise tasks and stops once a pair is labeled. We present GVD (Governed Versioning and Deduplication), a framework that unifies cross-document version linking with rule-level conflict resolution under an auditable update policy. Incoming documents are assigned to version families through bidirectional rule alignment, and their rules are compared against the family memory to identify duplicates, contradictions, asymmetric refinements, and new knowledge, with Counterfactual Span Probing (CSP) resolving related pairs that inference misclassifies as neutral. Relation-specific policies suppress duplicates and escalate only consequential changes for review, retaining version lineage as an audit trail. The pipeline runs fully locally, with no large language model. On 120 enterprise documents processed as 140 ingestions across 59 version families, GVD reaches an F1 of 0.97 for version-family construction and 0.94 for rule-level consistency, with CSP raising rule consistency from 0.90 to 0.94.
Chinese Translation
文档库持续演化。指南与政策会被修订、取代并重新上传，因此相同内容会以不同的措辞反复出现，而更新的版本则对较早版本加以细化或与之相矛盾。这些不一致性属于不断增长的集合整体，而非任何单一文档，然而现有工作将版本管理、重复检测与矛盾检测视为彼此孤立的成对任务，并且一旦对某一对文档打上标签便就此停止。我们提出 GVD（Governed Versioning and Deduplication，受治理的版本管理与去重），这是一个在可审计的更新策略下将跨文档版本链接与规则级冲突消解统一起来的框架。新传入的文档通过双向规则对齐被分配到版本家族中，其规则会与家族记忆进行比较，以识别重复、矛盾、非对称细化和新知识，其中反事实片段探测（Counterfactual Span Probing，CSP）用于解决被推理错误分类为中立的相关文档对。针对不同关系的策略会抑制重复项，并仅将有实质性影响的变更升级以供审核，同时保留版本谱系作为审计追踪。该流水线完全在本地运行，不使用任何大语言模型。在 120 份企业文档（以 140 次摄入处理、跨越 59 个版本家族）上，GVD 在版本家族构建上达到 0.97 的 F1，在规则级一致性上达到 0.94，而 CSP 将规则一致性从 0.90 提升至 0.94。
cs.AI / 4 / 2609.17804
A Four-Stage Decomposition of Word-Problem Solving and Mechanistic Fragility in LLM Math Reasoning
LLM 数学推理中应用题求解的四阶段分解与机制脆弱性
large language model
大语言模型相关
Abstract
Large language models solve grade-school math word problems with high accuracy, yet a single irrelevant clause inserted into the problem can collapse it. We reconcile these observations with a mechanistic account. We show that the model's internal computation decomposes into a four-stage sequential pipeline, Schema Abstraction, Operation Planning, Operand Binding, and Computation, each stage producing a distinct intermediate representation in an identifiable band of layers. Using the same scaffold to diagnose distractor-induced failure, we localize the corruption to a single stage, Operation Planning, implemented by a set of attention heads whose causal role we validate bidirectionally. In short, we provide a mechanistic interpretation of math word problem reasoning in LLMs, and their failure when distracted.
Chinese Translation
大型语言模型能以高准确率求解小学阶段的数学应用题，然而只要在题目中插入一个无关分句，就可能使其崩溃。我们通过一种机制性解释来调和这些观察结果。我们表明，模型的内部计算分解为一条四阶段顺序流水线：图式抽象、运算规划、操作数绑定和计算；每个阶段都在一个可识别的层带中产生一种不同的中间表示。使用同一脚手架来诊断由干扰项引发的失败时，我们将损坏定位到单个阶段——运算规划，该阶段由一组注意力头实现，我们双向验证了这些注意力头的因果作用。简而言之，我们为 LLM 中数学应用题推理以及它们在受到干扰时的失败提供了一种机制性解释。
cs.AI / 5 / 2609.17987
Multimodal Conditioning of Fine-Tuned Stable Diffusion XL for Controllable and Culturally Faithful Ulos Motif Generation
面向可控且文化忠实的 Ulos 纹样生成的微调 Stable Diffusion XL 多模态条件控制
diffusionlarge language model
扩散模型相关
大语言模型相关
Abstract
The traditional Batak Ulos weaving industry faces growing challenges in producing diverse, innovative motifs due to limitations in conventional, manually driven design methods. This study proposes a multimodal generative framework integrating a fine-tuned Latent Diffusion Model (Stable Diffusion XL v1.0 via LoRA) with a Multimodal Large Language Model (LLaMA 1.5-7B) to enable controllable, culturally faithful Ulos motif generation. Four complementary conditioning mechanisms: text, image, representation, and semantic map (via ControlNet) jointly guide the generation process, each governing a distinct aspect from semantic intent to spatial layout. A five level ablation study across three scenarios (shape transformation, colour variation, and high-complexity input) shows that conditioning effectiveness is not proportional to the number of mechanisms combined: Text + Image + Semantic Map achieved the best FID (270) and CLIP Score (0.65 - 0.70) but the weakest SSIM (0.65), while Text + Image + Representation offered the best overall balance, with stable SSIM (0.84) and competitive FID (280). Combining all four mechanisms yielded the weakest FID (330), indicating conflicting optimization signals. Qualitative evaluation by nine weavers and thirty public participants confirmed statistically significant positive acceptance (Wilcoxon, p=0.007 and p<0.001, respectively). A web-based prototype supporting text-to-image and image-to-image generation was also developed, offering a practical digital design tool for cultural heritage preservation.
Chinese Translation
传统的巴塔克（Batak）Ulos 织造产业由于传统、人工驱动的设计方法的局限，在生成多样、创新的纹样方面面临日益严峻的挑战。本研究提出一个多模态生成框架，将经过微调的潜在扩散模型（通过 LoRA 微调的 Stable Diffusion XL v1.0）与多模态大语言模型（LLaMA 1.5-7B）相结合，以实现可控且文化忠实的 Ulos 纹样生成。四种互补的条件控制机制：文本、图像、表征和语义图（通过 ControlNet），共同引导生成过程，每一种机制掌控着从语义意图到空间布局的一个不同方面。在三种场景（形状变换、颜色变化和高复杂度输入）下进行的五级消融研究表明，条件控制的有效性并不与所组合机制的数量成正比：文本 + 图像 + 语义图取得了最佳的 FID（270）和 CLIP 分数（0.65 - 0.70），但 SSIM 最弱（0.65），而文本 + 图像 + 表征提供了最佳的总体平衡，具有稳定的 SSIM（0.84）和具有竞争力的 FID（280）。将全部四种机制组合则产生了最弱的 FID（330），表明存在相互冲突的优化信号。由九名织工和三十名公众参与者进行的定性评估证实了在统计上显著的正向接受度（Wilcoxon 检验，分别为 p=0.007 和 p<0.001）。还开发了一个支持文本到图像和图像到图像生成的基于网页的原型，为文化遗产保护提供了一种实用的数字化设计工具。
cs.AI / 6 / 2609.18004
Missing Bridges: Composition-Aware Active Imitation Learning
缺失的桥梁：组合感知的主动模仿学习
diffusion
扩散模型相关
Abstract
Active imitation learning reduces expert effort by allowing a learner to request the demonstrations it needs. Existing methods typically select these requests for their expected information gain about the expert policy. In structured multi-task domains, however, the number of start-goal tasks may grow combinatorially despite their solutions sharing reusable behavior. This makes composable behaviors especially valuable, since a single demonstration may help solve many tasks at once. Prior methods do not explicitly account for this value when selecting which demonstration to request. We introduce Adaptive Agents via Latent Topologies (AALT), which requests demonstrations that maximize expected gains in start-goal connectivity. We further show that this objective is formally tied to information gain about task reachability. AALT organizes existing demonstrations into a topology of latent hub states connected by learned behaviors, identifies high-value bridge demonstrations that are likely to enable many tasks at once, and grounds each to an expert query. At inference, it plans through the resulting topology and conditions a diffusion policy on each successive hub transition. In a simulated UR5e robot ordered-retrieval domain with 72 tasks, AALT improved from 42/72 to 72/72 (100%) successful tasks consistently using only 3 demonstrations totaling 5 transitions beyond the initial dataset. After 20 demonstrations, the strongest baseline averaged 88.6% success using 98 transitions.
Chinese Translation
主动模仿学习通过允许学习器请求其所需的演示，减少了专家的付出。现有方法通常根据这些请求对专家策略的预期信息增益来选择请求。然而，在结构化多任务领域中，起点-目标任务的数量可能会组合式增长，尽管它们的解共享可复用的行为。这使得可组合行为尤其有价值，因为单个演示可能有助于一次性解决许多任务。先前方法在选择请求哪个演示时并未明确考虑这一价值。我们提出通过潜在拓扑的自适应智能体（Adaptive Agents via Latent Topologies, AALT），它请求能够最大化起点-目标连通性预期增益的演示。我们进一步表明，该目标在形式上与关于任务可达性的信息增益相关。AALT将现有演示组织为由所学行为连接的潜在枢纽状态的拓扑，识别可能一次性使许多任务变得可行的高价值桥梁演示，并将每个这样的演示落实为一次专家查询。在推理时，它通过所得拓扑进行规划，并在每个连续的枢纽转移上对扩散策略进行条件化。在一个具有72个任务的模拟UR5e机器人有序取回领域中，AALT在初始数据集之外仅使用3个演示、总共5次转移，便始终从42/72提升到72/72（100%）成功任务。在20个演示后，最强基线平均达到88.6%成功率，使用了98次转移。
cs.AI / 7 / 2609.18123
AutoTuneBench: Trustworthy Measurement for Agent Auto-Tuning of LLM Serving Engines
AutoTuneBench：面向 LLM 服务引擎智能体自动调优的可信测量
large language model
大语言模型相关
Abstract
Large language model agents tune GPU kernels and serving engines through a closed loop of propose, measure, and keep, but the measurements behind this loop are not trustworthy. We characterize four failure modes from a four-day pilot corpus of 619 model calls: strawman baselines manufacture speedups, absolute times do not transfer across machines, saturated tasks nullify comparisons, and infrastructure defects impersonate science. We present AutoTuneBench, a benchmark and measurement protocol that makes trust architectural. The protocol is frozen as code with test-enforced provenance; a database-level validator rejects out-of-protocol results; anti-cheat checks run outside the agent's modification surface; comparisons follow pre-registered readouts; and measurements anchor to externally published results, grounded in paired-seed statistics with a 5\% cross-run coefficient-of-variation cap. Honest measurement rewrites the headlines: our best kernel reads 10.6x against a naive baseline but 2.03x against the honest one; one configuration delivers 1.174x on one machine and 1.0049x on another; a pre-registered on/off comparison nulls at a shared wall (2.4840 vs 2.4957\,ms); and the KernelBench Level-1 suite admits 51\% of tasks with median speedup 1.0001x over PyTorch eager. The protocol, the two-engine corpus (vLLM and SGLang), and its audit trail are released as open artifacts.
Chinese Translation
大语言模型智能体通过由“提出、测量和保留”构成的闭环来调优 GPU 内核与服务引擎，但这一闭环背后的测量并不可信。我们从一个为期四天、包含 619 次模型调用的试点语料库中刻画了四种失败模式：稻草人基线制造出加速比，绝对时间无法跨机器迁移，饱和任务使比较失效，基础设施缺陷冒充科学。我们提出 AutoTuneBench，一个将信任架构化的基准与测量协议。该协议被冻结为代码，并具有由测试强制执行的来源追踪；数据库级验证器会拒绝协议外的结果；反作弊检查在智能体修改面之外运行；比较遵循预先注册的读数；测量锚定到外部已发布结果，并以配对种子统计为基础，同时受 5\% 的跨运行变异系数上限约束。诚实的测量改写了头条结论：我们最好的内核相对于朴素基线读数为 10.6x，但相对于诚实基线为 2.03x；一个配置在一台机器上带来 1.174x，在另一台机器上带来 1.0049x；一项预先注册的开/关比较在共享墙处归零（2.4840 对 2.4957\,ms）；而 KernelBench Level-1 套件接纳了 51\% 的任务，其相对于 PyTorch eager 的中位加速比为 1.0001x。该协议、双引擎语料库（vLLM 和 SGLang）及其审计追踪均作为开放工件发布。
cs.AI / 8 / 2609.18128
Symbolic Temporal Supervision of LLM Agents Using Contracts
使用契约对LLM智能体进行符号时序监督
large language model
大语言模型相关
Abstract
Large language model (LLM) agents augmented by tools can automate complex, multi-step tasks, such as web navigation, code generation, and workflow orchestration, by acting on external systems through tool calls. However, hallucinations, distributional instability, and adversarial manipulations in LLMs, and the irreversible consequences of certain tool calls can lead to harmful outcomes. Existing safeguards either grade recorded trajectories post hoc with stochastic LLM judges or block unsafe actions one call at a time, and no single deterministic artifact supports both roles. We present ContrAgent, a contract-based framework for symbolic temporal supervision of LLM agents. ContrAgent captures an agent's behavior as a sequence of tool calls and formalizes it as a trace over a fixed set of checkable predicates. It then specifies required behaviors using assume-guarantee contracts in linear temporal logic over finite traces (LTLf). Each contract is compiled to a deterministic finite automaton (DFA) that serves two roles: gating agent actions online and evaluating recorded traces offline. A contract library, acting as a reusable knowledge base, is maintained independently of the agent's model and can be applied across different agents within the same task domain. We show the effectiveness of our approach on four benchmarks spanning both roles, where ContrAgent matches state-of-the-art LLM-judge and rule-based guardrail baselines while producing deterministic, reproducible verdicts and, in the online mode, orders-of-magnitude lower per-call latency.
Chinese Translation
由工具增强的大型语言模型（LLM）智能体可以通过工具调用作用于外部系统，从而自动化复杂的多步任务，例如网页导航、代码生成和工作流编排。然而，LLM中的幻觉、分布不稳定性、对抗性操纵，以及某些工具调用的不可逆后果，可能导致有害结果。现有防护措施要么使用随机LLM评判器对已记录轨迹进行事后评分，要么逐个调用来阻止不安全动作，并且没有任何单一的确定性制品能同时支持这两种角色。我们提出ContrAgent，一个用于对LLM智能体进行符号时序监督的基于契约的框架。ContrAgent将智能体的行为捕获为一系列工具调用，并将其形式化为在一组固定可检查谓词上的迹。然后，它使用有限迹上的线性时序逻辑（LTLf）中的假设-保证契约来指定所需行为。每个契约都被编译为确定性有限自动机（DFA），其承担两种角色：在线门控智能体动作，以及离线评估已记录轨迹。契约库充当可复用知识库，独立于智能体模型进行维护，并且可以在同一任务领域内跨不同智能体应用。我们在涵盖这两种角色的四个基准上展示了我们方法的有效性；其中，ContrAgent与最先进的LLM评判器和基于规则的护栏基线相匹配，同时产生确定性、可复现的判定结果，并且在线模式下每次调用的延迟低数个数量级。
cs.AI / 9 / 2609.18270
BENCHCOMPASS: From Scores to Signals for Training and Harness Decisions in Payment-Domain LLMs
BENCHCOMPASS：从分数到信号，用于支付领域大语言模型的训练与 Harness 决策
large language model
大语言模型相关
Abstract
Payment operations are a critical financial infrastructure, but the value of large language models in this domain remains unclear because payment rules change quickly, evidence is fragmented, and decisions depend on transaction state, participant role, region, and payment rail. Existing benchmarks do not isolate whether failures come from missing payment-rule knowledge, poor use of supplied evidence, or brittleness under imperfect harness inputs. We introduce BENCHCOMPASS, a payment-domain benchmark whose construction pipeline builds scenario-grounded tasks from typed evidence packs, applies LLM-based quality checks, creates task-input attack variants, and reserves final item admission for domain experts. The release contains an expert-reviewed Pro benchmark covering payment knowledge, context-grounded scenario reasoning, and Attacked Open robustness, plus a lower-assurance Normal pool for inspection and future curation. Across 16 model variants, BENCHCOMPASS shows qualitatively different failure modes: missing parametric payment knowledge, incomplete reasoning over supplied rules, and failure to reject plausible but invalid workflows. The benchmark remains unsaturated: the best frontier model reaches 89.6% on Open Context-Grounded Reasoning and 81.7% under attacked inputs, while a representative 32B open-weight model reaches 69.8% and 42.6%. Benchmark data and code are available at https://github.com/ant-intl/BenchCompass.
Chinese Translation
支付运营是一项关键的金融基础设施，但大语言模型在这一领域的价值仍不明确，因为支付规则变化迅速、证据碎片化，且决策取决于交易状态、参与者角色、地区和支付通道。现有的基准测试无法区分失败究竟源于缺失的支付规则知识、对所提供证据的使用不当，还是在不完美 harness 输入下的脆弱性。我们提出 BENCHCOMPASS，一个支付领域基准，其构建流水线从带类型的证据包构建有场景依据的任务，应用基于大语言模型的质量检查，创建任务输入的对抗攻击变体，并将最终条目的准入留给领域专家。该发布包含一个经专家评审的 Pro 基准，覆盖支付知识、基于上下文的场景推理以及 Attacked Open 鲁棒性，另有一个保证程度较低的 Normal 池，用于检查和未来策展。在 16 个模型变体上，BENCHCOMPASS 显示出性质不同的失败模式：缺失参数化支付知识、对所提供规则的推理不完整，以及未能拒绝看似合理但无效的工作流。该基准仍未饱和：最佳前沿模型在 Open Context-Grounded Reasoning 上达到 89.6%，在受攻击输入下达到 81.7%，而一个具有代表性的 32B 开放权重模型则达到 69.8% 和 42.6%。基准数据与代码可在 https://github.com/ant-intl/BenchCompass 获取。
cs.AI / 10 / 2609.18283
Where Should Agents Live? Energy-Memory Characterization of Agentic AI for the Edge-Cloud Continuum
智能体应驻留于何处？面向边缘-云连续体的智能体AI的能量-内存表征
large language model
大语言模型相关
Abstract
As telecommunication networks evolve toward autonomous 5G-Advanced and 6G operations, agentic artificial intelligence (AI) workflows, where large language models (LLMs) execute multi-step reasoning, invoke diagnostic tools, retrieve domain knowledge, and coordinate across agent teams, are increasingly embedded across the edge-cloud continuum. While the biological brain accomplishes complex cognition on an exceptionally modest metabolic power budget of approximately 20W contemporary LLMs are profoundly energy- and memory-intensive, making sustainable lifecycle orchestration a critical operational priority. However, existing AI lifecycle metrics evaluate only isolated, single-model inferences or overlook multi-agent execution graphs entirely. Consequently, network operators lack foundational models to determine whether distributed agent communication incurs meaningful energy costs and where across edge-cloud tiers agent teams should physically reside. To address this gap, we introduce agentic-eCAL, generalizing the Energy Cost of AI Lifecycle (eCAL) metric to directed multi-agent workflows by coupling a closed-form two-rate single-call energy model (compute-bound prefill and memory-bound decode) with 7-layer OSI data transport. Grounded in hundreds of GPU benchmark configurations on NVIDIA A100 and H100, 16 open-weight models and 8 orchestration topologies, we validate components of the metric and study workflow placement implications. Our findings demonstrate that inter-agent text transport incurs 0.25% of workflow energy across 5G RAN, metro, and optical links. Therefore in edge-cloud agent placement the dominant energy cost of distribution is often not the transmission of inter-agent text itself, but the additional inference and context processing induced by that communication.
Chinese Translation
随着电信网络向自主的5G-Advanced和6G运营演进，智能体人工智能（AI）工作流——其中大语言模型（LLM）执行多步推理、调用诊断工具、检索领域知识，并在智能体团队之间进行协调——正日益嵌入到边缘-云连续体中。尽管生物大脑以约20W的极低代谢功率预算完成复杂认知，但当代LLM却极其能耗密集和内存密集，这使得可持续的生命周期编排成为关键的运营优先事项。然而，现有的AI生命周期指标仅评估孤立的单模型推理，或完全忽略多智能体执行图。因此，网络运营商缺乏基础模型来确定分布式智能体通信是否会带来显著的能量成本，以及智能体团队在边缘-云各层级中应物理驻留在何处。为弥补这一空白，我们引入agentic-eCAL，通过将闭式双速率单次调用能量模型（计算受限的预填充和内存受限的解码）与7层OSI数据传输相结合，将AI生命周期能量成本（eCAL）指标推广到有向多智能体工作流。基于NVIDIA A100和H100上的数百个GPU基准配置、16个开放权重模型和8种编排拓扑，我们验证了该指标的组成部分，并研究了工作流放置的影响。我们的研究结果表明，跨5G RAN、城域和光链路，智能体间文本传输仅产生工作流能量的0.25%。因此，在边缘-云智能体放置中，分布的主要能量成本通常不是智能体间文本本身的传输，而是该通信所引发的额外推理和上下文处理。
cs.AI / 11 / 2609.18286
What Counts as Strategic Reasoning? A Systematic Mapping of Chess Research on Humans, Engines, and Language Models
什么才算战略性推理？对关于人类、引擎和语言模型的国际象棋研究的系统性映射
large language model
大语言模型相关
Abstract
Chess has long served as a model domain for studying search, expertise, decision-making, and artificial intelligence. The emergence of large language models (LLMs) has renewed the relevance of chess as a controlled environment for investigating strategic reasoning and comparing human and artificial decision-making. We present a systematic mapping study of recent research spanning human players, classical chess engines, neural and reinforcement-learning systems, LLMs, and hybrid approaches. The final map comprises 84 core study families, classified according to agent type, strategic-reasoning stages, and evaluation dimensions. The map reveals a literature strongly concentrated on situation assessment, evaluation, and action selection, while explicit planning, explanation, metacognition, and human--AI collaboration remain less explored. LLM research places particular emphasis on state representation and generalization, whereas grounded explanation appears more frequently in hybrid approaches combining language models with engines, expert knowledge, or other external structures. Two distinctions emerge that the map aggregates rather than resolves: hybrid systems differ in where and when heterogeneous capabilities combine, and evaluations that show improved human performance do not thereby establish human--AI synergy. We propose both as extensions of the mapping framework. We argue that chess provides a useful bridge between cognitive and computational perspectives on strategic reasoning, and identify explicit planning, grounded and faithful explanation, metacognitive calibration, and human--AI complementarity as directions for future research.
Chinese Translation
国际象棋长期以来一直充当研究搜索、专业技能、决策与人工智能的模型领域。大语言模型（LLM）的出现，使国际象棋作为一种用于考察战略性推理并比较人类与人工决策的受控环境，重新获得了重要性。我们呈现了一项系统性映射研究，涵盖人类棋手、经典国际象棋引擎、神经网络与强化学习系统、LLM 以及混合方法等近期研究。最终的图谱包含 84 个核心研究族，依据智能体类型、战略性推理阶段和评估维度进行分类。该图谱揭示出，相关文献强烈集中于局面评估、评估与行动选择，而显式规划、解释、元认知以及人机协作则仍较少被探索。LLM 研究尤为强调状态表示与泛化，而有据的解释则更频繁地出现在将语言模型与引擎、专家知识或其他外部结构相结合的混合方法中。浮现出两种区分，而该图谱只是将其聚合而非加以解决：混合系统在异质能力于何处、何时结合上存在差异，而显示人类表现有所提升的评估并不因此就确立了人机协同。我们提出将这两者作为该映射框架的扩展。我们认为，国际象棋在关于战略性推理的认知视角与计算视角之间提供了一座有用的桥梁，并指出显式规划、有据且忠实的解释、元认知校准以及人机互补性作为未来研究的方向。
cs.AI / 12 / 2609.18346
Faithful yet Collusive: Why Chain-of-Thought Monitoring Cannot Detect Collusion in LLM Pricing Agents under Oligopolistic Competition
忠实却合谋：为何思维链监控无法在寡头竞争下检测LLM定价代理中的合谋
large language model
大语言模型相关
Abstract
Large language models (LLM) deployed as autonomous pricing agents may sustain supracompetitive prices through tacit coordination. We develop a causal graph divergence framework that separately measures structural faithfulness and intent faithfulness of LLM pricing agents in Bertrand competition. Across nine LLMs under duopoly and triopoly conditions, collusive behavior and chain-of-thought (CoT) faithfulness dissociate along both dimensions: the most collusive model accurately reports cooperative intent yet reasons structurally unfaithfully, while the most structurally faithful model sustains supra-Nash pricing under both market structures. These findings establish that CoT monitoring alone cannot serve as a standalone safeguard against algorithmic collusion.
Chinese Translation
作为自主定价代理部署的大型语言模型（LLM）可能通过默契协调维持超竞争价格。我们开发了一个因果图分歧框架，该框架分别衡量伯特兰竞争中的LLM定价代理的结构忠实性和意图忠实性。在双寡头和三家寡头条件下对九个LLM的研究中，合谋行为与思维链（CoT）忠实性在这两个维度上都出现分离：最具合谋性的模型准确报告了合作意图，却在结构上进行不忠实的推理；而结构上最忠实的模型在两种市场结构下都维持超纳什定价。这些发现表明，仅靠CoT监控不能作为防止算法合谋的独立保障。
cs.AI / 13 / 2609.18357
Market Signal Injection: Adversarial Context Manipulation of LLM Pricing Agents
市场信号注入：对LLM定价代理的对抗性上下文操纵
large language model
大语言模型相关
Abstract
Large language model (LLM) pricing agents may respond to how market data is presented, even when its numerical values remain unchanged. We introduce market signal injection (MSI), an attack that manipulates numerical formatting, competitor ordering, or qualitative market commentary without issuing explicit instructions. We evaluate nine open-weight models in simulated Bertrand duopoly and triopoly markets and three proprietary models in duopoly markets. Sentiment-based attacks produce the largest behavioral shifts, which propagate to other firms and alter profits and consumer surplus. Susceptibility varies across model families, and larger models are not consistently more robust. Matched neutral-text controls and a rule-based agent support a framing-based account of these shifts under the fixed demand parameters of our simulation. Episode-held-out probes distinguish baseline from attacked activations in all eleven re-evaluated model--condition pairs: linear AUC is 1.00 and MLP AUC ranges from 0.93 to 0.99. This separability does not by itself identify harmful pricing decisions. Input canonicalization removes the tested sentiment attacks, while decision boundary anchoring, which combines prompt constraints with output projection, provides partial mitigation under the tested adaptive attacks. These results identify data presentation as an attack surface for LLM pricing agents and motivate defenses that account for interactions among agents.
Chinese Translation
大语言模型（LLM）定价代理可能会对市场数据的呈现方式作出反应，即使其数值保持不变。我们提出市场信号注入（MSI），这是一种在不发出明确指令的情况下操纵数值格式、竞争对手排序或定性市场评论的攻击。我们在模拟的伯特兰双寡头和三寡头市场中评估了九个开放权重模型，并在双寡头市场中评估了三个专有模型。基于情感的攻击产生了最大的行为转变，这些转变会传播到其他企业并改变利润和消费者剩余。易感性因模型家族而异，并且更大的模型并非始终更稳健。匹配的中性文本对照组和一个基于规则的代理支持了对这些转变的一种基于框架的解释，且是在我们模拟的固定需求参数下。留出回合的探针在所有十一个重新评估的模型–条件对中区分了基线激活与受攻击激活：线性AUC为1.00，MLP AUC范围为0.93至0.99。这种可分性本身并不能识别有害的定价决策。输入规范化消除了所测试的情感攻击，而决策边界锚定，其将提示约束与输出投影相结合，在所测试的自适应攻击下提供了部分缓解。这些结果将数据呈现方式确定为LLM定价代理的一个攻击面，并促使人们设计考虑代理之间交互的防御措施。
cs.AI / 14 / 2609.18394
Cultural Competence in Context: A Large Language Model Passes the Turing Test in Finland
语境中的文化能力：一个大型语言模型在芬兰通过了图灵测试
large language model
大语言模型相关
Abstract
We report the results of a Turing Test conducted in Finland in the Finnish language. Because languages and cultural contexts are unevenly represented in LLM training data, we expected the model (ChatGPT 5.2) to perform worse in a Finnish-language Turing Test than in previously studied English-language US contexts. We also present model-generated role prompting as a replicable technique for conducting comparative LLM-based Turing Tests designed to improve construct validity. Contrary to our expectations, the LLM passed the Finnish Turing Test. A prominent source of error was participants' reliance on linguistic cues, particularly colloquial Finnish, as markers of human authorship. We reframe the Turing Test from a test of intelligence to a comparative method for examining whether an AI system can display credible membership in a particular social world. Because its outcome reflects model capabilities, prompted identity, insider competence among human participants, and their AI literacy, the method provides a useful probe of the human-machine boundary across domains.
Chinese Translation
我们报告了在芬兰以芬兰语进行的一项图灵测试的结果。由于语言和文化语境在 LLM 训练数据中的呈现并不均衡，我们预期该模型（ChatGPT 5.2）在芬兰语图灵测试中的表现会逊于此前已研究过的英语美国语境。我们还提出，由模型生成的角色提示是一种可复现的技术，用于开展旨在提高构念效度的、基于 LLM 的比较性图灵测试。与我们的预期相反，该 LLM 通过了芬兰语图灵测试。一个突出的错误来源是参与者依赖语言线索，尤其是口语化的芬兰语，将其视为人类作者的标志。我们将图灵测试从一种智力测试重新界定为一种比较方法，用以考察一个 AI 系统能否在某个特定社会世界中显示出可信的成员身份。由于其结果反映了模型能力、被提示的身份、人类参与者的内部人能力以及他们的 AI 素养，该方法为跨领域探究人机边界提供了一种有用的探针。
cs.AI / 15 / 2609.18515
Beyond Routine Compliance: Cunning Data Cultivates Safety Vigilance in Large Language Models
超越例行合规：狡黠数据在大语言模型中培育安全警觉性
large language model
大语言模型相关
Abstract
Safety alignment teaches large language models (LLMs) to recognize harmful requests and reject risky instructions. Yet aligned models can fail when harmful intent is concealed within seemingly benign contexts. Robust safety therefore requires both knowledge of safety boundaries and \textbf{vigilance}: the ability to detect unusual premises, misleading reasoning, and latent risks beneath surface-level semantics. Vigilance requires models to scrutinize a request's underlying intent and assumptions before acting. To cultivate this capability, we introduce \textbf{cunning questions}, which are not necessarily safety-related but contain misleading premises, atypical reasoning, or subtle inconsistencies. We hypothesize that learning to look beyond such reasoning traps can transfer to safety-critical scenarios. Experiments show that Cunning training improves robustness to out-of-distribution jailbreak attacks and strengthens subsequent safety fine-tuning. Furthermore, augmenting an existing state-of-the-art safety alignment pipeline with Cunning establishes a new state of the art across our evaluated settings, reducing mean ASR across nine backbone--benchmark combinations from 17.40\% to 15.05\%. Trace analysis after matched safety fine-tuning suggests that safety judgments are more likely to govern responses before harmful planning begins. A conditional theoretical analysis further characterizes when invariance learned from cunning data can transfer to safety-related inputs. These findings suggest that cunning data can strengthen model vigilance and complement conventional safety alignment.
Chinese Translation
安全对齐教会大语言模型（LLM）识别有害请求并拒绝高风险指令。然而，当有害意图被隐藏在看似无害的语境中时，已对齐的模型仍可能失效。因此，稳健的安全性既需要对安全边界的知识，也需要\textbf{警觉性}：即察觉表面语义之下的异常前提、误导性推理与潜在风险的能力。警觉性要求模型在行动之前审视请求背后的意图与假设。为培育这一能力，我们引入了\textbf{狡黠问题}（cunning questions），它们未必与安全相关，但包含误导性前提、非典型推理或细微的不一致。我们假设，学会看穿此类推理陷阱的能力可以迁移到安全关键场景中。实验表明，狡黠训练提升了对分布外越狱攻击的鲁棒性，并增强了后续的安全微调。此外，用狡黠数据增强一个现有的最先进安全对齐流程后，在我们评估的各项设定中确立了新的最先进水平，将九个主干模型--基准组合上的平均 ASR 从 17.40\% 降至 15.05\%。在匹配的安全微调之后进行的迹分析表明，安全判断更有可能在有害规划开始之前就主导响应。一项条件性理论分析进一步刻画了从狡黠数据中学到的不变性何时能够迁移到与安全相关的输入上。这些发现表明，狡黠数据可以增强模型的警觉性，并补充常规的安全对齐。
cs.AI / 16 / 2609.18520
AeroWeaver: An Embodied-Agent Harness for Weaving Aerial Skills into Distributed, Adaptive Swarm Execution
AeroWeaver：一个将空中技能编织为分布式、自适应集群执行的具身智能体框架
large language model
大语言模型相关
Abstract
Collective intelligence is a collaborative autonomy paradigm in which multiple agents pursue shared objectives through local perception, information exchange, and coordinated action. UAV swarms embody this paradigm by coordinating multiple vehicles in tasks such as search, inspection, and tracking. Recent advances in large language model (LLM) agents have strengthened natural-language task understanding and high-level planning, providing a flexible semantic interface between mission descriptions and collective behavior. While these advances expand semantic reasoning, applying LLM agents to UAV swarms raises challenges in grounding model decisions in executable capabilities, reconciling global task reasoning with distributed execution, and using mission-specific experience for continual adaptation. To address these challenges, we introduce AeroWeaver, an embodied-agent harness that weaves individual UAV skills into coordinated mission-level behavior. AeroWeaver connects semantic decisions to governed skills, organizes role-conditioned local agents for distributed coordination, and uses role-indexed state-action-reward experience to refine skill selection online. Experiments and runtime validation show that AeroWeaver maintains valid skill execution under tested conditions and supports body-local multi-UAV operation without a central agent generating joint actions from global context, while reward-guided online updates provide a training-free path for adaptive learning swarm agents from accumulated execution experience. Code: https://github.com/Admire-ljb/AeroWeaver.
Chinese Translation
集群智能是一种协同自主范式，其中多个智能体通过局部感知、信息交换与协同行动来追求共同目标。无人机集群通过在搜索、巡检和跟踪等任务中协调多架飞行器来体现这一范式。近期大语言模型（LLM）智能体的进展增强了自然语言任务理解与高层规划能力，在任务描述与集群行为之间提供了一种灵活的语义接口。尽管这些进展扩展了语义推理，但将 LLM 智能体应用于无人机集群仍提出了若干挑战：将模型决策落地到可执行能力、协调全局任务推理与分布式执行，以及利用任务特定经验进行持续适应。为应对这些挑战，我们提出了 AeroWeaver，一种具身智能体框架，其将单个无人机技能编织为协调的任务级行为。AeroWeaver 将语义决策与受管控的技能相连接，组织以角色为条件的本地智能体以实现分布式协调，并利用以角色为索引的状态-动作-奖励经验来在线优化技能选择。实验与运行时验证表明，AeroWeaver 在测试条件下保持有效的技能执行，并支持机体本地的多无人机运行，而无需一个中央智能体根据全局上下文生成联合动作，同时奖励引导的在线更新为从累积执行经验中自适应学习集群智能体提供了一条免训练路径。代码：https://github.com/Admire-ljb/AeroWeaver。
cs.AI / 17 / 2609.18591
Recursive Reasoning or Statistical Extrapolation? In-Context Learning in Multi-Agent Interdependent Decision-Making
递归推理还是统计外推？多智能体相互依赖决策中的上下文学习
large language model
大语言模型相关
Abstract
In-context learning (ICL) enables large language model (LLM) agents to improve decisions using interaction history, yet it remains unclear whether such improvement reflects refined internal reasoning or mere extrapolation of statistical patterns. To disentangle these mechanisms, we study LLM agents in multi-agent incomplete-information games that require recursive belief reasoning. By constructing a public goods game and manipulating the statistical structure of historical feedback, we evaluate decision quality against a history-independent rational expectations equilibrium (REE) benchmark. Our experiments reveal that when historical statistical patterns are disrupted, the benefits of longer context largely vanish, degrading decision quality to the no-context baseline in a way sharply amplified by stronger strategic interdependence. These results suggest that, in such strategic environments, ICL behavior is more consistent with statistical extrapolation than with strategic reasoning. Our work extends the mechanistic study of ICL to strategic multi-agent settings, introduces REE as a diagnostic tool for distinguishing reasoning from extrapolation, and provides a reusable framework for probing the boundaries of LLM reasoning in recursive belief tasks.
Chinese Translation
上下文学习（ICL）使大语言模型（LLM）智能体能够利用交互历史来改进决策，然而这种改进究竟反映的是被精炼的内部推理，还是仅仅是对统计模式的外推，目前仍不清楚。为了厘清这些机制，我们研究了处于需要递归信念推理的多智能体不完全信息博弈中的LLM智能体。通过构建一个公共品博弈并操控历史反馈的统计结构，我们以与历史无关的理性预期均衡（REE）为基准来评估决策质量。我们的实验揭示，当历史统计模式被破坏时，更长上下文带来的收益在很大程度上消失，使决策质量退化至无上下文基线水平，而更强的策略相互依赖会显著放大这种退化。这些结果表明，在此类策略性环境中，ICL行为与统计外推更为一致，而非与策略推理一致。我们的工作将ICL的机制性研究拓展到策略性多智能体场景，引入REE作为区分推理与外推的诊断工具，并为探究LLM在递归信念任务中推理的边界提供了一个可复用的框架。
cs.AI / 18 / 2609.18597
Reasoning through Evolution: Automatic Meta-path Discovery for LLM-based Fake News Detection
通过进化进行推理：面向基于LLM的虚假新闻检测的自动元路径发现
large language model
大语言模型相关
Abstract
Propagation structures provide crucial evidence for fake news detection, yet existing approaches primarily rely on supervised GNN-based models, which require substantial labeled data and exhibit limited generalization. Although large language models (LLMs) exhibit strong reasoning capabilities, directly feeding them raw propagation graphs creates a significant modality mismatch and severe information overload, making structure-aware reasoning unreliable in zero-shot and few-shot settings. To bridge this gap, we propose MAGER, a multi-agent genetic evolution framework that automatically discovers meta-paths optimized for LLM reasoning. By compressing complex propagation graphs into informative subgraphs, the evolved meta-paths alleviate both information overload and modality mismatch, enabling frozen LLMs to perform structure-aware veracity reasoning. We further introduce a graph in-context learning strategy that retrieves semantically and structurally similar demonstrations to strengthen classification and reasoning. Extensive experiments show that MAGER substantially improves frozen LLMs as standalone fake news detectors in data-efficient settings. Our code is available at https://github.com/SenticNet/MAGER.
Chinese Translation
传播结构为虚假新闻检测提供了关键证据，然而现有方法主要依赖基于监督式GNN的模型，这类模型需要大量标注数据，且泛化能力有限。尽管大语言模型（LLM）展现出强大的推理能力，但直接将原始传播图输入其中会造成显著的模态不匹配与严重的信息过载，使得结构感知推理在零样本和少样本场景下并不可靠。为弥合这一差距，我们提出MAGER，一个多智能体遗传进化框架，能够自动发现针对LLM推理优化的元路径。通过将复杂的传播图压缩为信息丰富的子图，进化得到的元路径同时缓解了信息过载与模态不匹配，使冻结的LLM能够执行结构感知的真实性推理。我们进一步引入了一种图上下文学习策略，该策略检索在语义与结构上相似的示例，以增强分类与推理。大量实验表明，在数据高效设定的场景下，MAGER显著提升了冻结LLM作为独立虚假新闻检测器的性能。我们的代码可在 https://github.com/SenticNet/MAGER 获取。
cs.AI / 19 / 2609.18723
Beyond Truncation: Rethinking LLM Decoding as Ensemble Pruning
超越截断：将LLM解码重新思考为集成剪枝
large language model
大语言模型相关
Abstract
We introduce Mahalanobis-Ensemble Decoding (ME-Decoding), a novel Large Language Model (LLM) decoding framework that frames candidate token selection as ensemble pruning. Existing selection strategies rely predominantly on scalar probabilities, ignoring geometric semantic relationships and causing candidate redundancy. Meanwhile, current geometry-aware methods often require complex optimization or directly reweighting the original token probabilities, leading to significant computational overhead or inference instability. To address this, we formulate decoding as a subset optimization problem using a Mahalanobis distance-driven objective to enhance semantic diversity while preserving high probabilities. Specifically, we dynamically discount redundant generation paths using a token similarity matrix, constructed via an adaptive-bandwidth kernel over token embeddings. We further devise an efficient greedy selection algorithm with near-linear complexity in the candidate size under early stopping, while establishing its theoretical approximation guarantees. This renders ME-Decoding a robust, plug-and-play module with negligible inference overhead. Extensive experiments across diverse reasoning and generation tasks demonstrate that our method consistently achieves strong performance.
Chinese Translation
我们提出了马氏集成解码（Mahalanobis-Ensemble Decoding，ME-Decoding），这是一种新颖的大语言模型（LLM）解码框架，它将候选词元选择视为集成剪枝。现有的选择策略主要依赖于标量概率，忽略了语义的几何关系，并导致候选冗余。与此同时，当前的几何感知方法通常需要复杂的优化，或直接对原始词元概率重新加权，从而导致显著的计算开销或推理不稳定。为解决这一问题，我们将解码形式化为一个子集优化问题，采用马氏距离驱动的目标函数，以在保持高概率的同时增强语义多样性。具体而言，我们利用词元相似度矩阵动态地对冗余生成路径进行折扣，该矩阵通过在词元嵌入上使用自适应带宽核构建。我们进一步设计了一种高效的贪心选择算法，在提前停止条件下其复杂度相对于候选规模接近线性，同时建立了其理论近似保证。这使得 ME-Decoding 成为一个稳健、即插即用的模块，且推理开销可忽略不计。在多种推理与生成任务上的大量实验表明，我们的方法始终取得强劲的性能。
cs.AI / 20 / 2609.18731
Which LLM is Best for Translating Natural Language Goals to PDDL
哪个 LLM 最适合将自然语言目标转换为 PDDL
large language model
大语言模型相关
Abstract
Bridging the gap between human intent and machine execution remains a challenge in automated planning, where expressing goals in formal languages like PDDL restricts accessibility to non-experts. This paper empirically evaluates whether current Large Language Models (LLMs) can reliably translate natural language testing goals, written in informal language by video game testers, into well-formed PDDL targets suitable for classical planning. We present a carefully designed prompt template, integrating insights from iterative experimentation, aimed at maximizing both accuracy and response coherence from multiple state-of-the-art LLMs. Six contemporary models are systematically assessed on correctness, speed, and error tendencies using real-world, domain-specific benchmarks. All models demonstrate high correctness, exceeding 92\%, with Gemini 2.5 Flash achieving the highest accuracy at 96\% and the lowest incidence of false positives, while GPT-4.1 leads in response speed. Despite these advances, critical distinctions exist in model performance, and occasional failures arise from language ambiguity and limitations in domain representation. Our analysis underscores both the significant progress and ongoing gaps in enabling LLMs to act as robust bridges between natural language objectives and automated planning pipelines.
Chinese Translation
在自动规划中，弥合人类意图与机器执行之间的差距仍然是一项挑战，其中用 PDDL 等形式语言表达目标限制了非专家的可访问性。本文实证评估了当前的大型语言模型（LLM）能否可靠地将由视频游戏测试人员用非正式语言编写的自然语言测试目标转换为适合经典规划的良构 PDDL 目标。我们提出了一个精心设计的提示模板，整合了来自迭代实验的洞见，旨在最大化多个最先进 LLM 的准确性和响应连贯性。使用真实的、特定领域的基准，对六个当代模型在正确性、速度和错误倾向方面进行了系统评估。所有模型都表现出高正确性，超过 92\%，其中 Gemini 2.5 Flash 达到了最高的 96\% 准确率以及最低的假阳性发生率，而 GPT-4.1 在响应速度方面领先。尽管有这些进展，模型性能仍存在关键差异，并且偶发失败源于语言歧义和领域表示的局限性。我们的分析强调了在使 LLM 能够充当自然语言目标与自动规划流水线之间的稳健桥梁方面，既取得了显著进展，也存在持续存在的差距。
cs.AI / 21 / 2609.18736
Clueing up LLMs with Tool-Augmented Deductive Reasoning
以工具增强的演绎推理为 LLM 提供线索
large language model
大语言模型相关
Abstract
Despite recent advances in large language models (LLMs), performing logically consistent deductive reasoning over extended interactions remains challenging. Tasks that require integrating evidence across multiple reasoning steps, maintaining consistency with prior inferences, and updating beliefs under new constraints can surface limitations in current models while providing a useful testbed for evaluating reasoning enhancements. In this paper, we implement a text-based, multi-agent version of the classic board game Clue as an environment to evaluate multi-step, agentic deductive reasoning. In this setting, agents must infer hidden information from a sequence of observations, maintain consistency across turns, and reason over an evolving set of logical constraints. We instantiate six LLM-based agents (GPT-4o-mini and Gemini-2.5-Flash) as players that engage in turn-based gameplay; using three agents per model family, we establish baseline performance across repeated games. We then introduce a tool-augmented approach in which a structured possibility matrix converts implicit game state from generated reasoning logs into an explicit representation of remaining possibilities. The possibility matrix encodes extended-turn memory and deductive constraints, offloading these tasks from the agent. We compare this approach against the baseline to evaluate how tool augmentation supports reasoning quality and task success for autonomous agents in a strategic reasoning environment.
Chinese Translation
尽管大语言模型（LLM）近来取得了诸多进展，在扩展的交互过程中执行逻辑一致的演绎推理仍然颇具挑战性。那些需要跨多个推理步骤整合证据、与先前的推断保持一致、并在新约束下更新信念的任务，既可能暴露出当前模型的局限性，也为评估推理增强提供了有用的测试平台。在本文中，我们实现了经典棋盘游戏 Clue 的一个基于文本的多智能体版本，作为评估多步骤、智能体式演绎推理的环境。在这一设定中，智能体必须从一系列观测中推断隐藏信息，在多个回合之间保持一致，并在一个不断演化的逻辑约束集合上进行推理。我们将六个基于 LLM 的智能体（GPT-4o-mini 和 Gemini-2.5-Flash）实例化为参与回合制博弈的玩家；每个模型系列使用三个智能体，我们通过重复对局建立了基线性能。随后，我们引入一种工具增强的方法，其中结构化的可能性矩阵将生成推理日志中的隐式博弈状态转换为剩余可能性的显式表示。该可能性矩阵编码了扩展回合的记忆与演绎约束，从而将这些任务从智能体身上卸载。我们将这一方法与基线进行比较，以评估在策略推理环境中，工具增强如何支持自主智能体的推理质量和任务成功。
cs.AR / 22 / 2609.18675
HBFlex: A Flexible Memory System for Bridging Fine-Grained LLM States and Coarse-Grained HBF Parallel Execution
HBFlex：一种用于桥接细粒度 LLM 状态与粗粒度 HBF 并行执行的灵活内存系统
large language model
大语言模型相关
Abstract
Large language models (LLMs) require increasing memory capacity to accommodate growing model weights and KV caches. High-Bandwidth Flash (HBF) offers high memory density and aggregate read bandwidth through massive plane-level parallelism, making it an attractive option for LLM serving. However, serving LLMs entirely from HBF introduces three challenges: fine-grained KV reads create placement and access imbalance, incremental writes interfere with foreground reads, and mixed KV lifetimes amplify garbage collection. Hybrid HBM/HBF designs retain HBM to support dynamic KV management, but this allocation reduces the HBF resources available under a fixed packaging budget, limiting aggregate HBF bandwidth. We present HBFlex, a full-HBF memory system with coordinated optimizations for KV reads, writes, and reclamation. HBFlex balances KV placement and attention accesses to improve plane utilization. It aggregates incremental updates and schedules writeback within sufficiently long compute windows to reduce write--read interference. It also combines lifetime-guided block packing with deferred reclamation to reduce valid-page migration. We evaluate HBFlex through trace-driven simulation across different configurations. HBFlex achieves average throughput speedups of up to 1.58$\times$ over FlashAccel and 3.30$\times$ over H3, benefiting from higher HBF bandwidth and more efficient management of dynamic KV-cache reads, writes, and erases.
Chinese Translation
大型语言模型（LLM）需要不断增长的内存容量，以容纳日益增大的模型权重和 KV 缓存。高带宽闪存（HBF）通过大规模平面级并行提供高内存密度和聚合读带宽，使其成为 LLM 服务的一个有吸引力的选择。然而，完全从 HBF 服务 LLM 会带来三个挑战：细粒度 KV 读取造成放置和访问不均衡，增量写入干扰前台读取，以及混合的 KV 生命周期放大垃圾回收。混合 HBM/HBF 设计保留 HBM 以支持动态 KV 管理，但这种分配减少了在固定封装预算下可用的 HBF 资源，限制了聚合 HBF 带宽。我们提出 HBFlex，一种全 HBF 内存系统，针对 KV 读取、写入和回收进行了协同优化。HBFlex 平衡 KV 放置和注意力访问，以提高平面利用率。它将增量更新聚合起来，并在足够长的计算窗口内调度写回，以减少写—读干扰。它还将生命周期引导的块打包与延迟回收相结合，以减少有效页迁移。我们通过跨不同配置的轨迹驱动仿真来评估 HBFlex。HBFlex 相比 FlashAccel 实现了最高 1.58$\times$ 的平均吞吐量加速，相比 H3 实现了最高 3.30$\times$ 的加速，这得益于更高的 HBF 带宽以及对动态 KV 缓存读取、写入和擦除更高效的管理。
cs.CL / 23 / 2609.18005
A Calibrated Instrument for Measuring How Inference Optimizations Affect Output Quality
一种用于测量推理优化如何影响输出质量的经校准仪器
large language model
大语言模型相关
Abstract
Large language model optimization is an active research area, spanning quantization of model weights, early-exit methods for skipping layers, and speculative decoding. Each track uses its own quality measures, typically an idiosyncratic benchmark score. Few approach the measurement precision required by other scientific disciplines. We propose a rigorous methodology for measuring output quality, suitable for cross-system and cross-technique comparison. We score outputs with an LLM as a judge, but calibrate the judge formally: we compare its scores on two ordinary runs of a model given the same prompts, verifying that it shows no systematic preference between statistically equivalent outputs and measuring its per-sample noise. Each design also includes a 'null' condition, provably identical in distribution to the unmodified model, whose measured difference must be zero. With this one instrument we measure several acceleration techniques on the same prompts, so their quality costs can be compared. Perceived quality proves highly dependent on the domain of discourse. A 4-bit model was indistinguishable from its 16-bit original down to our design's +/-0.3-point resolution, in English prose and Chinese alike. At 3-bit precision the same prompts lost 0.5 points in English prose, 0.9 in Chinese, and 1.1 on multi-step math; early exit that cost 0.7 points on prose cost 2.5 on math, cutting correctly solved problems from 19 of 27 to 6. The pattern held for models from Alibaba and from Meta, but not its magnitude: the same quantizer cost Meta's model 1.8 points where it cost Alibaba's 0.7. A model's certainty about a token predicts how likely it is to differ from the full model's choice, but not how much that difference affects judged quality, so acceptance rules relying on certainty cannot distinguish errors that matter from errors that don't.
Chinese Translation
大语言模型优化是一个活跃的研究领域，涵盖模型权重量化、用于跳过层的提前退出方法以及推测解码。每条路线都使用自己的质量度量，通常是一种各自特有的基准分数。很少有方法能达到其他科学学科所要求的测量精度。我们提出一套严格的输出质量测量方法，适用于跨系统和跨技术比较。我们使用 LLM 作为评判者对输出评分，但对评判者进行形式化校准：我们比较它在给定相同提示的模型两次普通运行上的评分，验证它在统计上等价的输出之间没有系统性偏好，并测量其逐样本噪声。每个设计还包括一个“空”条件，其在分布上可证明与未修改模型相同，其测量差异必须为零。借助这一单一仪器，我们在相同提示上测量若干加速技术，从而可以比较它们的质量代价。感知质量被证明高度依赖于所讨论的领域。一个 4 比特模型与其 16 比特原始模型在我们设计的 +/-0.3 分分辨率下无法区分，在英语散文和中文中都是如此。在 3 比特精度下，相同提示在英语散文中损失了 0.5 分，在中文中损失了 0.9 分，在多步数学中损失了 1.1 分；在散文上造成 0.7 分损失的提前退出在数学上造成 2.5 分损失，将正确解决的问题从 27 个中的 19 个减少到 6 个。这一模式在来自阿里巴巴和来自 Meta 的模型上均成立，但幅度并非如此：同一个量化器使 Meta 的模型损失 1.8 分，而使阿里巴巴的模型损失 0.7 分。模型对某个 token 的确定性可以预测它偏离完整模型选择的概率，但不能预测这种差异对评判质量的影响程度，因此依赖确定性的接受规则无法区分重要错误与不重要错误。
cs.CL / 24 / 2609.18068
From a River in Gilead to the Inference Distributions of Large Language Models: Covert Dialect Bias and Linguistic Profiling at Scale
从基列的一条河到大型语言模型的推断分布：隐蔽方言偏见与大规模语言画像
large language model
大语言模型相关
Abstract
Large language models (LLMs) are increasingly deployed in high-stakes domains such as housing screening. While alignment techniques mitigate explicit racial bias in generated text, they often leave covert attitudinal associations in internal probability distributions untouched. Adapting the matched-guise sociolinguistic paradigm, we examine covert dialect bias in housing-related social judgments across four varieties: Standard American English (SAE), African American Vernacular English (AAVE), Nigerian Standard English (NSE), and Nigerian Pidgin (NP). AAVE reflects the racialized dialect studied in prior covert-bias evaluations, whereas NSE and NP represent Black African, postcolonial varieties absent from this literature. Using 260 meaning-matched sentence quadruples and log-probability scoring over housing-relevant adjectives, we probe ten open-weight LLMs across three contexts varying in social proximity: tenant screening, neighbor acceptance, and roommate selection. Across all ten models, AAVE and NP are consistently associated with more negative adjectives than SAE, with NP penalized most severely. Crucially, each dialect is penalized via distinct stereotype clusters rather than a generic non-standard category. NSE, which carries institutional prestige, displays a context-dependent shift: favored over SAE in formal tenant screening but increasingly penalized as social proximity grows. Our findings reveal that LLMs inherit covert dialect bias along both racial identity and prestige dimensions, echoing documented human housing discrimination and demonstrating its reach across postcolonial English varieties.
Chinese Translation
大型语言模型（LLMs）正越来越多地部署于住房筛选等高风险领域。尽管对齐技术减轻了生成文本中的显性种族偏见，但它们常常使内部概率分布中的隐蔽态度性关联保持不变。通过采用配对变体（matched-guise）社会语言学范式，我们考察跨四个语言变体在住房相关社会判断中的隐蔽方言偏见：标准美国英语（SAE）、非洲裔美国人白话英语（AAVE）、尼日利亚标准英语（NSE）和尼日利亚皮钦语（NP）。AAVE 反映了此前隐蔽偏见评估中研究过的种族化方言，而 NSE 和 NP 则代表该文献中缺席的黑非洲、后殖民语言变体。使用 260 组意义匹配的句子四元组以及对住房相关形容词的对数概率评分，我们在三种社会亲近度不同的语境中探测了十个开放权重 LLM：租户筛选、邻居接纳和室友选择。在所有十个模型中，AAVE 和 NP 一贯比 SAE 与更多负面形容词相关联，其中 NP 受到的惩罚最严重。至关重要的是，每种方言都是通过不同的刻板印象簇而非一个通用的非标准类别而受到惩罚的。NSE 带有制度性声望，表现出依赖语境的转变：在正式的租户筛选中比 SAE 更受青睐，但随着社会亲近度增加而日益受到惩罚。我们的发现揭示，LLM 在种族身份和声望两个维度上都继承了隐蔽方言偏见，这与已有记录的人类住房歧视相呼应，并表明其影响范围跨越后殖民英语变体。
cs.CL / 25 / 2609.18106
Linguistic Triggers of Gender and Racial Bias in Open-Weight LLMs Applied to Recruitment
应用于招聘的开放权重LLM中性别与种族偏见的语言触发因素
large language model
大语言模型相关
Abstract
Open-weight large language models are rapidly entering hiring pipelines, yet their discriminatory failure modes -- and the regulatory exposure these create under the EU AI Act high-risk classification (Annex III) and U.S. EEOC adverse-impact analysis -- remain poorly understood. We present the first systematic, multi-model audit of open-weight LLMs that treats job-posting language as the primary experimental variable, evaluating six models (Llama 3.2, Mistral, Gemma 3, Qwen 3, Phi 3, DeepSeek-R1) across four controlled experiments that jointly probe recruiter-simulation and job-seeker-simulation tasks. We find that (1) agentic posting language depresses recruiter recommendation scores for female candidates (r_rb = 0.309, p_Bonf = 7x10^-5; model-fixed-effects r_rb = 0.448), while communal language partially reverses the penalty; and (2) coded-exclusion language suppresses non-White recruiter scores at large effect sizes (r_rb = 0.646-0.758) and, on the job-seeker side, selectively deters non-White personas from expressing interest -- operationalizing a chilling-effect mechanism at scale. A label-ablation experiment isolates the explicit demographic persona label as the primary causal driver, and Word Embedding Association Tests corroborate these findings at the representational level (d = 1.01-1.45 under Caliskan et al.'s multi-word gender attribute lists). We translate these results into a concrete pre-deployment audit protocol -- posting-vocabulary scoring, persona-conditioned LLM probing, and adverse-impact flagging against the four-fifths threshold -- that operationalizes the documentation and risk-management obligations Annex III imposes on high-risk AI in recruitment.
Chinese Translation
开放权重的大语言模型正迅速进入招聘流程，但它们的歧视性失效模式——以及这些模式在《欧盟人工智能法案》高风险分类（附件III）和美国平等就业机会委员会（EEOC）不利影响分析下所产生的监管风险——仍鲜为人知。我们提出首个针对开放权重LLM的系统性、多模型审计，将职位发布语言作为主要实验变量，在四项受控实验中评估六个模型（Llama 3.2、Mistral、Gemma 3、Qwen 3、Phi 3、DeepSeek-R1），这些实验共同探查招聘者模拟和求职者模拟任务。我们发现：(1) 能动性招聘用语会降低女性候选人的招聘者推荐评分（$r_{rb} = 0.309$, $p_{Bonf} = 7x10^-5$；模型固定效应 $r_{rb} = 0.448$），而社群性语言会部分逆转这一惩罚；并且 (2) 编码排斥语言在较大效应量下抑制非白人招聘者评分（$r_{rb} = 0.646-0.758$），而且在求职者一侧，选择性地阻止非白人角色表达兴趣——在规模上操作化了一种寒蝉效应机制。一项标签消融实验将明确的人口统计人设标签识别为主要因果驱动因素，并且词嵌入关联测试在表征层面证实了这些发现（在 Caliskan 等人的多词性别属性列表下，d = 1.01-1.45）。我们将这些结果转化为一套具体的部署前审计协议——职位发布词汇评分、以人设为条件的LLM探测，以及对五分之四阈值的不利影响标记——该协议将附件III对招聘中的高风险AI所施加的文档和风险管理义务操作化。
cs.CL / 26 / 2609.18156
TeochewBench: A Human-Reviewed Benchmark for Teochew Hanzi Translation
TeochewBench：一个经人工审核的潮州话汉字翻译基准
large language model
大语言模型相关
Abstract
Teochew has a substantial speaker community and exhibits distinctive lexical, syntactic, and pragmatic features, yet textual resources for evaluating large language models remain limited. We present TeochewBench, a human-reviewed benchmark comprising 300 Teochew Hanzi expressions for evaluating translation from Teochew Hanzi into Mandarin Chinese and English. The dataset covers five categories: basic vocabulary; everyday sentences; Teochew-specific expressions; tone, politeness, and context; and idiomatic, ambiguous, and culturally specific expressions. A primary Teochew-speaking reviewer examined all entries individually and revised them as needed, while two additional Teochew speakers verified selected items. Our main evaluation covers 11 official general-purpose post-trained models on the reviewed dataset in both translation directions, yielding 6,600 predictions. Two official base checkpoints provide 1,200 predictions for supplementary diagnostics, bringing the total to 13 models and 7,800 predictions. We additionally include a Hanzi-copy control, which returns the source input unchanged, to assess how shared Hanzi affect automatic scores for translation into Mandarin Chinese. Qwen3.5-27B achieved the highest overall chrF-style score among the evaluated checkpoints, at 60.63, followed by Qwen2.5-72B-Instruct at 56.61, Gemma-3-27B-IT at 56.36, and GLM-4-32B-0414 at 55.82. Across the 11 main-evaluation models, the mean chrF-style score decreased from 69.25 for low-specificity items to 27.52 for high-specificity items. High-specificity expressions received lower scores and exhibited smaller cross-model differences, suggesting that they constitute a shared low-scoring region across the model families evaluated here. The Hanzi-copy control further indicates that surface overlap in low-specificity items can substantially affect automatic scores for translation into Mandarin Chinese.
Chinese Translation
潮州话拥有庞大的使用者群体，并展现出独特的词汇、句法和语用特征，然而用于评估大语言模型的文本资源仍然有限。我们提出 TeochewBench，一个经人工审核的基准，包含 300 条潮州话汉字表达，用于评估从潮州话汉字到普通话中文和英语的翻译。该数据集涵盖五个类别：基础词汇；日常句子；潮州话特有表达；语调、礼貌与语境；以及习语性、歧义性和文化特定性表达。一位主要的潮州话使用者审核者逐条检查了所有条目，并按需进行了修订，另有两位潮州话使用者对选定条目进行了核验。我们的主评估在经审核的数据集上覆盖了 11 个官方通用后训练模型，涵盖两个翻译方向，共产生 6,600 条预测。两个官方基础检查点提供了 1,200 条预测用于补充性诊断，使总数达到 13 个模型和 7,800 条预测。我们额外加入了一个汉字复制对照（Hanzi-copy control），它原样返回源输入，以评估共享汉字如何影响译入普通话中文的自动评分。在被评估的检查点中，Qwen3.5-27B 取得了最高的总体 chrF 类分数，为 60.63，其次是 Qwen2.5-72B-Instruct 的 56.61、Gemma-3-27B-IT 的 56.36 以及 GLM-4-32B-0414 的 55.82。在 11 个主评估模型中，平均 chrF 类分数从低特定性条目的 69.25 下降到高特定性条目的 27.52。高特定性表达获得的分数较低，且表现出较小的跨模型差异，这表明它们构成了此处所评估的各个模型家族共有的低分区域。汉字复制对照进一步表明，低特定性条目中的表层重叠会显著影响译入普通话中文的自动评分。
cs.CL / 27 / 2609.18204
Beyond Accuracy: How Procedural Traces Shift the Decision Criterion of LLM Overseers
超越准确率：程序性轨迹如何改变LLM监督者的决策标准
large language model
大语言模型相关
Abstract
Organizations increasingly use oversight loops where one large language model (LLM) audits another's outputs alongside procedural traces of claimed steps. A common concern about such LLM-as-a-judge pipelines is that detailed traces make overseers gullible. Using signal detection theory, we audit five LLM overseers on 19 compliance tasks (4,551 analyzed judgments), varying only trace detail and evidence labeling. With disconfirming evidence always visible, error detection remains near ceiling. Instead, elaborate traces shift the decision criterion toward rejection, increasing false alarms in susceptible overseers. Without option labels, human-validated reason coding shows about 60% of false alarms cite an inability to tie evidence to its option. Labels eliminate this stated reason, yet residual rejection of correct work persists in those overseers and rises with trace detail. Procedural traces thus act as governance artifacts that shape oversight decisions. AI auditors should be evaluated by their decision criterion and false-alarm behavior, alongside accuracy.
Chinese Translation
组织越来越多地使用监督回路，其中一个大型语言模型（LLM）会结合所声称步骤的程序性轨迹，审计另一个模型的输出。对于此类“LLM作为评判者”的流程，一个常见的担忧是，详细的轨迹会使监督者变得轻信。我们运用信号检测理论，在19项合规任务上审计了五个LLM监督者（分析了4,551条判断），仅改变轨迹详细程度和证据标注。在否定性证据始终可见的情况下，错误检测仍接近上限。相反，详尽的轨迹会使决策标准向拒绝方向偏移，从而在易受影响的监督者中增加虚警。在没有选项标签的情况下，经人工验证的原因编码显示，约60%的虚警提到无法将证据与其选项对应起来。标签消除了这一所述原因，但在那些监督者中，对正确工作的残余拒绝仍然存在，并随轨迹详细程度上升而增加。因此，程序性轨迹作为治理人工制品，塑造着监督决策。AI审计者应依据其决策标准和虚警行为，与准确率一并接受评估。
cs.CL / 28 / 2609.18282
Too Good to Be Real? Diagnosing and Reducing the Gap Between AI Preference and Real User Engagement
好得不真实？诊断并缩小 AI 偏好与真实用户参与之间的差距
large language model
大语言模型相关
Abstract
Large language models are increasingly used to generate and evaluate online content, yet it remains unclear whether the qualities they associate with higher engagement match what real users respond to. We study this question using 1.17 million answers to 25,978 questions from Zhihu, Quora, and Reddit, comparing real platform answers and AI-generated answers across four within-question engagement levels. We introduce Ontological Preference Measurement, which represents answers along three dimensions: logic, affect, and expression. We find a systematic gap between AI preference and real user engagement: as target engagement increases, LLMs add more explicit logical structure, while real user engagement is more strongly associated with affective and expressive salience. We call this tendency logic overbinding. Based on this diagnosis, we propose Ontology-Masked Reasoning Autoencoding (OMRA), a controlled intervention that masks and reconstructs over-explained spans while preserving stance, factual content, and coherence. Across four LLM families, OMRA reduces the measured gap by an average of 54.4%. In human evaluation, OMRA wins 62.4% of pairwise preference judgments against matched real platform answers, even though the real answers are more often judged to be human-written.
Chinese Translation
大型语言模型越来越多地被用于生成和评估在线内容，但它们与更高参与度相关联的品质是否与真实用户所回应的内容相匹配，仍不清楚。我们利用来自知乎、Quora 和 Reddit 的 25,978 个问题的 117 万条回答来研究这一问题，在四个问题内参与度水平上比较真实平台回答和 AI 生成回答。我们引入本体论偏好测量（Ontological Preference Measurement），它沿三个维度表示回答：逻辑、情感和表达。我们发现 AI 偏好与真实用户参与之间存在系统性差距：随着目标参与度上升，LLM 添加更明确的逻辑结构，而真实用户参与则与情感显著性和表达显著性关联更强。我们将这一倾向称为逻辑过度绑定（logic overbinding）。基于这一诊断，我们提出本体掩码推理自编码（Ontology-Masked Reasoning Autoencoding, OMRA），一种受控干预方法，它掩码并重构过度解释的片段，同时保留立场、事实内容和连贯性。在四个 LLM 系列中，OMRA 将测得的差距平均缩小 54.4%。在人工评估中，面对匹配的真实平台回答，OMRA 在成对偏好判断中赢得 62.4%，尽管真实回答更常被判定为人类撰写。
cs.CL / 29 / 2609.18304
Rollback the World, Keep the Reflection: Rollback-Induced Reflection for Long-Horizon LLM Agents
回滚世界，保留反思：面向长时程 LLM 智能体的回滚诱导反思
large language model
大语言模型相关
Abstract
Large language model (LLM) agents increasingly tackle long-horizon tasks through multi-step environment interaction, yet a single erroneous action can alter subsequent states and observations, causing errors to compound over time. Existing methods either correct the context without repairing altered environment states or restore earlier states while discarding useful experience, making it difficult to both eliminate failure conditions and avoid repeating past mistakes. We argue that reliable recovery should instead be treated as a rollback-boundary control problem that jointly determines when to intervene, where to resume, and what information should survive recovery. Based on this view, we propose Rollback-Induced Reflection (RIR), a unified recovery framework that restores execution to a selected prior state while carrying forward reusable knowledge distilled from the abandoned trajectory to guide subsequent decisions. We further characterize recovery through a unified operator over rollback depth and retained memory, providing a general view of state restoration and knowledge retention. Experiments on three long-horizon benchmarks demonstrate that RIR consistently improves task performance across multiple LLM backbones, with structured reflection memory preserving useful experience and selective rollback enabling efficient recovery.
Chinese Translation
大语言模型（LLM）智能体越来越多地通过多步环境交互来解决长时程任务，然而单个错误动作就可能改变后续状态和观测，导致错误随时间推移不断累积。现有方法要么在不修复已被改变的环境状态的情况下纠正上下文，要么在恢复较早状态的同时丢弃有用经验，因而难以既消除失败条件又避免重复过去的错误。我们认为，可靠的恢复应被视为一个回滚边界控制问题，该问题联合决定何时干预、从何处恢复，以及哪些信息应在恢复后保留。基于这一观点，我们提出回滚诱导反思（RIR），一个统一的恢复框架，它将执行恢复到选定的先前状态，同时向前传递从被放弃轨迹中提炼出的可复用知识，以指导后续决策。我们进一步通过一个作用于回滚深度和保留记忆的统一算子来刻画恢复，从而提供关于状态恢复与知识保留的一般性视角。在三个长时程基准上的实验表明，RIR 在多个 LLM 主干模型上持续提升任务性能，其中结构化反思记忆保留有用经验，而选择性回滚实现高效恢复。
cs.CL / 30 / 2609.18317
Knowledge-Graph Based Augmentation versus Retrieval Augmented Generation for Cultural-Related Question Answering
基于知识图谱的增强与检索增强生成在文化相关问答中的对比
large language model
大语言模型相关
Abstract
Large language models (LLMs) suffer from a long-tail deficit: culturally specific facts, particularly those concerning underrepresented regions such as Latin America, appear too rarely in pretraining corpora to be reliably memorized. Retrieval-Augmented Generation (RAG) addresses this by grounding generation in external text, but structured alternatives such as Knowledge Graphs (KGs) offer tighter control over what enters the context, along with potential gains in explainability and updatability. We benchmark Graph-RAG against standard RAG on LatamQA, a culturally grounded multiple-choice dataset spanning eight thematic categories. The graphs are built end-to-end from Wikipedia articles with KGGen, a recent open-domain extractor, without manual curation in our main setting. G-Retriever is competitive with RAG and reduces the error of the base LLM by 72\% with a standard KG and 78\% with a benchmark-aware variant, the gap to RAG narrowing further as the graph is oriented toward task-relevant content. The trained projection transfers zero-shot to Portuguese without target-language fine-tuning, indicating multilingual reach.
Chinese Translation
大型语言模型（LLMs）存在长尾缺陷：文化特定事实，尤其是那些涉及拉丁美洲等代表性不足地区的事实，在预训练语料中出现得过于稀少，以至于无法被可靠地记住。检索增强生成（RAG）通过将生成建立在外部文本之上来解决这一问题，但诸如知识图谱（KGs）之类的结构化替代方案对进入上下文的内容提供了更严格的控制，并可能在可解释性和可更新性方面带来收益。我们在 LatamQA 上对 Graph-RAG 与标准 RAG 进行基准测试，LatamQA 是一个具有文化根基、涵盖八个主题类别的多选题数据集。这些图谱是使用 KGGen（一种近期的开放域抽取器）从维基百科文章端到端构建的，在我们的主要设置中没有进行人工整理。G-Retriever 与 RAG 相比具有竞争力，并且在使用标准 KG 时将基础 LLM 的错误降低了 72\%，在使用基准感知变体时降低了 78\%，随着图面向任务相关内容，与 RAG 的差距进一步缩小。训练得到的投影无需目标语言微调即可零样本迁移到葡萄牙语，表明其具有多语言覆盖能力。
cs.CL / 31 / 2609.18320
Attention Dispersion as a Diagnostic Signal for Hallucination in Large Language Models
注意力分散作为大型语言模型幻觉的诊断信号
large language model
大语言模型相关
Abstract
Large Language Models (LLMs) frequently exhibit hallucinations, presenting a major barrier to reliability in complex reasoning tasks. While traditional detection methods rely on output-based confidence metrics, these logits are often miscalibrated by modern alignment techniques. In this paper, we investigate the temporal volatility of internal attention mechanisms as an alternative diagnostic signal for hallucination that does not depend on output calibration. By introducing an unsupervised metric for attention dispersion, we show that epistemic uncertainty leaves a measurable trace within intermediate layers, where spikes in attention entropy are associated with reasoning breakdowns. We evaluate our approach on mathematical reasoning benchmarks (GSM8K and MATH-500) using the Qwen2.5 model family (1.5B and 3B parameters), finding statistically significant AUC improvements of up to +0.076 over output-based baselines across all tested conditions. These findings suggest that attention dispersion is a promising complement to traditional hallucination detection methods, requiring further investigation across broader model families and task domains.
Chinese Translation
大型语言模型（LLMs）经常表现出幻觉，这为复杂推理任务中的可靠性构成了主要障碍。虽然传统检测方法依赖于基于输出的置信度指标，但这些 logits 往往被现代对齐技术错误校准。在本文中，我们研究内部注意力机制的时间波动性，作为不依赖于输出校准的幻觉诊断替代信号。通过引入一种用于注意力分散的无监督度量，我们表明认知不确定性会在中间层中留下可测量的痕迹，其中注意力熵的尖峰与推理崩溃相关。我们使用 Qwen2.5 模型系列（1.5B 和 3B 参数）在数学推理基准（GSM8K 和 MATH-500）上评估我们的方法，发现在所有测试条件下，相较于基于输出的基线，AUC 提升高达 +0.076，且具有统计显著性。这些发现表明，注意力分散是传统幻觉检测方法的一种有前景的补充，需要在更广泛的模型系列和任务领域中进一步研究。
cs.CL / 32 / 2609.18494
Size Matters: Foundation Model for Czech HTML documents
规模至关重要：面向捷克语 HTML 文档的基础模型
large language model
大语言模型相关
Abstract
Creating universal, high-quality representations of web documents in high-traffic industrial environments requires models that are both performant and economic. Existing approaches, however, often depend on large models, overlook the structural information inherent in HTML, or are constrained by short context windows, limiting their ability to process real-world web pages. We present HTML-LM, a compact foundation model with 154 million parameters that addresses these limitations through HTML-aware training and a ModernBERT-based architecture. It was trained on 100 million web documents using multiple objectives, including masked language modeling, bag-of-words prediction, and contrastive distillation from large language models. Consequently, HTML-LM sets a new state-of-the-art for classification and regression applications in the Czech Internet domain, surpassing both larger encoders and small-sized LLMs. The model is deployed in production, processing thousands of web documents per second, and released to the community under the CC BY-NC 4.0. https://huggingface.co/Seznam/html-lm.
Chinese Translation
在高流量的工业环境中创建通用、高质量的网页文档表示，需要既性能出色又经济的模型。然而，现有方法往往依赖大型模型、忽视 HTML 固有的结构信息，或受限于短上下文窗口，从而限制了它们处理真实世界网页的能力。我们提出 HTML-LM，一个拥有 1.54 亿参数的紧凑型基础模型，它通过 HTML 感知训练和基于 ModernBERT 的架构解决了这些限制。它使用多个目标在 1 亿份网页文档上进行训练，包括掩码语言建模、词袋预测，以及来自大型语言模型的对比蒸馏。因此，HTML-LM 在捷克互联网领域的分类和回归应用中树立了新的最先进水平，超越了更大的编码器和小型 LLM。该模型已部署在生产环境中，每秒处理数千份网页文档，并以 CC BY-NC 4.0 许可向社区发布。https://huggingface.co/Seznam/html-lm.
cs.CL / 33 / 2609.18516
Align, Integrate, and Fire: Efficient Token-Level Alignment for Zero-Shot SpeechLLMs
对齐、整合与激发：面向零样本 SpeechLLM 的高效 Token 级对齐
large language model
大语言模型相关
Abstract
While Large Language Models excel in natural language processing, efficiently extending their capabilities to spoken input remains a significant challenge. Existing methods for building SpeechLLMs often rely on computationally expensive full-model fine-tuning, or employ parameter-efficient projectors that suffer from inefficient token sequence lengths and costly full-model supervision. In this paper, we introduce Aligned Continuous Integrate-and-Fire, a highly efficient framework for zero-shot speech processing. Our method dynamically compresses continuous acoustic frames into the exact discrete token length of the target text utilizing explicit Dynamic Time Warping alignments. This allows our initial training stage to establish a robust acoustic-to-semantic bridge using lightweight distance metrics, entirely bypassing the computationally expensive LLM forward pass. For subsequent fine-tuning, we propose a memory-efficient knowledge distillation objective that targets a single LLM layer, performing competitively with full-model cross-entropy training at a fraction of the computational cost. Through extensive evaluations on Automatic Speech Recognition and Speech Translation, we demonstrate that our method achieves superior performance compared to prior parameter-efficient baselines.
Chinese Translation
尽管大语言模型在自然语言处理方面表现出色，但将其能力高效地扩展到语音输入仍然是一项重大挑战。现有构建 SpeechLLM 的方法往往依赖于计算代价高昂的全模型微调，或者采用参数高效的投影器，而后者受困于低效的 token 序列长度和代价高昂的全模型监督。在本文中，我们提出了 Aligned Continuous Integrate-and-Fire，一个用于零样本语音处理的高效框架。我们的方法利用显式的动态时间规整对齐，将连续的声学帧动态压缩到目标文本精确的离散 token 长度。这使得我们的初始训练阶段能够使用轻量级距离度量建立起稳健的声学到语义的桥梁，完全绕过了计算代价高昂的 LLM 前向传播。对于后续的微调，我们提出了一种内存高效的知识蒸馏目标，其仅针对单个 LLM 层，以一小部分计算成本实现了与全模型交叉熵训练相当的性能。通过在自动语音识别和语音翻译上的大量评估，我们证明我们的方法相较于先前的参数高效基线取得了更优的性能。
cs.CL / 34 / 2609.18642
STRETCH the Boundaries: A Unified Self-Taught Framework for Progressive LLM Evolution
拓展边界：一个用于渐进式 LLM 演化的统一自教框架
large language model
大语言模型相关
Abstract
Large language models (LLMs) often suffer from capability stagnation in self-improvement training because fixed difficulty levels fail to adapt to their evolving proficiency. To address this issue, we propose STRETCH (Self-Taught Reasoning Evolution via Targeted CHallenge), a unified framework inspired by cognitive scaffolding theory. STRETCH introduces a dynamic Stretch Zone mechanism that continuously aligns question difficulty with the model's solving capability. Within a single parameter space, the model alternates between a Scaffolder that generates adaptive, boundary-pushing challenges and a Learner that that optimizes its solving trajectories through reinforcement learning. This dual-loop co-evolution effectively stabilizes training, mitigates reward hacking and promote progressive reasoning growth. Experiments on both negotiation and operation research benchmarks demonstrate that STRETCH consistently outperforms strong prompting and domain-specific baselines. Further scaffolder configuration analysis shows that dynamic difficulty alignment is critical for sustained capability improvement and synchronized reasoning evolution.
Chinese Translation
大语言模型（LLM）在自我改进训练中常常遭遇能力停滞，因为固定的难度水平无法适应其不断演进的熟练程度。为了解决这一问题，我们提出了 STRETCH（Self-Taught Reasoning Evolution via Targeted CHallenge，通过定向挑战进行的自学推理演化），一个受认知脚手架理论启发的统一框架。STRETCH 引入了一种动态拉伸区（Stretch Zone）机制，该机制持续地将问题难度与模型的求解能力对齐。在单一参数空间内，模型在脚手架生成器（Scaffolder）和学习器（Learner）之间交替：前者生成自适应的、推动边界的挑战，后者通过强化学习优化其求解轨迹。这种双循环协同演化有效地稳定了训练，缓解了奖励黑客行为，并促进了渐进式推理增长。在谈判和运筹学基准上的实验表明，STRETCH 始终优于强大的提示方法和领域特定基线。进一步的脚手架生成器配置分析表明，动态难度对齐对于持续的能力提升和同步的推理演化至关重要。
cs.CL / 35 / 2609.18739
A Scalable Framework for Automated NER Annotation Correction in Low-Resource Languages
一个面向低资源语言的可扩展自动化 NER 标注纠错框架
large language model
大语言模型相关
Abstract
Poor quality or noisy annotations in Named Entity Recognition (NER), as in any other NLP task, make it challenging to achieve state-of-the-art performance. In this paper, we present a multi-step framework to enhance the annotation quality of NER datasets by employing automated techniques. We propose a frequency-based iterative approach that leverages self-training and a dual-threshold mechanism to enhance inference confidence. Experimental evaluations on different NER datasets demonstrate significant improvements in NER performance with respect to the original datasets. This work further explores the potential of generative Large Language Models (LLMs) to perform NER for low-resource languages.
Chinese Translation
命名实体识别（NER）中质量差或有噪声的标注，如同在任何其他 NLP 任务中一样，会使达到最先进性能变得困难。在本文中，我们提出了一个多步骤框架，通过采用自动化技术来提升 NER 数据集的标注质量。我们提出了一种基于频率的迭代方法，该方法利用自训练和双阈值机制来增强推理置信度。在不同 NER 数据集上的实验评估表明，相对于原始数据集，NER 性能取得了显著提升。这项工作进一步探索了生成式大语言模型（LLM）在低资源语言上执行 NER 的潜力。
cs.CL / 36 / 2609.19144
A Zeroth-Order Paradigm for LLM Preference Alignment
一种用于 LLM 偏好对齐的零阶范式
large language model
大语言模型相关
Abstract
Direct preference alignment methods are widely used to align large language models (LLMs) with human preferences because of their computational and memory efficiency. However, likelihood displacement motivates alternative ways to extract information from preference pairs with small likelihood margins. In this paper, we propose and analyze Comparison-based Preference Optimization (ComPO), a zeroth-order alignment method based on comparison oracles. ComPO extracts directional information from these pairs without directly optimizing a differentiable preference loss on them. We establish a convergence guarantee for its basic offline scheme under smoothness, gradient sparsity, and compatibility between the oracle and a latent objective. We further introduce online ComPO, which retains the offline comparison mechanism and uses unlabeled policy generations for reverse-KL control relative to a reference policy. Following the coverage perspective of preference fine-tuning, we establish a performance guarantee for a basic constrained scheme under local coverage and in-distribution pairwise reward accuracy. Experiments on Mistral, Llama, Gemma-2, Qwen3, and Gemma-3 models demonstrate improvements over existing direct alignment methods, including length-controlled win rates, with pair-level diagnostics providing evidence consistent with mitigating likelihood displacement.
Chinese Translation
直接偏好对齐方法因其计算和内存效率而被广泛用于将大语言模型（LLMs）与人类偏好对齐。然而，似然位移促使人们探索其他方式，从具有小似然间隔的偏好对中提取信息。在本文中，我们提出并分析基于比较的偏好优化（Comparison-based Preference Optimization, ComPO），一种基于比较预言机的零阶对齐方法。ComPO 从这些偏好对中提取方向信息，而不直接在这些偏好对上优化可微的偏好损失。我们在平滑性、梯度稀疏性以及预言机与潜在目标之间的兼容性条件下，为其基本离线方案建立了收敛保证。我们进一步引入在线 ComPO，它保留离线比较机制，并使用未标注的策略生成来进行相对于参考策略的反向-KL 控制。遵循偏好微调的覆盖视角，我们在局部覆盖和分布内成对奖励准确率下，为一个基本约束方案建立了性能保证。在 Mistral、Llama、Gemma-2、Qwen3 和 Gemma-3 模型上的实验表明，相较于现有直接对齐方法取得了改进，包括长度控制胜率，同时成对级诊断提供了与缓解似然位移相一致的证据。
cs.CR / 37 / 2609.17839
Evaluating the Impact of Personalization in Conversational Cybersecurity Assistants
评估个性化在对话式网络安全助手中的影响
large language model
大语言模型相关
Abstract
Users increasingly turn to Large Language Models to answer a variety of questions, including cybersecurity questions. We study how personalization strategies can help improve the effectiveness of answers to questions asked to an LLM-based cybersecurity assistant. Beyond accuracy, we focus on the understandability, actionability and, most importantly, motivating power of answers, given how often users fail to follow cybersecurity recommendations. Specifically, we investigate four personalization strategies, ranging from static user profiles to interaction-history-based personalization, using a corpus of 1,045 real-world cybersecurity questions and a 7-day deployment involving 57 participants and 1,066 user questions. Across both a large-scale automated LLM-based evaluation and human evaluation, conversation-based personalization is consistently favored in comparative ratings of perceived helpfulness and likelihood of following security advice. Importantly, the relative trends observed in the LLM-based evaluation align with those obtained from human evaluation, suggesting that LLM-based evaluation can provide a scalable mechanism for comparing personalization strategies before costly user studies. These results indicate that behavior-driven personalization is a promising direction for LLM-powered cybersecurity assistants and highlight the value of combining LLM-based and human evaluation when studying personalized language-model systems.
Chinese Translation
用户越来越多地求助于大型语言模型来回答各种问题，包括网络安全问题。我们研究个性化策略如何能够帮助提高对基于LLM的网络安全助手所提问题的回答的有效性。除了准确性之外，我们关注回答的可理解性、可操作性，以及最重要的激励力，因为用户经常未能遵循网络安全建议。具体而言，我们研究了四种个性化策略，从静态用户画像到基于交互历史的个性化，使用包含1,045个真实世界网络安全问题的语料库，以及一项涉及57名参与者和1,066个用户问题的为期7天的部署。在大规模自动化基于LLM的评估和人工评估这两者中，基于对话的个性化在感知有用性和遵循安全建议可能性的比较评分中始终更受青睐。重要的是，在基于LLM的评估中观察到的相对趋势与从人工评估中获得的趋势一致，这表明基于LLM的评估可以提供一种可扩展的机制，用于在昂贵的用户研究之前比较个性化策略。这些结果表明，行为驱动的个性化是LLM驱动的网络安全助手的一个有前景的方向，并凸显了在研究个性化语言模型系统时结合基于LLM的评估和人工评估的价值。
cs.CR / 38 / 2609.18526
The Illusion of Local Privacy: Confidentiality Boundary Failures in Consumer LLM Serving Systems
本地隐私的幻觉：消费级 LLM 服务系统中的机密性边界失效
large language model
大语言模型相关
Abstract
Running large language models (LLMs) locally is often considered more private than cloud-hosted inference because user prompts remain on the device. We ask whether keeping inference local is, by itself, sufficient to keep those prompts confidential. Our results show that it is not: prompt confidentiality also depends on how the surrounding serving software handles prompt data before, during, and after inference. We examine four boundaries at which prompt confidentiality can fail in consumer local-LLM serving systems: model loading, runtime memory, wrapper-level persistence, and the serving interface. To study these boundaries, we develop LLAnalyzer, a measurement framework that tests each boundary separately and traces observed failures to the responsible software component. Applying LLAnalyzer to four open-weight model families and two consumer deployment platforms, we find markedly different behaviour across boundaries. In a 24-hour AFL++ campaign with more than 12 million executions, we observe no parser crashes or successful malformed GGUF loads within the explored state space. Runtime memory tells a different story: we recover prompts after inference because multiple plaintext representations survive in allocator-managed memory, and sanitisation reduces this residue without eliminating it. We also find that consumer wrappers can extend prompt lifetime through plaintext persistence. At the serving boundary, we uncover a previously undocumented authorization flaw in llama.cpp that allows one authenticated client to restore another tenant's saved conversation state; the attack succeeds in 200/200 controlled trials. Separately, shared prompt-prefix caching exposes a remote timing oracle that remains distinguishable under WAN conditions. We argue that local LLM systems need explicit guarantees for prompt lifetime, persistent storage, and tenant isolation.
Chinese Translation
在本地运行大语言模型（LLM）通常被认为比云端托管的推理更具隐私性，因为用户提示词保留在设备上。我们追问：将推理保持在本地，其本身是否足以使这些提示词保持机密。我们的结果表明并非如此：提示词的机密性还取决于周边的服务软件在推理之前、之中和之后如何处理提示词数据。我们考察消费级本地 LLM 服务系统中提示词机密性可能失效的四个边界：模型加载、运行时内存、封装层持久化和服务接口。为研究这些边界，我们开发了 LLAnalyzer，这是一个测量框架，它分别测试每一个边界，并将观察到的失效追溯到应负责的软件组件。将 LLAnalyzer 应用于四个开放权重模型家族和两个消费级部署平台，我们发现在不同边界上的行为存在显著差异。在一场持续 24 小时、执行次数超过 1200 万次的 AFL++ 测试活动中，在所探索的状态空间内，我们没有观察到解析器崩溃或畸变 GGUF 加载成功的情况。运行时内存则讲述了不同的故事：我们在推理之后恢复出提示词，因为多种明文表示形式在由分配器管理的内存中存活下来，而清理操作减少了这种残留却并未将其消除。我们还发现，消费级封装程序可以通过明文持久化延长提示词的生命周期。在服务边界上，我们揭示了 llama.cpp 中一个此前未被记录的授权缺陷，它允许一个已认证的客户端恢复另一个租户保存的对话状态；该攻击在 200/200 次受控试验中均成功。另外，共享的提示词前缀缓存暴露了一个远程计时预言机，其在广域网条件下仍保持可区分性。我们认为，本地 LLM 系统需要对提示词生命周期、持久化存储和租户隔离提供明确的保证。
cs.AI / 39 / 2609.18007
Newer Is Not Fairer: Gender Stereotyping in Text-to-Image AI Across Model Generations
更新的并不更公平：文本到图像 AI 跨模型世代的性别刻板印象
diffusion
扩散模型相关
Abstract
Text-to-image generative models are widely used in professional and creative settings, yet how they represent gender across occupations -- and whether newer models are fairer -- remains poorly understood across multiple generations. We evaluate gender representation across 20 occupations, 5 prompt templates, and 4 Stable Diffusion model generations (SD 1.5, SD 2.1, SDXL, SD 3 Medium), generating 8,000 images with n = 100 per occupation-model cell (5 prompts x 20 images), and classifying all with DeepFace. Across the 8,000 open-source images, 76.4% show male subjects (95% CI [75.1%, 78.7%], p < 2.2 x 10^-16, Benjamini-Hochberg adjusted). More strikingly, 57.6% of images for historically female-coded occupations show male subjects (raw p = 3.43 x 10^-22, BH-adjusted p = 1.71 x 10^-21). All nine significant tests reported in this paper survive BH correction across 10 tests. When compared against U.S. Bureau of Labor Statistics workforce data, models underrepresent women by 20-46pp on average, with particularly large deviations for near gender-balanced occupations: scientist (48% female in BLS, 82-99% male in model outputs) and cleaner (46% female in BLS, 80-92% male in outputs). Model generations do not improve steadily: bias worsens from SD 1.5 to SDXL before partially recovering in SD 3 Medium. A preliminary comparison with GPT-image-1 on five occupations suggests lower bias than open-source models, though the practical effect is small (Cramer's V = 0.080) and the comparison is exploratory. No model achieves gender parity.
Chinese Translation
文本到图像生成模型在专业和创意环境中被广泛使用，然而它们在各个职业中如何表征性别——以及更新的模型是否更公平——在多个世代中仍鲜为人知。我们评估了 20 个职业、5 个提示模板和 4 个 Stable Diffusion 模型世代（SD 1.5、SD 2.1、SDXL、SD 3 Medium）中的性别表征，生成了 8,000 张图像，每个职业-模型单元格 n = 100（5 个提示词 x 20 张图像），并用 DeepFace 对全部图像进行分类。在这 8,000 张开源图像中，76.4% 显示男性主体（95% 置信区间 [75.1%, 78.7%]，p < 2.2 x 10^-16，经 Benjamini-Hochberg 校正）。更引人注目的是，历史上女性编码的职业中，57.6% 的图像显示男性主体（原始 p = 3.43 x 10^-22，BH 校正后 p = 1.71 x 10^-21）。本文报告的所有九项显著检验在经过 10 项检验的 BH 校正后仍然显著。当与美国劳工统计局劳动力数据比较时，模型平均将女性代表性低估了 20-46 个百分点，对于接近性别平衡的职业偏差尤其大：科学家（BLS 中女性占 48%，模型输出中男性占 82-99%）和清洁工（BLS 中女性占 46%，输出中男性占 80-92%）。模型世代并没有稳步改善：偏差从 SD 1.5 到 SDXL 变得更严重，然后在 SD 3 Medium 中部分恢复。与 GPT-image-1 在五个职业上的初步比较表明，其偏差低于开源模型，尽管实际效应很小（Cramer's V = 0.080），且该比较是探索性的。没有模型实现性别均等。
cs.LG / 40 / 2609.18077
vidax: A Unified JAX Framework for Video Generative Models on Accelerator Meshes
vidax：面向加速器网格上视频生成模型的统一 JAX 框架
diffusion
扩散模型相关
Abstract
Open-source video generative models ship almost exclusively as PyTorch/CUDA reference implementations. This leaves Cloud TPU pods without a production-ready inference path, despite offering large, cost-effective accelerator memory pools ideal for long-sequence spatiotemporal attention. We present vidax, an open-source JAX/Flax inference engine and zero-copy PyTorch-to-JAX weight translator for modern video generation architectures. vidax covers a diverse set of spatiotemporal models --- including Diffusion Transformers, omnimodal Mixture-of-Transformers, 3D VAEs, text encoders, and native samplers --- with zero PyTorch dependency in the execution path. The framework unifies 1D tensor parallelism with DeepSpeed-Ulysses sequence parallelism on a single JAX sharding mesh, integrates TPU flash-attention kernels, and implements per-layer weight offloading to support reference resolutions that exceed single-device memory. We benchmark compile times, latency, and peak memory utilization on TPU v4-8 hardware, and document real-world numerical bugs surfaced during checkpoint translation. vidax is released open-source as a baseline for JAX and TPU video generation research.
Chinese Translation
开源视频生成模型几乎仅以 PyTorch/CUDA 参考实现的形式发布。这使得 Cloud TPU pod 缺乏生产就绪的推理路径，尽管它们提供了大规模、高性价比的加速器内存池，非常适合长序列时空注意力。我们提出 vidax，一个开源的 JAX/Flax 推理引擎以及面向现代视频生成架构的零拷贝 PyTorch 到 JAX 权重转换器。vidax 覆盖了多样化的时空模型集合——包括 Diffusion Transformer、全模态 Mixture-of-Transformers、3D VAE、文本编码器和原生采样器——且在执行路径中对 PyTorch 零依赖。该框架在单一 JAX 分片网格上统一了 1D 张量并行与 DeepSpeed-Ulysses 序列并行，集成了 TPU flash-attention 内核，并实现了逐层权重卸载，以支持超出单设备内存的参考分辨率。我们在 TPU v4-8 硬件上对编译时间、延迟和峰值内存利用率进行了基准测试，并记录了在检查点转换过程中暴露出的真实场景数值 bug。vidax 已作为 JAX 与 TPU 视频生成研究的基线开源发布。
cs.LG / 41 / 2609.18488
Beyond Random Couplings: Contrastive Noise Alignment in Generative Flows
超越随机耦合：生成流中的对比噪声对齐
diffusion
扩散模型相关
Abstract
Diffusion and flow-matching models are typically trained by corrupting data through independently sampled Gaussian noise. While simple and scalable, this forward process induces arbitrary data-noise couplings, forcing the network to learn high-curvature transports between unrelated endpoints. Existing optimal-transport methods reduce this burden by reassigning fixed noise samples to data, but the source noise distribution itself remains passive. To address this, we introduce Contrastive Noise Alignment (CNA), a training-time method that creates dynamic, contrastive couplings by optimizing the noise representations directly. By modeling the noise batch as an interacting particle system, CNA employs a cross-modal InfoNCE objective to align noise particles with their paired data targets. To prevent spatial collapse, this alignment is regularized using an angular entropy term and a radial norm penalty. We show theoretically that this equilibrium asymptotically preserves Gaussian structures, maintaining tractability during inference. Empirically, CNA improves the alignment between noise and data, reduces flow curvature, and provides better generation quality with fewer required sampling steps. For few-step, pixel-space generation (2-4 NFEs), CNA reduces FID by over 50\% compared to standard rectified flow, and by at least 24\% against Optimal Transport baselines.
Chinese Translation
扩散模型和流匹配模型通常通过独立采样的高斯噪声对数据进行破坏来进行训练。尽管简单且可扩展，这种前向过程会引入任意的数据-噪声耦合，迫使网络学习无关端点之间的高曲率传输。现有的最优传输方法通过将固定的噪声样本重新分配给数据来减轻这一负担，但源噪声分布本身仍然是被动的。为了解决这个问题，我们引入对比噪声对齐（Contrastive Noise Alignment, CNA），这是一种训练时方法，通过直接优化噪声表示来创建动态的、对比性的耦合。通过将噪声批次建模为相互作用的粒子系统，CNA采用跨模态InfoNCE目标来将噪声粒子与其配对的数据目标对齐。为了防止空间坍缩，这种对齐使用角度熵项和径向范数惩罚进行正则化。我们从理论上表明，这种平衡渐近地保持高斯结构，从而在推理期间保持可处理性。经验上，CNA改善了噪声与数据之间的对齐，降低了流曲率，并在所需采样步骤更少的情况下提供更好的生成质量。对于少步、像素空间生成（2-4 NFEs），与标准整流流相比，CNA将FID降低超过50\%，而相对于最优传输基线至少降低24\%。
cs.AI / 42 / 2609.17842
Lexara-RF: Reference-Free Metrics for Evaluating Conversational Visual Analytics Agents
Lexara-RF：用于评估对话式可视分析智能体的无参考指标
large language model
大语言模型相关
Abstract
Conversational visual analytics (CVA) agents powered by large language models generate visualizations and natural-language explanations from open-ended queries. Evaluating these multimodal outputs is challenging: curated reference benchmarks are costly to author, cannot comprehensively capture the space of valid responses, and are unavailable in production. Building on the Lexara evaluation framework, we introduce Lexara-RF, a reference-free set of metrics that scores CVA outputs using only the prompt, data, and model response. We reformulate evaluation as verification: 13 metrics operationalize visualization design theory and Gricean cooperative principles as computable consistency, intent-alignment, and design validity checks. On a human-rated corpus of CVA test-cases, Lexara-RF achieves alignment comparable to reference-based formulations, outperforms surface-similarity NLG baselines, and localizes structurally grounded failures with high accuracy.
Chinese Translation
由大语言模型驱动的对话式可视分析（CVA）智能体能够根据开放式查询生成可视化和自然语言解释。评估这些多模态输出具有挑战性：精心构建的参考基准的编写成本高昂，无法全面覆盖有效响应的空间，并且在生产环境中不可用。在 Lexara 评估框架的基础上，我们引入了 Lexara-RF，这是一组无参考指标，仅使用提示、数据和模型响应即可对 CVA 输出进行评分。我们将评估重新表述为验证：13 项指标将可视化设计理论和 Gricean 合作原则操作化为可计算的一致性、意图对齐和设计有效性检查。在一个由人工评分的 CVA 测试用例语料库上，Lexara-RF 达到了与基于参考的表述相当的对齐效果，优于表面相似度 NLG 基线，并以高准确率定位出具有结构依据的失败。
cs.AI / 43 / 2609.18384
GYROval: A Robust Benchmark for Cultural Value Orientation in Large Language Models
GYROval：大型语言模型文化价值取向的稳健基准
large language model
大语言模型相关
Abstract
We present a robust benchmark for measuring cultural value orientation in large language models on the two Inglehart-Welzel axes over several domains and roles (hence GYROval - Gridded Yielding of Robust value Orientation), together with the results of administering it to twenty models. Items are binary contrastive scenarios in the sense introduced by CDEval: both options are legitimate courses of action, neither is correct, there is no answer key, and a model's score on an axis is the proportion of its responses falling on the counted pole. Eleven of the twenty models were additionally administered a paired Russian translation of the identical items and a second sampling temperature. The instrument is publicly released in both languages. Stability was assessed by treating the vignette as the unit of analysis, ranking the models within the levels of each perturbation factor, and summarising the agreement between levels by tie-corrected Kendall's \emph{W} against an empirical permutation null.
Chinese Translation
我们提出了一个稳健基准，用于测量大型语言模型在两个 Inglehart-Welzel 轴上的文化价值取向，涵盖若干领域和角色（因此称为 GYROval——Gridded Yielding of Robust value Orientation），并给出了将其施测于二十个模型的结果。题项是 CDEval 所引入意义上的二元对比情景：两个选项都是正当的行动路径，二者都不正确，没有答案键，而模型在某一轴上的得分是其落在被计入的一极上的回答所占的比例。二十个模型中有十一个还额外接受了相同题项的配对俄语翻译以及第二个采样温度。该测量工具以两种语言公开发布。稳定性通过以下方式评估：将情景短文作为分析单位，在每个扰动因素的水平内对模型进行排序，并通过经平局校正的 Kendall's \emph{W} 与经验置换零分布相对照，来概括各水平之间的一致性。
cs.LG / 44 / 2609.17682
DSD: Learning Diverse and Reusable Motor Skills via Diffusion Skill Discovery
DSD：通过扩散技能发现学习多样且可复用的运动技能
diffusion
扩散模型相关
Abstract
Humans efficiently learn new tasks by reusing a rich repertoire of motor skills across different goals and contexts. A similar strategy can also be used to enable simulated characters to efficiently perform new tasks by leveraging reusable motor skills. To support a wide range of downstream tasks, the learned repertoire should be diverse, consisting of distinct behaviors as well as spatial and temporal variation within each behavior. A commonly used method for learning diverse skills is by maximizing the mutual information between skill latents and the states produced by a policy. The marginal state entropy promotes broad behavioral coverage, while the conditional entropy encourages consistent behaviors from each latent. However, directly estimating the marginal state entropy is intractable in high-dimensional control problems. Prior methods therefore rely on indirect latent-space approximations or coarse estimators of the state distribution. These approximations may not effectively promote broad coverage of the state space, resulting in skills with limited behavioral diversity and reduced utility for downstream tasks. In this work, we propose Diffusion Skill Discovery (DSD), a skill discovery method that uses a diffusion model to approximate the entropy gradient of the policy-induced state distribution through score matching. The resulting objective encourages the discovery of skills that produce a broader range of behaviors for high-dimensional humanoid control. The learned skills are reused in two downstream control settings: hierarchical control with a task-specific high-level policy and zero-shot control through latent selection from offline trajectories. Our experiments show that DSD discovers a broader repertoire of reusable motor skills than prior skill discovery methods, leading to the emergence of complex and agile behaviors that can be reused across downstream tasks.
Chinese Translation
人类通过在不同目标和情境中复用丰富的运动技能库，从而高效地学习新任务。类似策略也可用于使模拟角色通过利用可复用的运动技能来高效执行新任务。为支持广泛的下游任务，所学技能库应当是多样的，既包含彼此不同的行为，也包含每种行为内部的空间和时间变化。一种常用的学习多样技能的方法是通过最大化技能潜变量与策略产生的状态之间的互信息。边缘状态熵促进广泛的行为覆盖，而条件熵鼓励每个潜变量产生一致的行为。然而，在高维控制问题中直接估计边缘状态熵是难以处理的。因此，先前的方法依赖于间接的潜空间近似或对状态分布的粗略估计器。这些近似可能无法有效促进对状态空间的广泛覆盖，从而导致技能的行为多样性有限，并降低其对下游任务的效用。在这项工作中，我们提出扩散技能发现（DSD），一种技能发现方法，它使用扩散模型通过分数匹配来近似策略诱导的状态分布的熵梯度。由此得到的目标鼓励发现能够在高维人形控制中产生更广泛行为的技能。学到的技能在两个下游控制设置中被复用：带有任务特定高层策略的分层控制，以及通过从离线轨迹中选择潜变量实现的零样本控制。我们的实验表明，与先前的技能发现方法相比，DSD 发现了更广泛的可复用运动技能库，从而导致能够跨下游任务复用的复杂且敏捷行为的涌现。
cs.LG / 45 / 2609.17691
Accelerating Diffusion Sampling via Speculative Draft Trees
通过投机草稿树加速扩散采样
diffusion
扩散模型相关
Abstract
Speculative sampling accelerates diffusion model generation by drafting inexpensive candidate states and correcting them under a coupling that preserves the target distribution exactly, reducing the number of expensive target evaluations. Existing diffusion samplers, notably those based on reflection maximal coupling, are topologically constrained: their lookahead drafts form a chain graph, a single linear sequence, which inherently limits the acceptance rate per target evaluation. We connect speculative sampling in diffusion models to relative entropy coding (REC). This perspective shows the lookahead need not be linear and motivates our central contribution, draft trees, which enrich the candidates considered per round and lower the target function evaluations. We further adopt greedy rejection sampling, an REC algorithm, as the draft-target coupling, improving acceptance while guaranteeing exact target samples. Experiments across diverse target and draft models demonstrate up to 8.3% acceleration over the reflection coupling baseline in practical settings.
Chinese Translation
投机采样通过草拟廉价的候选状态，并在一种精确保持目标分布的耦合下对其进行校正，从而加速扩散模型生成，减少昂贵目标评估的次数。现有的扩散采样器，尤其是那些基于反射最大耦合的采样器，在拓扑上受到约束：它们的前瞻草稿形成链图，即单个线性序列，这本质上限制了每次目标评估的接受率。我们将扩散模型中的投机采样与相对熵编码（REC）联系起来。这一视角表明，前瞻不必是线性的，并启发了我们的核心贡献——草稿树，它丰富了每轮所考虑的候选，并降低了目标函数评估。我们进一步采用贪心拒绝采样，一种 REC 算法，作为草稿-目标耦合，在保证精确目标样本的同时提高接受率。在不同目标模型和草稿模型上的实验表明，在实际设置中，相较于反射耦合基线，最高可获得 8.3% 的加速。
cs.LG / 46 / 2609.17837
Adaptive hybrid coupling with operator inference, the overlapping Schwarz alternating method and reinforcement learning
基于算子推断、重叠 Schwarz 交替方法和强化学习的自适应混合耦合
diffusion
扩散模型相关
Abstract
Hybrid domain decomposition methods provide a flexible framework for coupling full order models (FOMs) and reduced order models (ROMs), but typically assume the model assigned to each subdomain is fixed throughout a simulation. This is limiting for transient problems in which localized features propagate through the domain and the regions requiring high-fidelity resolution change over time. We introduce a reinforcement learning (RL)-based approach for online adaptation of FOM-ROM models coupled via the overlapping Schwarz alternating method (O-SAM), an iterative domain decomposition method that solves subdomain-local problems while exchanging solution information through transmission boundary conditions on overlapping interfaces. Deep Q-networks (DQNs) are trained offline to select among subdomain-local FOMs and pre-trained Operator Inference (OpInf) ROMs using a reward balancing accuracy, cost, and model-switching frequency. Once trained, the policies are deployed predictively on problem instances not seen during training, without requiring a reference FOM solution. We demonstrate the approach on two examples: a 1D advection-diffusion problem with a moving front, and a 3D linear elastic wave propagation problem implemented in the Norma.jl solid mechanics code. For the advection-diffusion benchmark, the learned policy dynamically allocates high-fidelity resolution as the front propagates and outperforms static FOM/ROM assignments; letting the agent also adapt the domain decomposition provides no further benefit. For the elastic wave benchmark, learned policies for two and three subdomain decompositions track the propagating wave by assigning FOMs to subdomains containing the wave and ROMs elsewhere, as expected. Our results demonstrate the potential of RL to enable predictive online adaptation of model fidelity within Schwarz-based hybrid simulations.
Chinese Translation
混合区域分解方法为耦合全阶模型（FOMs）和降阶模型（ROMs）提供了灵活的框架，但通常假设分配给每个子域的模型在整个模拟过程中是固定的。这对瞬态问题而言具有局限性，因为在这类问题中局部特征会在区域内传播，需要高保真分辨率的区域会随时间变化。我们提出了一种基于强化学习（RL）的方法，用于在线自适应通过重叠 Schwarz 交替方法（O-SAM）耦合的 FOM-ROM 模型；O-SAM 是一种迭代区域分解方法，它在求解子域局部问题的同时，通过在重叠界面上的传输边界条件交换解信息。深度 Q 网络（DQNs）通过离线训练，在子域局部 FOMs 和预训练的算子推断（OpInf）ROMs 之间进行选择，其奖励权衡了精度、成本和模型切换频率。一旦训练完成，这些策略会被预测性地部署到训练期间未见过的问題实例上，而不需要参考 FOM 解。我们在两个示例上展示了该方法：一个具有移动前沿的一维对流-扩散问题，以及一个在 Norma.jl 固体力学代码中实现的三维线弹性波传播问题。对于对流-扩散基准，学习到的策略在前沿传播时动态分配高保真分辨率，并优于静态 FOM/ROM 分配；让智能体同时自适应区域分解不会带来进一步收益。对于弹性波基准，针对两个和三个子域分解的学习策略通过将 FOMs 分配给包含波的子域、将 ROMs 分配给其他子域来跟踪传播中的波，这符合预期。我们的结果证明了 RL 有潜力在基于 Schwarz 的混合模拟中实现模型保真度的预测性在线自适应。
cs.LG / 47 / 2609.17841
Hybrid coupling with numerics-informed neural networks and the overlapping Schwarz alternating method
数值信息神经网络与重叠 Schwarz 交替方法的混合耦合
diffusion
扩散模型相关
Abstract
We develop a hybrid modeling framework for coupling pre-trained numerics-informed neural networks (NINNs) with classical full order models (FOMs) using the overlapping Schwarz alternating method. We consider the two-dimensional advection-diffusion equation in the advection-dominated, Peclet-number 10^6 regime. We first demonstrate that, unlike the corresponding physics-informed neural network (PINN), a monolithic NINN can be accurately trained on our model problem without domain decomposition. We then employ overlapping multiplicative Schwarz as a deployment mechanism for coupling a pre-trained, subdomain-local NINN with a neighboring FOM, with the NINN weights held fixed throughout the Schwarz iteration. We consider two training approaches for the subdomain-local NINNs: a top-down approach, in which boundary data are obtained from a coupled Schwarz solve on the full domain with a FOM on each subdomain (FOM-FOM Schwarz), and a bottom-up approach, in which boundary traces are generated synthetically on the NINN subdomain without requiring any full-domain solves. The resulting hybrid NINN-FOM solutions agree closely with the corresponding FOM-FOM Schwarz solutions, with the top-down and bottom-up training approaches yielding comparable accuracy.
Chinese Translation
我们开发了一个混合建模框架，用于使用重叠 Schwarz 交替方法将预训练的数值信息神经网络（NINNs）与经典的全阶模型（FOMs）耦合起来。我们考虑对流占优、Peclet 数为 10^6 情形下的二维对流-扩散方程。我们首先证明，与相应的物理信息神经网络（PINN）不同，整体式 NINN 可以在我们的模型问题上无需区域分解即可被准确训练。随后，我们采用重叠乘性 Schwarz 作为一种部署机制，将预训练的、子域局部的 NINN 与相邻的 FOM 耦合起来，并在整个 Schwarz 迭代过程中保持 NINN 权重固定不变。对于子域局部的 NINN，我们考虑两种训练方法：一种是自顶向下方法，其中边界数据由在整个区域上、每个子域均使用 FOM 的耦合 Schwarz 求解（FOM-FOM Schwarz）获得；另一种是自底向上方法，其中边界迹是在 NINN 子域上合成生成的，不需要任何全区域求解。由此得到的混合 NINN-FOM 解与相应的 FOM-FOM Schwarz 解高度吻合，且自顶向下和自底向上两种训练方法产生的精度相当。
cs.LG / 48 / 2609.18148
LIGE-GR: A Smooth Leap from Ranking to Generative Recommendation in the LLM Era
LIGE-GR：LLM 时代从排序到生成式推荐的一次平滑跨越
large language model
大语言模型相关
Abstract
The remarkable success of large language models (LLMs) has provided important inspiration for the next generation of recommender systems. Structurally, recommendation and language generation share a similarity: both aim to produce an ordered sequence that optimizes the user's experience. However, how to precisely absorb the essence of the LLM paradigm into mature industrial recommender systems remains an open problem. There are two challenges. First, it is unclear how to incorporate sequence-level generation and optimization from the LLM paradigm into recommendation. Second, real-world recommender systems are mature systems that have been iteratively customized for years around specific products, business constraints, serving infrastructure, and organizational ownership. Replacing such systems wholesale is often technically risky and organizationally disruptive. In this paper, we propose LIGE-GR, a listwise generation and evaluation recommendation framework that upgrades from a traditional ranking system based on itemwise recommendation toward a generative recommendation paradigm. Instead of rebuilding the entire recommendation stack from scratch, LIGE-GR generalizes the existing pointwise recommendation system into a listwise generation system. This allows mature recommender systems to benefit from listwise optimization while preserving compatibility with existing models, value functions, and serving infrastructure. We validate LIGE-GR in short-video recommendation on Instagram Reels and Facebook Video. On these recommendation surfaces, LIGE-GR improves time spent by 1.14 percent on Instagram Reels and 0.72 percent on Facebook Video, while requiring only modest additional inference resources.
Chinese Translation
大语言模型（LLM）的显著成功为下一代推荐系统提供了重要启发。从结构上看，推荐与语言生成具有相似性：二者都旨在生成一个优化用户体验的有序序列。然而，如何将 LLM 范式的精髓精确地吸收到成熟的工业推荐系统中，仍然是一个开放问题。这里有两个挑战。首先，如何将 LLM 范式中的序列级生成与优化融入推荐仍不明确。其次，现实世界的推荐系统是成熟系统，多年来围绕特定产品、业务约束、服务基础设施和组织归属进行了迭代定制。整体替换此类系统通常具有技术风险，并会造成组织上的破坏。在本文中，我们提出 LIGE-GR，一种列表级生成与评估推荐框架，它将基于逐物品推荐的传统排序系统升级为生成式推荐范式。LIGE-GR 不是从头重建整个推荐栈，而是将现有的逐点推荐系统泛化为列表级生成系统。这使成熟的推荐系统能够从列表级优化中受益，同时保持与现有模型、价值函数和服务基础设施的兼容性。我们在 Instagram Reels 和 Facebook Video 的短视频推荐中验证了 LIGE-GR。在这些推荐界面上，LIGE-GR 在 Instagram Reels 上使使用时长提升了 1.14%，在 Facebook Video 上提升了 0.72%，同时仅需适度的额外推理资源。
cs.LG / 49 / 2609.18273
Behavioral Fingerprinting and Navigation Prediction in Web Browsing
Web 浏览中的行为指纹识别与导航预测
large language model
大语言模型相关
Abstract
Web browsing often appears ephemeral: users visit a few websites, complete a task, and move on. However, even short fragments of browsing activity can contain rich and structured behavioral signals. In this work, we conduct a comparative empirical study of two complementary behavioral inference tasks: session-level user identification and next-domain prediction. Both tasks are derived from the same cleaned event stream and evaluated on large-scale anonymous browsing traces, with sessionization and splitting adapted to the temporal requirements of each task. For user identification, we evaluate classical and neural models operating on session-level behavioral and domain features. For next-domain prediction, we combine graph-based modeling with Large Language Models (LLMs). Experimental results show that short browsing sessions are highly identifiable, while future navigation actions are highly predictable from long-term interaction structure combined with recent behavioral context. Furthermore, LLM-derived semantic features yield only marginal gains over purely structural and sequential models, indicating that repeated interaction patterns remain the dominant predictive signal in the evaluated web-browsing setup. These findings highlight the extent to which interaction history substantially contributes to both user identifiability and navigation predictability in browsing traces.
Chinese Translation
Web 浏览往往看似短暂易逝：用户访问几个网站，完成一项任务，然后便继续前行。然而，即便是短暂的浏览活动片段，也可能包含丰富且结构化的行为信号。在本研究中，我们对两项互补的行为推断任务开展了比较性实证研究：会话级用户识别与下一域名预测。这两项任务均源自同一清洗后的事件流，并在大规模匿名浏览轨迹上进行评估，其中会话划分与数据集切分均根据各任务的时间要求进行了适配。对于用户识别，我们评估了在会话级行为特征与域名特征上运行的经典模型与神经模型。对于下一域名预测，我们将基于图的建模与大型语言模型（LLMs）相结合。实验结果表明，短浏览会话具有高度可识别性，而未来的导航行为可通过长期交互结构与近期行为上下文相结合而被高度预测。此外，与纯结构模型和序列模型相比，由 LLM 衍生的语义特征仅带来边际增益，这表明在所评估的 Web 浏览设置中，重复的交互模式仍是主导性的预测信号。这些发现凸显了交互历史在浏览轨迹中对用户可识别性与导航可预测性两方面均有显著贡献的程度。
cs.LG / 50 / 2609.18306
Bias Amplification in Multi-Agent Network: How Biased Agents Shape Opinions and Rhetoric
多智能体网络中的偏见放大：有偏智能体如何塑造观点与修辞
large language model
大语言模型相关
Abstract
Large language models (LLMs) are increasingly deployed in applications involving interaction between agents, where their output plays a role in collective reasoning and decision-making processes. Despite significant research into the functioning of LLMs in such multi-agent systems, the processes of bias propagation in such systems are still a challenge. This work studies how biased opinions are propagated in the form of textual interaction in an environment of LLMs, in which a minority of agents maintain persistent extreme opinions, while the remaining agents iteratively update their beliefs through structured textual interactions. The findings show that even the presence of a small percentage of biased agents in such a system leads to significant shifts in the opinions of non-biased agents. It suggests that for the same percentage of biased agents, the shifts occur more quickly for the Llama~3.2 model when compared to a classical Friedkin-Johnsen (FJ) model. Further semantic analysis demonstrates that rhetorical consistency in textual explanations increases systematically with biased exposure and, importantly, is partially decoupled from numerical convergenumericalutral agents adopt the vocabulary employed by the biased agents even in configurations where their numerical opinion shifts remain moderate. The research helps explain how bias and language develop together in multi-agent language model ecosystems.
Chinese Translation
大型语言模型（LLMs）越来越多地被部署在涉及智能体之间交互的应用中，其输出在集体推理和决策过程中发挥作用。尽管已有大量研究探讨 LLMs 在此类多智能体系统中的运作方式，但此类系统中的偏见传播过程仍然是一个挑战。本文研究在 LLMs 环境中，有偏观点如何以文本交互的形式传播；在该环境中，少数智能体维持持续性的极端观点，而其余智能体则通过结构化的文本交互迭代更新其信念。研究发现表明，即使此类系统中只有很小比例的有偏智能体存在，也会导致无偏智能体的观点发生显著偏移。这表明，对于相同比例的有偏智能体，与经典的 Friedkin-Johnsen (FJ) 模型相比，Llama~3.2 模型中的偏移发生得更快。进一步的语义分析表明，文本解释中的修辞一致性随着有偏暴露的增加而系统性地提高，并且重要的是，它与数值收敛部分脱钩。中性智能体会采用有偏智能体所使用的词汇，即使在它们的数值观点偏移仍然温和的配置中也是如此。该研究有助于解释偏见与语言如何在多智能体语言模型生态系统中共同发展。
cs.LG / 51 / 2609.18466
Spatially Adaptive Noise Injection
空间自适应噪声注入
diffusion
扩散模型相关
Abstract
Diffusion samplers reverse a learned noising process using either stochastic (DDPM) or deterministic (DDIM) updates, which represent endpoints of a single family controlled by a scalar noise-injection variance that is applied identically at every spatial location. This uniform approach neglects the geometry of natural images: high-curvature regions such as edges and textures, where the denoiser is uncertain, benefit from stochastic correction, whereas smooth regions, where the score is precise, are degraded by injected noise. This work investigates whether each pixel requires stochastic correction at a given timestep and introduces Spatially Adaptive Noise Injection (SANI), a novel sampling framework that dynamically adjusts noise application on a per-pixel basis. SANI integrates a probabilistic gating mechanism with a derived spatially adaptive variance, ensuring that noise is injected precisely where needed to refine complex features while preserving well-formed structures. Experimental results and decoupling ablations demonstrate that SANI consistently improves Fréchet Inception Distance (FID) over the vanilla DDPM and DDIM endpoint samplers across diverse sampling timesteps, while remaining competitive with variance-learning baselines, highlighting the importance of spatial adaptivity in diffusion sampling.
Chinese Translation
扩散采样器利用随机（DDPM）或确定性（DDIM）更新来逆转一个已学习的加噪过程，这两者代表了由标量噪声注入方差所控制的同一族方法的两个端点，而该方差在每个空间位置上都被完全相同地施加。这种均匀的做法忽略了自然图像的几何结构：诸如边缘和纹理等高曲率区域，即去噪器不确定之处，能够从随机校正中获益；而平滑区域，即分数精确之处，则会因注入的噪声而退化。本文研究在给定时间步上每个像素是否需要随机校正，并提出空间自适应噪声注入（SANI），一种新颖的采样框架，它在逐像素的基础上动态调整噪声的施加。SANI 将概率门控机制与推导出的空间自适应方差相结合，确保噪声被精确注入到需要之处，以细化复杂特征，同时保持结构完好。实验结果与解耦消融表明，在多种采样时间步下，SANI 相较于原始的 DDPM 与 DDIM 端点采样器持续改善了 Fréchet Inception Distance（FID），同时与方差学习基线相比仍具竞争力，凸显了空间自适应性在扩散采样中的重要性。
cs.LG / 52 / 2609.18560
The evolution of sex for artificial intelligence: a population-genetic framework for multigenerational model populations
人工智能中有性生殖的演化：一个面向多代模型群体的群体遗传学框架
large language model
大语言模型相关
Abstract
Some aspects of AI development resemble a population process in which models are specialised, retrained on the output of peers, or combined by averaging weights. These practices lead to generations of models, in the biological sense studied by population genetics. Here, I develop this parallelism and interpret multigenerational model populations in terms of sexual and asexual reproduction, formally recombining the two fields. I test these analogies in an exact inheritance model, in trained networks (recurrent, feedforward and variational autoencoder generators) and in large language models, and show that they hold generally, with some measurable architecture-specific biases. Training recursively on model output is known to lead to model collapse, a process previously described as akin to genetic drift; I develop all that follows. A minimal model of a learner retrained on its parent's output reproduces the Wright-Fisher process exactly; verified real data added to each generation play the role of immigration, with the surprising finding that the absolute number of real data samples matters, not their share, exactly as in population genetics. Training a child on the average of its parents' outputs cancels the benefit of having several parents, matching blending inheritance (and reviving Jenkin's objection to Darwin), whereas combining parents so that each keeps its strongest contribution preserves it; merged language-model specialists exceeded every parent across seeds (the Fisher-Muller effect); and lineages become reproductively isolated, losing the ability to merge at all, when they have learned conflicting conventions and not when they have merely drifted apart. As AI societies become societies in time as well as in space, a mathematical framework for their inheritance acquires predictive power. Remarkably, that framework can be adapted almost wholesale from biology.
Chinese Translation
AI 发展的某些方面类似于一个群体过程，其中模型被专门化、在同伴的输出上重新训练，或通过平均权重进行组合。这些做法产生了模型的一个个世代，其意义正是群体遗传学所研究的生物学意义上的世代。在这里，我发展了这一平行类比，并以有性生殖和无性生殖来解释多代模型群体，从而在形式上重组这两个领域。我在一个精确的遗传模型、在训练过的网络（循环网络、前馈网络和变分自编码器生成器）以及在大型语言模型中检验了这些类比，并表明它们普遍成立，同时存在一些可测量的、特定于架构的偏差。已知在模型输出上递归训练会导致模型崩溃，这一过程此前被描述为类似于遗传漂变；我发展接下来的全部内容。一个在被其亲本输出上重新训练的学习者的最小模型精确地再现了 Wright-Fisher 过程；每一代加入的已验证真实数据扮演迁入的角色，并且有一个令人惊讶的发现：真正重要的是真实数据样本的绝对数量，而不是其占比，这与群体遗传学中完全一样。用其亲本输出的平均值来训练一个子代，会抵消拥有多个亲本的好处，符合融合遗传（并重新引发詹金对达尔文的反对），而将亲本组合起来、使每个亲本保留其最强贡献，则能保留这一好处；合并后的语言模型专家在不同随机种子下超过了每一个亲本（Fisher-Muller 效应）；并且，当谱系学到了相互冲突的约定时，它们会变得生殖隔离，完全失去合并的能力，而当它们仅仅是漂移分开时则不会如此。随着 AI 社会在空间上以及时间上都成为社会，一个关于其遗传的数学框架便获得了预测能力。值得注意的是，该框架几乎可以整体地从生物学中改编而来。
cs.LG / 53 / 2609.18612
Weakening Neurons: An Input-Output Functionality in Transformers with Outsize Influence
削弱神经元：Transformer 中一种具有超大影响力的输入-输出功能
large language model
大语言模型相关
Abstract
We analyze the learned input-output behavior of GLU-based neurons in large language models (LLMs). We propose a simple analysis method: For each neuron, we compute the cosine similarities between its input (reading) and output (writing) weight vectors. In this scheme, a strong negative cosine similarity indicates the neuron weakens the direction it detects in the residual stream, so we call this a weakening neuron. This allows us to gain a number of novel insights. First, we show that nine different LLMs have similar patterns: weakening neurons appear mostly in late layers whereas their counterparts, (conditional) strengthening neurons, are frequent in early-middle layers. Second, we find that weakening neurons display surprising behavior: even though there are few, they activate often and have a large influence on model behavior. Third, weakening neurons have a strong effect on model output when gate values are negative -- which is surprising since negative gate values are not expected to encode functionality.
Chinese Translation
我们分析大型语言模型（LLM）中基于 GLU 的神经元所学到的输入-输出行为。我们提出一种简单的分析方法：对于每个神经元，我们计算其输入（读取）与输出（写入）权重向量之间的余弦相似度。在该方案中，强烈的负余弦相似度表明该神经元削弱它在残差流中检测到的方向，因此我们将其称为削弱神经元。这使我们能够获得若干新颖的洞见。首先，我们表明九个不同的 LLM 具有相似的模式：削弱神经元主要出现在后层，而它们的对应物——（条件性）增强神经元——则频繁出现在早中期层。其次，我们发现削弱神经元表现出令人惊讶的行为：尽管数量很少，它们却频繁激活，并对模型行为产生很大影响。第三，当门控值为负时，削弱神经元对模型输出有很强的影响——这令人惊讶，因为负门控值本不应被认为编码了功能。
cs.LG / 54 / 2609.18708
Rethinking Critic Learning in PPO: Understanding and Mitigating Value Flattening
重新思考PPO中的评论家学习：理解并缓解价值平坦化
large language model
大语言模型相关
Abstract
In reinforcement learning for large language models, Proximal Policy Optimization (PPO) commonly uses a critic to estimate state values and reduce the variance of policy updates. However, we uncover a systematic failure mode in PPO critics, which we call Value Flattening: state values, estimated from multiple Monte Carlo continuations, change sharply across intermediate states while critic predictions remain comparatively flat. We further observe this phenomenon in a controlled FrozenLake environment and find that it becomes more pronounced as the state space grows. Our theoretical and empirical analyses relate Value Flattening to an implicit variance penalty in the critic loss and redundant updates from temporally correlated states with similar gradients. Motivated by these findings, we introduce SParse Proximal Policy Optimization (SP$^3$O), which applies the value loss to only a few well-separated states in each response to mitigate both effects. Experiments on Qwen3-Base show that SP$^3$O with only three states supervised per response can mitigate Value Flattening and consistently improve the learned policy across model sizes and evaluation suites. Together, our results identify Value Flattening as an important yet overlooked failure mode of critic learning in standard PPO and show that a simple sparse supervision strategy can mitigate it.
Chinese Translation
在面向大语言模型的强化学习中，近端策略优化（PPO）通常使用一个评论家来估计状态价值并降低策略更新的方差。然而，我们揭示了PPO评论家中一种系统性的失效模式，我们将其称为价值平坦化：由多条蒙特卡洛延续估计得到的状态价值在中间状态之间急剧变化，而评论家的预测却保持相对平坦。我们进一步在受控的FrozenLake环境中观察到这一现象，并发现随着状态空间增大，它变得更加明显。我们的理论和实证分析将价值平坦化与评论家损失中的隐式方差惩罚以及来自时间相关且梯度相似状态的冗余更新联系起来。受这些发现的启发，我们提出了稀疏近端策略优化（SParse Proximal Policy Optimization，SP$^3$O），它在每个响应中仅将价值损失应用于少数彼此分离良好的状态，以缓解这两种效应。在Qwen3-Base上的实验表明，每个响应仅监督三个状态的SP$^3$O可以缓解价值平坦化，并在不同模型规模和评估套件上持续改进学到的策略。总之，我们的结果将价值平坦化识别为标准PPO中评论家学习的一个重要但被忽视的失效模式，并表明一种简单的稀疏监督策略可以缓解它。
cs.LG / 55 / 2609.18878
Preventing Model Collapse: A Fisher-Rao Perspective on the Dynamics of Training with Synthetic Data
防止模型崩溃：Fisher-Rao 视角下使用合成数据训练的动力学
large language model
大语言模型相关
Abstract
Large Language Models (LLMs) are now routinely trained using synthetic data, since high-quality human data has been exhausted by the ever increasing needs of larger and larger models. However, recursive training on synthetic data frequently induces model collapse, a degenerative feedback loop where models progressively forget the true underlying data distribution. Training on a mixture of synthetic and fresh human data is a logical countermeasure and can prevent model collapse. However, it is an open question as to what is the exact minimum required ratio of human-to-synthetic data to maintain training stability. In this paper, we establish rigorous theoretical guarantees on the minimum rate of human data required to prevent model collapse. Although previous work established a formal lower bound for this ratio, such bound can be vacuous for very high dimensions, as the analysis relies on the usual Euclidean metric in R^n and is not adapted to the space of categorical probability distributions. Instead, in this paper we explicitly leverage the information-geometric structure of the probability simplex by analyzing the dynamics of the process under the Fisher-Rao metric. We derive quantitative contraction and invariance bounds that are stable and do not become trivial as the dimensions increase. Thus, we show that the effective required data ratio to prevent model collapse is different than previously implied.
Chinese Translation
大型语言模型（LLMs）现在通常使用合成数据进行训练，因为高质量人类数据已被越来越大模型不断增长的需求耗尽。然而，在合成数据上进行递归训练常常会引发模型崩溃，这是一种退化性反馈回路，其中模型会逐渐遗忘真实的底层数据分布。在合成数据和新鲜人类数据的混合数据上进行训练是一种合乎逻辑的对策，并且可以防止模型崩溃。然而，维持训练稳定性所需的人类数据与合成数据之间的确切最小比例仍是一个未解问题。在本文中，我们建立了关于防止模型崩溃所需的人类数据最低比率的严格理论保证。尽管先前的工作为该比例建立了一个形式化下界，但此类下界在非常高维时可能是空洞的，因为该分析依赖于 R^n 中通常的欧几里得度量，并且并不适配于类别概率分布的空间。相反，在本文中，我们通过分析该过程在 Fisher-Rao 度量下的动力学，显式利用概率单纯形的信息几何结构。我们推导出定量的收缩界和不变界，这些界是稳定的，并且不会随着维度增加而变得平凡。因此，我们表明防止模型崩溃的有效所需数据比例与先前所暗示的不同。
cs.AI / 56 / 2609.18111
A Comprehensive Review of Generative Physical Artificial Intelligence
生成式物理人工智能的综合综述
diffusion
扩散模型相关
Abstract
The integration of large-scale foundation models with physical embodiments has led to significant advancements in robotics known as Generative Physical Artificial Intelligence (GPAI). These agentic AI systems autonomously perceive, reason, and act in complex real-world situations. This survey comprehensively analyzes GPAI systems, focusing on their architectural foundations, current applications, and key limitations. We introduce a taxonomy of five distinct approaches: Robot Foundation Models (RFMs) for cross-platform skill transfer; Vision-Language Action (VLA) models for end-to-end multi-modal perception and control; Large Behavior Models (LBMs) for human-like movement generation; Diffusion Policy Models (DPMs) for diffusion model-based temporally coherent action generation; and World Foundation Models (WFMs) for physics-compliant simulation and data generation. We examine how these approaches complement each other: WFMs generate training data for VLAs and DPMs, RFMs enable cross-platform deployment of learned policies, while LBMs provide motion priors for natural behavior. Through examples across autonomous vehicles, industrial automation, healthcare robotics, and humanoid systems, we identify significant performance improvements and summarize promising research directions in data-efficient learning, sim-to-real transfer, edge-compatible architectures, and safety frameworks. These insights advance embodied AI for IoT-connected environments where intelligent agents interact with networked sensors, actuators, and edge devices.
Chinese Translation
大规模基础模型与物理实体的集成已经在机器人领域带来了重大进展，这种进展被称为生成式物理人工智能（GPAI）。这些智能体式人工智能系统在复杂的现实世界情境中自主感知、推理和行动。本综述全面分析了GPAI系统，重点关注其架构基础、当前应用和关键局限性。我们引入了五种不同方法的分类体系：用于跨平台技能迁移的机器人基础模型（RFMs）；用于端到端多模态感知与控制的视觉-语言-动作（VLA）模型；用于类人运动生成的大型行为模型（LBMs）；用于基于扩散模型的时间连贯动作生成的扩散策略模型（DPMs）；以及用于符合物理规律的仿真和数据生成的世界基础模型（WFMs）。我们考察了这些方法如何相互补充：WFMs为VLAs和DPMs生成训练数据，RFMs使得学习到的策略能够跨平台部署，而LBMs为自然行为提供运动先验。通过跨自动驾驶汽车、工业自动化、医疗机器人和人形系统的示例，我们识别出显著的性能改进，并总结了在数据高效学习、仿真到现实迁移、边缘兼容架构和安全框架方面有前景的研究方向。这些见解推进了面向物联网连接环境的具身人工智能，在这些环境中，智能体与联网的传感器、执行器和边缘设备进行交互。
cs.AI / 57 / 2609.18216
CPR: Combining global composing, local performing and full-sequence refining in piano rendering with continuous autoregressive modelling
CPR：在连续自回归建模的钢琴渲染中结合全局作曲、局部演奏和全序列精炼
diffusion
扩散模型相关
Abstract
Prompt-conditioned piano MIDI-to-Music rendering aims to faithfully render target notes while reproducing the timbre of a reference recording. Existing approaches primarily follow two paradigms: autoregressive (AR) modeling and flow matching (or diffusion). Discrete-codec AR models provide causal temporal modeling, but quantization can discard acoustic detail. Flow matching better preserves acoustic structure in the cost of full-sequence attention costs and worse semantic structure. Continuous autoregressive models operate directly on continuous representations. It not only combines the condition-following ability of AR models and distribution-modeling capacity of flow matching but also bypasses the quantization bottleneck with lower computational costs. Building on this principle, we present Composer--Performer--Refiner (CPR) framework. Composer autoregressively predicts continuous hidden states, Performer generates 24kHz acoustic latents through local flow matching and Refiner then upsamples the waveform to 48 kHz. We further introduce Bottlenecked Representation Alignment (BREPA) and Modality--Time RoPE (MT-RoPE) to strengthen musical semantic structure in Composer hidden states and temporal alignments across modalities. Codes are available at https://github.com/FEAfeatherTHER/CPR_official
Chinese Translation
以提示为条件的钢琴 MIDI-to-Music 渲染旨在忠实渲染目标音符，同时再现参考录音的音色。现有方法主要遵循两种范式：自回归（AR）建模和流匹配（或扩散）。离散编解码器 AR 模型提供因果时间建模，但量化可能丢弃声学细节。流匹配以全序列注意力开销和更差的语义结构为代价，更好地保留了声学结构。连续自回归模型直接作用于连续表示。它不仅结合了 AR 模型的条件遵循能力和流匹配的分布建模能力，还以更低的计算成本绕过了量化瓶颈。基于这一原理，我们提出了 Composer--Performer--Refiner（CPR）框架。Composer 自回归地预测连续隐藏状态，Performer 通过局部流匹配生成 24kHz 声学潜变量，随后 Refiner 将波形上采样至 48 kHz。我们进一步引入瓶颈表示对齐（BREPA）和模态--时间 RoPE（MT-RoPE），以增强 Composer 隐藏状态中的音乐语义结构以及跨模态的时间对齐。代码可在 https://github.com/FEAfeatherTHER/CPR_official 获取。
cs.SE / 58 / 2609.17698
A Large-Scale Empirical Study of Quality Assurance Practices and Gaps in AI Agents
AI 智能体中质量保证实践与缺口的大规模实证研究
large language model
大语言模型相关
Abstract
Large language model (LLM)-based agents are increasingly used across software engineering, web automation, research, and productivity applications. Their integration of planning, memory, tool use, code execution, and external interactions enables greater autonomy but also introduces new reliability, safety, and security risks. We present a large-scale empirical study of quality assurance (QA) practices in 157 open-source LLM-based agent projects with at least 100 GitHub stars. We analyze documentation, source code, configurations, and tests to characterize QA practices across execution surfaces, safeguards, testing artifacts, risk scenarios, and recurring gaps. We find that current QA primarily focuses on basic functionality and high-risk actions, while coverage remains fragmented. Safeguards are inconsistently applied across equivalent execution routes, tests rarely examine boundary, adversarial, or multi-step tool-use failures, and identified risks are seldom translated into end-to-end QA checks. These findings highlight the need to move beyond feature-level testing toward systematic end-to-end validation that ensures agent workflows remain within intended boundaries when interacting with untrusted inputs, tools, persistent state, and external APIs.
Chinese Translation
基于大语言模型（LLM）的智能体正日益广泛地应用于软件工程、Web 自动化、研究和生产力应用中。它们对规划、记忆、工具使用、代码执行和外部交互的集成带来了更强的自主性，但也引入了新的可靠性、安全性和安全性风险。我们开展了一项大规模实证研究，考察了 157 个至少有 100 个 GitHub 星标的开源基于 LLM 的智能体项目中的质量保证（QA）实践。我们分析了文档、源代码、配置和测试，以刻画跨执行表面、防护措施、测试制品、风险场景和反复出现的缺口的 QA 实践。我们发现，当前的 QA 主要聚焦于基本功能和高风险操作，而覆盖范围仍然碎片化。防护措施在等效的执行路径上应用不一致，测试很少考察边界性、对抗性或涉及多步工具使用的失败，而已识别的风险也很少被转化为端到端的 QA 检查。这些发现凸显出有必要超越功能层面的测试，转向系统化的端到端验证，以确保智能体工作流在与不可信输入、工具、持久状态和外部 API 交互时仍保持在预期边界之内。
cs.SE / 59 / 2609.18052
An Empirical Evaluation of Cost-Efficient Large Language Models on Algorithmic Programming Tasks
对成本高效的大型语言模型在算法编程任务上的实证评估
large language model
大语言模型相关
Abstract
This study empirically evaluates whether cost-efficient Large Language Models (LLMs) can be trusted to generate enterprise code to a written specification. Three models (Gemini Flash 3, GPT-5.4 mini and Claude Haiku 4.5) were asked to solve 992 algorithmic problems as Java Spring Boot service methods conforming to a mandated signature and data-transfer-object specification, crossing four model and agentic coding tool combinations with two prompt variants to yield eight configurations, with iteration forbidden and hardcoded answers explicitly prohibited. Eight problem statements were withheld to probe how models respond to missing input. The 7,593 resulting methods were classified by an eight-class outcome taxonomy describing what each does about producing an answer, then deployed and executed, giving 7,936 measured requests joined to that classification. Structural conformance approached ceiling, yet 38.4% of methods do not compute the value they returned and only 12.9% of returned answers were correct. Conditioning on outcome class shows that response reliability and correctness are inversely related, whereas genuinely computing methods answered least often and were correct 19.3%. Limitations include single generation runs per configuration, partial harness coverage, single-pass timing, syntactic classification, and probable corpus contamination.
Chinese Translation
本研究实证评估了成本高效的大型语言模型（LLMs）能否被信任，以依据书面规范生成企业代码。三个模型（Gemini Flash 3、GPT-5.4 mini 和 Claude Haiku 4.5）被要求将 992 个算法问题求解为符合强制签名与数据传输对象规范的 Java Spring Boot 服务方法，将四种模型与智能体式编码工具的组合同两种提示变体交叉，得到八种配置，其中禁止迭代，并明确禁止硬编码答案。有八个问题陈述被隐去，以探究模型如何应对缺失的输入。由此产生的 7,593 个方法依据一个八类结果分类法进行了归类，该分类法描述每个方法在产生答案方面做了什么，随后这些方法被部署并执行，得到与上述分类相关联的 7,936 个被测请求。结构符合度接近上限，然而 38.4% 的方法并不计算它们所返回的值，且仅有 12.9% 的返回答案是正确的。以结果类别为条件进行分析表明，响应可靠性与正确性呈反向关系，而真正进行计算的方法作答频率最低，其正确率为 19.3%。局限性包括每种配置仅进行单次生成运行、测试框架覆盖不完整、单次计时、句法层面的分类，以及可能存在的语料污染。
cs.SE / 60 / 2609.18311
A Study on the Impact of Natural Language Differences in Prompts on Automatic Code Generation Using LLMs
关于提示中自然语言差异对使用LLM的自动代码生成影响的研究
large language model
大语言模型相关
Abstract
Large Language Models (LLMs) have demonstrated remarkable performance in automatic code generation tasks, thereby encouraging new research in this area. Although numerous studies have explored LLM-based code generation, the impact of the natural language in input prompts remains unexplored (language bias). This study aims to (1) quantify how the natural language of input prompts influences LLM-based code generation performance and (2) evaluate a mitigation strategy to reduce language bias in code generation. We assess code generation Accuracy on AtCoder, LeetCode, and BigCodeBench. To quantify the language bias on code generation, each problem is presented in English, Japanese, and Chinese. We use seven LLMs (GPT-4o, o3-mini, DeepSeek-V3.2, Llama-3, Qwen2.5-Coder-14B, Qwen2.5-Coder-0.5B, and GitHub Copilot) and assess their performance in terms of Accuracy (the number of problems for which generated code passes all test cases). We compare Accuracy before and after translation to evaluate the effectiveness of translation as a mitigation strategy. We observed that the natural language of problem statements affects LLM-based code generation performance. Specifically, the languages officially supported by each dataset achieved the highest median Accuracy. Also, translation improved Accuracy, but its effectiveness was not consistent across datasets and model types. We found that AtCoder contained a particularly high proportion of narrative-style problem statements and longer problem statements. Natural language significantly affects LLM code generation accuracy. Translation can mitigate language bias in some settings, but its effectiveness depends on the dataset and model type. Furthermore, the narrative aspects and context length of input prompts are important factors related to language bias and the effectiveness of translation as a mitigation strategy.
Chinese Translation
大型语言模型（LLMs）在自动代码生成任务中表现出色，从而促进了该领域的新研究。尽管大量研究已经探索了基于LLM的代码生成，但输入提示中自然语言的影响仍未得到探索（语言偏差）。本研究旨在（1）量化输入提示的自然语言如何影响基于LLM的代码生成性能，以及（2）评估一种减轻代码生成中语言偏差的缓解策略。我们在AtCoder、LeetCode和BigCodeBench上评估代码生成的准确率（Accuracy）。为了量化代码生成中的语言偏差，每个问题分别以英语、日语和中文呈现。我们使用七个LLM（GPT-4o、o3-mini、DeepSeek-V3.2、Llama-3、Qwen2.5-Coder-14B、Qwen2.5-Coder-0.5B和GitHub Copilot），并以准确率（生成的代码通过所有测试用例的问题数量）评估它们的性能。我们比较翻译前后的准确率，以评估翻译作为一种缓解策略的有效性。我们观察到，问题陈述的自然语言会影响基于LLM的代码生成性能。具体而言，每个数据集官方支持的语言取得了最高的中位数准确率。此外，翻译提高了准确率，但其有效性在不同数据集和模型类型之间并不一致。我们发现，AtCoder包含比例特别高的叙事风格问题陈述和更长的问题陈述。自然语言显著影响LLM代码生成准确率。翻译可以在某些设置中减轻语言偏差，但其有效性取决于数据集和模型类型。此外，输入提示的叙事方面和上下文长度是与语言偏差以及翻译作为缓解策略的有效性相关的重要因素。
cs.AI / 61 / 2609.17989
Whom Do AI Agents Work For? Role Assignment Induces Sponsorship Bias in LLM Recommenders
AI 智能体为谁工作？角色指派在 LLM 推荐器中诱发赞助偏误
large language model
大语言模型相关
Abstract
Large language models (LLMs) now serve as conversational shopping assistants on platforms that also sell advertising. These AI agents face a conflict of duty. They advise consumers who rely on their judgment, yet are deployed by platforms that benefit when sponsored listings are chosen. Sponsorship disclosures, designed to allow consumers to penalize paid placements, now reach the AI agent rather than the consumer, and the agent's evaluation of them is hidden from the consumer. Drawing on the fiduciary concept of conflict of duty, we argue that an agent's evaluation of a sponsored listing should not depend on which party deployed it. In controlled choice experiments, we manipulate assigned roles in the system prompt to name either a traveler or a booking platform as the agent's principal. Platform delegation significantly attenuates the penalty that agents apply to sponsored listings and weakens the skepticism that disclosure triggers in their reasoning traces. We replicate out findings across LLMs and reasoning depths. A second study decomposes the disclosure label and shows that the divergence between the two delegates widens significantly when the paid placement is attributed to the platform. Stricter terminology ("Sponsored" instead of "Promoted") lowers choice of paid listings but does not close this gap when the platform is named. The findings show that disclosure mandates designed for human consumers cannot by themselves protect consumers in AI-mediated commerce.
Chinese Translation
大语言模型（LLM）如今在同时销售广告的平台上充当对话式购物助手。这些 AI 智能体面临义务冲突。它们为依赖其判断的消费者提供建议，却由那些在赞助列表被选中时获益的平台所部署。赞助披露本意是让消费者能够对付费投放施加惩罚，如今却传递给了 AI 智能体而非消费者，而智能体对这些披露的评估对消费者不可见。借鉴信义义务中的义务冲突概念，我们认为智能体对赞助列表的评估不应取决于由哪一方部署了它。在受控选择实验中，我们操纵系统提示中指派给智能体的角色，将旅行者或预订平台指定为智能体的委托人。平台委托显著减弱了智能体对赞助列表施加的惩罚，并削弱了披露在其推理轨迹中引发的怀疑。我们在不同 LLM 和推理深度上重复了我们的发现。第二项研究对披露标签进行分解，表明当付费投放被归因于平台时，两类委托人之间的分歧显著扩大。更严格的术语（用"Sponsored"而非"Promoted"）降低了付费列表的选择率，但当平台被指定为委托人时，并不能弥合这一差距。这些发现表明，为人类消费者设计的披露强制要求本身无法在 AI 中介的商务中保护消费者。
cs.CL / 62 / 2609.17981
Encoder Awakening via Adapters: Effective Domain-Adaptive Fine-tuning of Speech-LLMs
通过适配器唤醒编码器：语音大语言模型的有效领域自适应微调
large language model
大语言模型相关
Abstract
Speech Large Language Models (Speech-LLMs), typically built from a pre-trained speech encoder, a modality projector, and an LLM fine-tuned with Low-Rank Adapters (LoRA), have shown strong Automatic Speech Recognition (ASR) performance on general-domain speech. However, adapting them to domain-shifted speech, such as child or dialectal speech, remains challenging under limited target-domain data. Given the dominant role of the LLM in Speech-LLMs, with cross-entropy loss applied only at the LLM output, the speech encoder may receive insufficient adaptation to new acoustic conditions. In this paper, we propose Encoder Awakening via Adapters (EAVA), a simple yet effective domain-adaptive fine-tuning method for Speech-LLM-based ASR. First, lightweight adapters are inserted into each encoder layer and trained exclusively, enabling target-domain acoustic knowledge to be incorporated into the encoder while preserving its pre-trained knowledge. Second, the full model is jointly fine-tuned on the target domain with LoRA applied to the LLM. Experiments on three domain-shifted ASR datasets, covering child and dialectal speech, show that EAVA consistently outperforms vanilla fine-tuning and other baselines, achieving new state-of-the-art performance.
Chinese Translation
语音大语言模型（Speech-LLMs）通常由预训练的语音编码器、模态投影器以及使用低秩适配器（LoRA）微调的大语言模型（LLM）构建而成，已在通用领域语音上展现出强大的自动语音识别（ASR）性能。然而，在目标领域数据有限的情况下，将它们适配到领域偏移语音（如儿童语音或方言语音）仍然具有挑战性。鉴于LLM在Speech-LLM中起主导作用，且交叉熵损失仅应用于LLM输出端，语音编码器可能对新声学条件适应不足。在本文中，我们提出通过适配器唤醒编码器（Encoder Awakening via Adapters，EAVA），一种简单而有效的、面向基于Speech-LLM的ASR的领域自适应微调方法。首先，将轻量级适配器插入每个编码器层并仅对其进行训练，使目标领域声学知识能够被纳入编码器，同时保留其预训练知识。其次，在目标领域上对完整模型进行联合微调，并对LLM应用LoRA。在三个涵盖儿童语音和方言语音的领域偏移ASR数据集上的实验表明，EAVA始终优于普通微调和其他基线，并取得了新的最先进性能。
cs.LG / 63 / 2609.18759
Stable Filters for Generative Modeling of Graph Signals
用于图信号生成建模的稳定滤波器
diffusion
扩散模型相关
Abstract
Generating signals on graphs requires permutation-equivariant models that exhibit stability with respect to relative structural perturbations. While recent graph-aware Schrödinger bridge models incorporate topology information directly into their reference dynamics, it is unclear how perturbations of the graph propagate through these dynamics and affect the resulting generated distributions. In this paper, we analyze the structural stability of graph-aware continuous-time generative models whose drift combines a graph filter with a learned graph neural network. We derive explicit Wasserstein stability bounds that quantify the effect of relative graph perturbations on the generated distributions. Motivated by these bounds, we introduce a principled framework for designing stable graph filters that preserve the smoothing behavior of graph heat diffusion, while boosting structural stability. Experiments on synthetic and fMRI signals show our stable filters enhance structural robustness while matching or exceeding the generative quality of the heat equation baseline.
Chinese Translation
在图上的信号生成需要具有置换等变性的模型，这些模型相对于相对结构扰动表现出稳定性。尽管近期的图感知 Schrödinger 桥模型将拓扑信息直接纳入其参考动力学中，但图的扰动如何通过这些动力学传播并影响最终生成的分布仍不清楚。在本文中，我们分析图感知连续时间生成模型的结构稳定性，这类模型的漂移项将图滤波器与学习得到的图神经网络相结合。我们推导出显式的 Wasserstein 稳定性界，用以量化相对图扰动对生成分布的影响。受这些界的启发，我们提出了一个有原则的框架，用于设计稳定图滤波器，这些滤波器在提升结构稳定性的同时保持图热扩散的平滑行为。在合成信号和 fMRI 信号上的实验表明，我们的稳定滤波器增强了结构鲁棒性，同时达到或超过了热方程基线的生成质量。
cs.LG / 64 / 2609.18928
Comprehensive reconstruction of collider events with hypergraph representation learning and graph-conditioned diffusion
基于超图表示学习与图条件扩散的对撞机事件综合重建
diffusion
扩散模型相关
Abstract
In particle collider experiments, event reconstruction is the task of inferring the kinematics of short-lived particles produced in the hard scatter from the stable final states recorded by detectors. We decompose event reconstruction into two primary tasks: assigning measured jets and charged leptons to parent particles, and predicting unmeasured neutrino kinematics. We present VyPER, a novel geometric learning framework that represents collider events as hypergraphs with a physics-inspired topology. VyPER combines the supervised classification of hyperedges for particle assignment with a diffusion model for predicting neutrino kinematics, leveraging a joint loss function to optimize both reconstruction tasks within a unified framework. We showcase VyPER across several proton-proton collision processes, comparing its performance to existing analytical and machine-learning-based reconstruction techniques. In doing so, we demonstrate that accurate event reconstruction is achievable across a diverse range of Standard Model physics processes, opening new avenues for precision measurements in the Higgs boson, electroweak, and top-quark sectors.
Chinese Translation
在粒子对撞机实验中，事件重建的任务是根据探测器记录到的稳定末态，推断硬散射过程中产生的短寿命粒子的运动学信息。我们将事件重建分解为两个主要任务：将测量到的喷注和带电轻子指派给母粒子，以及预测未被测量的中微子运动学。我们提出 VyPER，一种新颖的几何学习框架，它将对撞机事件表示为具有物理启发式拓扑的超图。VyPER 将用于粒子指派的超边监督分类与用于预测中微子运动学的扩散模型相结合，利用联合损失函数在统一框架内优化这两项重建任务。我们在若干质子-质子对撞过程中展示了 VyPER，并将其性能与现有的解析方法和基于机器学习的重建技术进行了比较。借此我们证明，在多样化的标准模型物理过程中实现精确的事件重建是可行的，这为希格斯玻色子、电弱以及顶夸克领域的精密测量开辟了新途径。
cs.AI / 65 / 2609.17777
Information Set Emulation: Causal Certificates for AI Derived EHR Features
信息集仿真：AI衍生EHR特征的因果证书
large language model
大语言模型相关
Abstract
AI and large language models can recover clinically meaningful features from electronic health records (EHRs), but predictive usefulness does not establish admissibility for causal inference. We introduce information set emulation: an AI typed lift attaches source evidence, clinical and recording times, decision-time availability, representation version, proposed causal roles, and unresolved ambiguity to extracted features under a locked target trial. Causal certificates record auditable evidence for those roles. Features with unresolved downstream roles are routed to compatible reporting or separate analyses. Typed evidence defines an observational fiber of causal worlds consistent with the observed law. The locked scalar estimand maps this fiber to a compatible image whose squared Chebyshev radius equals the residual minimax mean squared error when the image is nonempty and compact. This classical identity provides a target-specific measure of information ambiguity. The contribution is its integration with a joint EHR observation map and an auditable certificate architecture. Under explicit exchangeability, positivity, and nuisance-consistency conditions, we give identification and cross-fitted augmented inverse probability weighted estimation, distinguishing empirical and population targets. An EHR compression-drift identity separates the roles of frame presence, treatment assignment, and outcome observation. Artificial simulations and a common-law finite-world example illustrate estimation failures and information-radius reduction. Synthetic Phase 0 notes demonstrate audit diagnostics; a separate role-specific analysis spread illustrates routing and is not an exact fiber radius. All experiments are synthetic. The framework specifies when reconstructed information can support a point claim and when compatible reporting is required.
Chinese Translation
AI和大语言模型可以从电子健康记录（EHRs）中恢复具有临床意义的特征，但预测有用性并不能确立因果推断的可采纳性。我们引入信息集仿真：一种AI类型化提升在锁定的目标试验下，将来源证据、临床时间与记录时间、决策时可用性、表示版本、提议的因果角色以及未解决的模糊性附加到提取的特征上。因果证书记录这些角色的可审计证据。具有未解决下游角色的特征被路由至相容报告或单独分析。类型化证据定义了一个与观测律一致的因果世界的观测纤维。锁定的标量估计目标将此纤维映射到一个相容像，当该像非空且紧致时，其平方Chebyshev半径等于残差极小极大均方误差。这一经典恒等式提供了一个目标特定的信息模糊性度量。本文的贡献在于其与联合EHR观测映射以及可审计证书架构的整合。在显式的可交换性、正值性和干扰项一致性条件下，我们给出识别以及交叉拟合的增广逆概率加权估计，并区分经验目标和总体目标。一个EHR压缩-漂移恒等式分离了框架存在、处理分配和结果观测的角色。人工模拟和一个普通法有限世界例子说明了估计失败与信息半径缩减。合成第0阶段笔记展示了审计诊断；一个单独的角色特定分析散布说明了路由，并且不是精确的纤维半径。所有实验均为合成。该框架规定了重建信息何时可以支持点主张，以及何时需要相容报告。
cs.LG / 66 / 2609.17823
METALICA: METAdynamics and repLICA exchange for enhanced diffusion sampling
METALICA：用于增强扩散采样的元动力学与副本交换
diffusion
扩散模型相关
Abstract
Many proteins function through transitions between conformational states, yet rare states are rarely sampled by diffusion models trained on an equilibrium ensemble, demanding better sampling methods. We introduce METALICA, which implements Metadynamics on a pretrained diffusion model via Replica Exchange. It accumulates a bias potential along a Collective Variable, repels new samples from previous ones through biased sampling, and reweights samples onto the unbiased distribution. METALICA holds one replica per diffusion level, forming a Markov Chain that evolves through inter-replica communication and is refined in place as the bias grows. METALICA is the dual of sequential control, in which Sequential Monte Carlo parallelizes the sampler over a batch of particles. Parallelism over the levels of the diffusion-time schedule instead allows METALICA to generate samples from long chains, essential for the discovery of rare events, with accuracy set by run length rather than by the memory available. We validate on a bimodal target with known free energies, then apply METALICA to the unfolding of a protein. At a budget for which sequential control yields no unfolded structure, METALICA populates the basin and resolves a second free energy minimum.
Chinese Translation
许多蛋白质通过构象状态之间的转换来发挥功能，然而，在平衡系综上训练的扩散模型很少采样到稀有状态，这要求更好的采样方法。我们提出 METALICA，它通过副本交换在预训练扩散模型上实现元动力学。它沿一个集体变量累积偏置势，通过偏置采样使新样本远离先前的样本，并将样本重新加权到无偏分布上。METALICA 在每个扩散层级持有一个副本，形成一个马尔可夫链，该链通过副本间通信演化，并随着偏置增长而原地细化。METALICA 是顺序控制的对偶，其中顺序蒙特卡洛将采样器在一批粒子上并行化。相反，在扩散时间日程的各个层级上进行并行，使 METALICA 能够从长链生成样本，这对稀有事件的发现至关重要，其精度由运行长度而非可用内存决定。我们在一个具有已知自由能的双峰目标上进行了验证，然后将 METALICA 应用于蛋白质的去折叠。在一个顺序控制无法产生任何去折叠结构的预算下，METALICA 填充了该盆地并解析出第二个自由能极小值。
人工智能 (cs.AI)
89
cs.AI / 1 / 2609.18025
Physics-Informed Neural Networks for Fast Multilayer Spectral Inversion of Hα 6562.8 A and Ca II 8542.1 A Spectra
Abstract
Strong chromospheric absorption lines such as H$α$ 6562.8 A and Ca II 8542.1 A provide vital diagnostics of plasma dynamics and thermal structure in the solar chromosphere. Multilayer spectral inversion (MLSI) offers a physically interpretable framework for modeling these lines using a finite number of radiative-transfer layers, but conventional MLSI relies on pixel-by-pixel nonlinear least-squares fitting, making it computationally expensive for large imaging spectroscopic data sets. Here, we introduce a physics-informed neural-network (PINN) framework to accelerate MLSI while preserving its analytic radiative-transfer formulation. The network predicts MLSI parameters directly from observed line profiles and passes them through a differentiable MLSI forward model to synthesize spectra. Training follows a two-stage approach: an initial stage optimized solely via spectral reconstruction loss, followed by fine-tuning that combines spectral consistency with parameter-space supervision from conventional MLSI results on a single reference image. This strategy eliminates the need for large precomputed training sets while maintaining physical interpretability. Applied to Fast Imaging Solar Spectrograph (FISS) observations from the Goode Solar Telescope (GST) targeting both quiet-Sun and active-region regions, MLSI-PINN parameter maps reproduce the primary spatial structures of direct inversions, achieving an arithmetic mean pixel-wise Pearson correlation coefficient of 0.933 across all evaluated parameters. The reconstructed spectra closely match both observed profiles and conventional MLSI fits. Post-training, MLSI-PINN processes a raster in approximately 5-15 seconds compared to 3-5 minutes for conventional MLSI, delivering an inference speedup of about 12-60 times without substantial loss in reconstruction quality, enabling efficient MLSI analysis on large chromospheric data sets.
cs.AI / 2 / 2609.17688
CapMem: A Benchmark for Caption-Based Episodic Memory in Egocentric Video
Abstract
Wearable assistants require episodic memory over egocentric video, yet current vision-language models face bounded frame budgets, growing visual-token costs, and long-context retrieval failures. Under these practical constraints, we study whether textual captions can serve as reusable episodic memory. We define the Episodic Memory Video Caption QA task and introduce CapMem, a human-annotated benchmark with 75 videos totaling 33.7 hours, and 1,000 multiple-choice questions across 16 scenarios. On long videos (>20 min), full-coverage CaptionQA with 30s and 60s caption windows outperforms direct VideoQA for 10/12 and 8/12 models, respectively. On the same video subset, a matched-frame control across six Qwen models retains mean accuracy gains of 3.22 and 2.55 points, respectively. Our caption-guided retrieve-and-verify harness further improves accuracy by up to 5.3 points. These results support the effectiveness of caption memory for episodic reasoning over long egocentric video.
cs.AI / 3 / 2609.17699
NeMo Data Designer: An Extensible Framework for Multimodal Synthetic Data Generation
Abstract
We present NeMo Data Designer (NDD), an open-source, general-purpose framework for multi-modal synthetic data generation (SDG). Designed to be intuitive to use, NDD provides a declarative configuration format in which human and/or agent users define each dataset column, with column types spanning text, code, structured outputs, images, embeddings, and statistical samplers that are explicitly configured to steer dataset diversity. Additional column types and functionality can be introduced using the framework's flexible plugin system. NDD's configuration is an inspectable artifact, supporting workflow sharing and reproducibility. SDG is an inherently iterative process. NDD therefore builds a preview-and-revision loop into its core workflow, allowing users to generate and inspect a small number of records, refine the specification, and rerun generation at full scale. At runtime, NDD resolves dependencies, schedules calls to user-provided model endpoints, and retries failed requests. We describe NDD's architecture and programming model and present case studies spanning structured, agentic, multimodal, and domain-specialized tasks, including datasets used in Nemotron model development and in production enterprise deployments.
cs.AI / 4 / 2609.17731
A Systematic Evaluation of the COTQ Provincial Land Cover Product: Structural Consistency, Spectral Separability, and Relative Positioning Against ESA, ESRI, and Google Products
Abstract
High-resolution land use and land cover (LULC) products derived from Sentinel-2 imagery are widely used for environmental monitoring and land management, yet their performance can vary across regions with complex ecological gradients and heterogeneous surface conditions. In Quebec, these limitations motivated the development of a provincial 10-m land-cover product, the COTQ, designed to support annual monitoring of land occupation and soil artificialisation. This study presents a systematic evaluation of the COTQ product relative to three global 10-m LULC datasets: ESA WorldCover, ESRI LandCover, and Google DynamicWorld. This paper does not introduce a new mapping methodology but focuses on analysing the behaviour and consistency of the COTQ using complementary evaluation approaches. All products are harmonized under a common legend and compared using structural indicators (object-size distributions, shape complexity, Adjusted Rand Index, and Intersection over Union), spectral separability metrics derived from Sentinel-2 reflectance data, and a targeted photo-interpretation of disagreement areas. The analysis is conducted over eight Sentinel-2 tiles selected to represent the main bioclimatic domains of Québec, from temperate and boreal forests to northern tundra environments. The results show that the COTQ exhibits structural and spectral characteristics most similar to ESA WorldCover among the reference global products, while revealing systematic differences linked to class definitions and thematic priorities, particularly for urban areas, wetlands, and rocky or cryptogamic surfaces. This multi-criteria evaluation provides an objective characterization of the COTQ product and clarifies its relative positioning with respect to existing global land-cover datasets for operational land monitoring in Québec.
cs.AI / 5 / 2609.17757
Imitation Learning for Autonomous Driving in CARLA
Abstract
Behavioral cloning trains a policy offline on expert demonstrations, but deployment is closed loop: each action affects the observations the policy receives next. We study how much closed-loop driving competence a compact multimodal policy can acquire from offline demonstrations in the CARLA simulator. The policy uses five-frame histories of RGB images, LiDAR, vehicle telemetry, and lane waypoints to predict throttle, brake, and steering at 20 Hz. Demonstrations were collected in three stages, ending with a systematic route-generation procedure that enumerates spawn points and feasible maneuvers and verifies completed autopilot routes. The released 1.36 million parameter policy was trained on 236,882 windows, representing about 3.3 hours of driving from 448 captures. The resulting policy drives autonomously for hours on training and held-out routes. In our runs, it did so without collisions and also transferred qualitatively to an unseen CARLA town with different road geometry. We also observed recovery from large trajectory deviations, although we do not claim systematic recovery without controlled evaluation. We report offline metrics and distinguish measured results from qualitative closed-loop observations. We release the code, trained checkpoint, ONNX model, data sample, and an evidence audit for the reported claims.
cs.AI / 6 / 2609.17775
SAGE: Governed Artifact Generation from Enterprise Guidelines
Abstract
Enterprise guideline documents mix narrative text, complex tables, and embedded images, and converting them into structured work artifacts still takes two to three days of manual effort each. Current language and vision-language models extract from such documents but offer no governed workflow beyond extraction: no validation, no consistency checking, no traceable artifact generation. We introduce SAGE, a governed multi-stage LLM pipeline organized around a shared versioned rule store with stable identifiers, schema-validated inter-stage contracts, and end-to-end provenance tracking. Extracted rules undergo deterministic structural validation and LLM-based semantic scoring, then a consistency module that removes duplicates, flags contradictions, and surfaces specification gaps; only uncertain or flagged items reach reviewers, while high-confidence outputs are auto-approved. On 120 documents, SAGE cuts turnaround from days to 20-100 minutes, achieving a 96% document-level success rate with 3.2% hallucination, extracting 3,896 rules and producing 812 artifacts ready for human review; without governance, hallucination rises to 15.7%.
cs.AI / 7 / 2609.17786
FairCompressAgent: An Agentic Framework for Fairness-Aware Model Compression for FPGA Deployment
Abstract
Fairness-aware model compression requires selecting methods and configurations that balance accuracy, fairness, and deployment cost. These decisions become more difficult when compression methods are composed or the user's requirements change. In this paper, we propose FairCompressAgent (FCA), an agentic framework that integrates fairness-aware pruning, incremental quantization, and sparse low-rank factorization through a common operator interface. A language-model planner uses model profiles and measured outcomes to select compression configurations, while an execution layer performs compression, fine-tuning, evaluation, and constraint-based selection. FCA also supports requirement updates and reports the remaining violation when a request cannot be satisfied. Experiments on Fitzpatrick-17k with VGG-11 compare four search methods over 40 measured configurations. Under the accuracy-constrained request, FCA selects a compressed model with 59.54% less inference tensor storage, while validation average precision increases from 0.5141 to 0.5233 and equalized opportunity (EOpp) decreases from 0.2251 to 0.2168. It reaches the same final selection as one-shot planning with 7.33 versus 12 candidate evaluations on average, under their respective stopping policies. Repeated fine-tuning, held-out testing, and online requirement updates characterize the stability and interactive use of this compression workflow. The results demonstrate how measured feedback and explicit constraints support the selection and interactive refinement of fairness-aware compression configurations.
cs.AI / 8 / 2609.17847
Learning Heterogeneous Preferences
Abstract
Learning from human feedback has become a central paradigm for training modern AI systems, where models of human utility are used as reward models in policy learning. Existing methods typically assume a \emph{universal utility} function shared across a population and treat disagreement between annotators as stochastic variation. While suitable for objective tasks, this assumption breaks down in subjective domains where preferences vary systematically across individuals. We study the problem of subjective preference learning, in which observed choices arise from heterogeneous but internally consistent utility functions. Drawing upon rational choice theory, RCT \parencite{tversky1981framing}, we introduce \emph{individuated utility} functions conditioned on both the individual and their decision context, and propose a novel multi-stage architecture for estimating them from multi-modal data. We evaluate our framework on a newly collected dataset of more than $575{,}000$ pairwise aesthetic judgments from $2{,}398$ participants comparing automotive wheel designs. Our experiments show that individuated utility models substantially outperform universal utility models including foundation model baselines. Our results demonstrate that disagreement reflects meaningful preference heterogeneity rather than annotation noise. More broadly, our findings highlight the importance of collecting annotator attributes and learning individuated utility functions, enabling reward models that explicitly account for whose preferences they represent and faithfully capture human decision diversity.
cs.AI / 9 / 2609.17855
SNOMED CT Concept Recommendation from Masked Clinical Context
Abstract
Standardizing clinical language to SNOMED CT supports interoperability, analytics, and reusable phenotyping, but concept recommendation remains difficult when relevant concepts are rare or absent from training data. We present a masked-concept recommendation benchmark using the SNOMED CT Entity Linking Challenge v1.2.1 data derived from MIMIC-IV-Note. The dataset contains 75,491 annotations across 272 discharge summaries, with 204 notes used for training and 68 for historical testing. For each unique note-concept pair, the target mention is masked from a local clinical context and the system ranks SNOMED CT concepts observed during training. We compare a popularity baseline, sparse TF-IDF concept prototypes, dense latent semantic analysis embeddings, sparse-dense fusion, retrieved-note evidence, and a retrieval-augmented hybrid. Sparse TF-IDF performs best, achieving Recall@1 of 14.81%, Recall@10 of 33.43%, MRR of 0.2114, and nDCG@10 of 0.2297. Retrieval augmentation does not improve this baseline, with Recall@10 of 31.99% and MRR of 0.1937. Performance is strongly affected by concept frequency: Recall@10 is 7.74% for concepts appearing in only one or two training notes versus 43.90% for concepts appearing in more than ten. In addition, 9.66% of test note-concept pairs contain concepts unseen during training. These findings show that local lexical context and terminology coverage are major determinants of recommendation quality in low-resource settings and provide a reproducible baseline for future ontology-grounded and biomedical-encoder retrieval systems.
cs.AI / 10 / 2609.17863
The Inference Engineering Pareto Atlas: Which Optimizations Dominate the Cost, Quality, and Latency Frontier?
Abstract
LLM inference optimizations report speedups on different models, GPUs, prompts, and quality metrics, making them hard to compare or combine. We build a cost, quality, and latency Pareto atlas to identify the best configurations for different deployment constraints. Since exhaustive testing is impractical, we measure 54 configurations of Qwen2.5-7B-Instruct running on vLLM 0.12 across L4, A100, and H100 GPUs and use these anchors to calibrate a simulator. It reproduces measurements at anchored batch sizes, with cross campaign drift below 1.5 percent. A separate quality evaluation tests FP16, AWQ 4bit, FP8 weights, and FP8 KV cache on 200 GSM8K questions with five examples per prompt. Sparse attention is evaluated only in simulation. On the calibrated grid, 18 of 36 configurations reach the Pareto frontier. Combined methods reach it more often than individual methods, with 9 of 15 combinations versus 9 of 21 single methods. Quality testing changes the winners. AWQ 4bit reduces per token latency to 0.34 times baseline on L4 but loses 5.9 percent of strict GSM8K accuracy, narrowly missing the 95 percent quality floor within sampling uncertainty. Flexible answer extraction matches FP16 accuracy, suggesting the loss comes from formatting rather than arithmetic. FP8 weights retain 99.4 percent of baseline accuracy at 0.61 to 0.65 times baseline latency across all three GPUs and appear in three of four regime winners. A naive FP8 KV cache maintains normal throughput but answers none of the 200 questions correctly, showing why speed alone is insufficient. Under two prompt designs, n gram speculative decoding measures at 0.90 to 0.98 times baseline and adds no benefit on this stack. The best choice depends on the constraint and GPU: H100 wins for tight latency, while A100 wins for throughput and low cost at 0.106 dollars per million tokens.
cs.AI / 11 / 2609.17865
Do Frontier Models Seek Safety Evidence Before Acting?
Abstract
Frontier models are often evaluated on how they respond to safety information once it is already in context. We study an earlier decision point: whether models choose to acquire safety-relevant evidence before acting. We introduce SAFE, a controlled benchmark in which models make deployment decisions with optional evidence that varies in retrieval cost, probability, severity, and presentation. Across GPT-5.5, o3, Claude Opus 4.8, and Claude Sonnet 4.6, we find distinct evidence-acquisition policies: Opus inspects nearly by default, o3 is the most skip-heavy and threshold-sensitive, and GPT-5.5 and Sonnet occupy intermediate regimes. Inspection increases strongly with severity and decreases with retrieval cost, whereas probability has much weaker behavioral influence: increasing the stated likelihood of a problem from 10% to 70% changes inspection by at most 21 percentage points. Despite these differences, Stage 1 rationales are dominated by expected-value reasoning across models. A cost-obligation decomposition further shows that avoidance is driven primarily by retrieval friction and explicit threats to the deployment payoff rather than by the remediation duties created by knowing. Counterfactual interventions reveal a further mismatch between behavior and explanation: evidence framing can strongly change decisions near the inspection boundary while going largely unmentioned, whereas probability is frequently cited despite having little causal influence. These results suggest that deployment-time safety depends not only on how models respond to known risks, but also on whether they acquire the evidence needed to know that acting is safe.
cs.AI / 12 / 2609.17885
ERPBench: A State-Grounded Evaluation Paradigm for Computer-Use Agents in Enterprise Software
Abstract
Computer-use agents that operate through screenshots and simulated actions are advancing rapidly, yet their evaluation remains anchored to general desktop and web tasks. Enterprise Resource Planning (ERP) systems run the finance, procurement, inventory, and customer operations of organizations worldwide, and pose distinct challenges for computer-use agents: dense interfaces, coordinated multi-step interactions, and errors that alter persistent business records rather than surfacing on screen. Existing enterprise benchmarks rely on proprietary platforms or on simulated approximations of such software. We introduce ERPBench, a benchmark that evaluates screenshot-only agents on a live and reproducible ERP system and scores each task against ground-truth values in its database. Beyond the benchmark, we present a production-grade harness that gates agent actions behind human approval for safe deployment, which ERPBench runs autonomously. Evaluating six closed and open-source agents, we demonstrate that strong general GUI performance does not transfer to enterprise reliability. Even when an agent reaches the right form and saves it, the stored record is often wrong: some agents save in up to 85% of runs but write the correct value in as few as 3%. We further characterize failure modes specific to enterprise workflows.
cs.AI / 13 / 2609.17890
OBC-Prune: Outcome-Based Calibration for Large Reasoning Model Pruning
Abstract
Large reasoning models (LRMs) generate long chain-of-thought traces before answering, creating significant inference overhead. Pruning can reduce this cost, but its effectiveness depends on the calibration data used to estimate parameter importance. Recent work calibrates on the model's own rollouts instead of generic dataset, but treats all reasoning tokens uniformly, regardless of whether they contribute to successful reasoning. As a result, pruning protects weights by statistical salience rather than by their contribution to correct reasoning, so weights behind erroneous computation survive as readily as those behind correct computation. These erroneous patterns then get carried into the pruned model, degrading reasoning quality, producing both lower accuracy and longer reasoning traces. We propose Outcome-Based Calibration for Large Reasoning Model Pruning (OBC-Prune) to close this gap. OBC first constructs difficulty-matched pairs of correct and incorrect rollouts from problems the model answers inconsistently. It then estimates the causal importance of each reasoning sentence through intervention-based analysis, quantifying how removing its influence affects subsequent predictions. These causal importance scores are converted into per-token weights that rescale the calibration activations used by one-shot pruning methods (SparseGPT, Wanda, ALPS), without modifying the underlying pruning algorithms. Experiments on DeepSeek-R1-Distill-Qwen 1.5B, 7B, and 14B models at 40\% and 50\% sparsity demonstrate consistent improvements over state-of-the-art calibration baselines across most model sizes and sparsity levels on MATH500, LiveCodeBench, and AIME 2025. These results indicate that preserving causally important reasoning circuits is a substantially more effective pruning objective than uniformly preserving observed activations.
cs.AI / 14 / 2609.17921
Collaborative Memory for Multi-Agent VLM Systems
Abstract
Vision-language model (VLM) agents combine specialized perception, tools, and reasoning to address complex visual tasks. In multi-agent settings, different agents inspect different image regions, video frames, or visual representations, so collaboration extends beyond distributed reasoning to distributed perception. This makes shared visual context a central problem in VLM agent collaboration. In this paper, we frame memory hierarchy, cross-agent sharing, and consistency mechanisms around the need to reconcile interpretations and update dependent reasoning. Effective collaboration requires agents to build on contributions from other agents, recover missing visual context, and reconcile differing interpretations as new evidence emerges. Shared visual memory preserves not only images or textual summaries but also the dependencies among observations, agent interpretations, and subsequent reasoning. Together, these design considerations shape how information flows and evolves across VLM agents. The proposed framework provides a foundation for building reliable and resource-efficient agent teams.
cs.AI / 15 / 2609.17965
Measuring AI Leadership: Development and Validation of a Multidimensional Measure for AI-Native Organizations
Abstract
AI is changing what leaders must judge, explain, learn, and coordinate, yet existing measures do not capture these behaviors at the level needed to study leadership in AI-enabled work. We develop the AI Leadership Battery, which organizes 36 behaviorally specific subdimensions into 11 theory-specified content families. Following established scale-development procedures, the research used deductive item generation; content validation of definitional correspondence and definitional distinctiveness; exploratory factor analysis and item reduction; confirmatory factor analysis in independent samples; and tests of internal consistency reliability, convergent validity, discriminant validity, and criterion-related and incremental validity. Across the development and validation studies, the analyses provided evidence for the Battery's content, multidimensional structure, reliability, and distinction from selected orbiting constructs. The Battery also contributed additional information beyond orbiting constructs across organizational growth, decision speed, customer/stakeholder response capability, AI-enabled team performance, AI-enabled work experience, AI security and risk management, and AI adoption and integration. The resulting measure provides researchers with a behavioral framework for examining how leaders in AI-enabled work regulate judgment, learning, adaptation, transparency, and accountability.
cs.AI / 16 / 2609.17969
Memory Has Geometry: Non-Uniform Geometric Memory for Long-Horizon Personalized AI
Abstract
Long-term memory is becoming a core substrate for personalized AI, yet most systems still represent personalization as discrete records in a largely static latent space, accessed under one global similarity notion. For data mining, this creates a mismatch: the evidence is a temporal event stream, while the dominant abstraction is a searchable record set. We argue that long-horizon personalization should instead model memory as a user-specific dynamical state space with locally heterogeneous geometry. Geometry here is a computational language, not a literal claim about cognition: it captures stable versus volatile regions, variable-rate drift, heterogeneous neighborhoods, and uncertainty about current user state. Profiles and isolated events remain useful as points, but interaction, feedback, and elapsed time induce trajectories. Memory access then becomes trajectory-conditioned reconstruction of the relevant user state, not only nearest-neighbor lookup.
cs.AI / 17 / 2609.17977
When to Call an LLM: A Confidence-Gated Hybrid for Cost-Effective Emotion Recognition in Conversational AI
Abstract
Emotion recognition in conversation (ERC) is a production capability behind agent-assist prompts, escalation routing, and post-call analytics in contact-center-as-a-service (CCaaS) platforms, where cost and latency constraints matter as much as accuracy. We report a systems-level comparison of three deployment options for dialogue-contextual ERC: a low-cost stacked ensemble (sentence embeddings, windowed context, RandomForest/XGBoost/logistic-regression stacking), off-the-shelf LLM prompting (GPT-4o-mini; zero-shot, few-shot, chain-of-thought), and a confidence-gated hybrid that escalates only the ensemble's least-confident predictions to the LLM - modeled on IVA-to-human-agent escalation policies used in production contact centers. On IEMOCAP, the ensemble significantly outperforms every LLM configuration (0.595 vs. 0.460-0.536 weighted F1, p < 0.0001) at a fraction of the cost and sub-10ms latency; on MELD and CMU-MOSI the ranking reverses, showing neither pure system is a safe default. The confidence-gated hybrid resolves this by Pareto-dominating both pure systems on all three datasets (0.620, 0.643, 0.824 weighted F1) while routing the majority of traffic through the near-zero-cost ensemble, translating to roughly $10-85 per million utterances versus $99-170 for an LLM-only pipeline. The escalation policy is not an opaque cost/accuracy dial: escalated turns disproportionately follow an emotion or sentiment shift, giving operators an interpretable, auditable routing signal, and the ensemble's confidence is well-calibrated and safely under- rather than over-confident. The pattern holds across three datasets and two LLM providers. Confidence-gated cascading is established in general ML systems; our contribution is showing it transfers cleanly to dialogue-contextual ERC, yielding a concrete deployment recipe for CCaaS and conversational-AI platforms deciding how to allocate LLM spend.
cs.AI / 18 / 2609.17983
Contiguity, Not Importance: Budgeted Repair of Stale KV Caches After Document Edits
Abstract
KV-cache reuse can reduce inference cost in retrieval-augmented generation and agentic systems, but cached contexts may become stale when retrieved knowledge, working memory, or user state is edited. Under causal self-attention, even a local edit can affect downstream KV states. A full re-prefill reliably restores consistency but is costly, whereas refreshing only the edited span can leave downstream dependencies stale. We formulate in-place repair as budgeted recomputation and compare training-free position-selection policies on a factual RAG benchmark with matched direct and derived edits. Across three model families, all policies repair direct cases, but derived cases clearly separate them. At the primary budget, a contiguous edit-local window recovers at least 0.94 of the post-edit answer margin and substantially outperforms attention-based, KV-deviation, and structural selectors. Mechanistic analysis shows that position sets effective under clean-state transplantation can fail under actual recomputation because scattered positions inherit surrounding staleness. The edit-local advantage also depends on adjacency and largely disappears when the answer-bearing text moves downstream. Because answer-relevant edits almost always corrupt model behavior, failure severity is difficult to predict, and repair is 13-21 times faster than full re-prefill, our results support unconditional edit-local repair when the dependent text remains adjacent to the edit.
cs.AI / 19 / 2609.17984
TuiML: Machine Learning for AI Agents
Abstract
Machine-learning libraries such as Weka and scikit-learn were designed for human programmers. Language-model agents now use these same libraries by recalling APIs from memory and writing code, an approach that hides what a library offers, delays errors until runtime, and loses experimental state between turns. We present TuiML, a self-contained machine-learning library built for AI agents, with native algorithms across supervised, unsupervised, time-series, data handling, tuning, and evaluation tasks. Every component describes itself through machine-readable metadata and parameter schemas, so an agent can search the library, inspect components, compose validated workflows, and register new ones that become discoverable in turn. Every call is validated, seeded, and traced, and sessions export as runnable notebooks, making experiments reproducible by construction. One specification layer drives the Model Context Protocol (MCP), agent-framework adapters, a Python API, a CLI, and local model serving, while data and models never leave the machine. Benchmarks show TuiML remains predictively competitive with scikit-learn and Weka. While looking like a conventional library to a human user, TuiML is designed for agents first, allowing them to read, extend, and operate machine learning autonomously. TuiML is open source, with documentation at https://tuiml.ai.
cs.AI / 20 / 2609.17985
RideWay: Benchmarking Efficient Task Completion for Tool-Using Language Agents
Abstract
AI agents are usually evaluated by whether they complete a task. In interactive service settings, a successful agent can still frustrate users by asking repeated questions, performing redundant searches, or making avoidable revisions. We introduce RideWay, an efficiency-centered benchmark for ridehailing agents in a stateful tool-calling environment, together with Efficiency Utility, a success-gated metric that discounts successful trajectories for excess tool calls and user-facing turns relative to task-specific reference effort. Human paired preferences calibrate the relative penalties, reflecting an aggregate service-workflow trade-off: extra dialogue often creates visible friction, whereas extra tool use can sometimes verify constraints or preserve user intent. Across 58 tasks and 24 models, the fitted penalty for excess turns is about twice that for excess tool calls. On task-disjoint held-out preferences, Efficiency Utility achieves 78.7% accuracy overall: 90.6% when trajectories differ in turns, but chance-level accuracy when they differ solely in tool calls - the axis on which human annotators agree least. RideWay therefore makes interaction efficiency measurable alongside task success, while exposing the boundary of count-based tool-use evaluation.
cs.AI / 21 / 2609.18057
Anchoring What Matters: A Dual-Level Learning Framework for Visually-Grounded Multimodal Reasoning
Abstract
Reinforcement learning with verifiable rewards (RLVR) has significantly improved the reasoning capabilities of large vision-language models (LVLMs). However, standard on-policy RLVR algorithms face a critical optimization bottleneck in preserving and reinforcing visually grounded reasoning behaviors: valuable visually-grounded reasoning trajectories are discarded after a single update, while uniform token advantage allocation prevents the model from reinforcing critical perception or reasoning steps. To bridge this gap, we propose PIVOT, a dual-level learning framework that anchors policy optimization around informative visual reasoning signals. Specifically, PIVOT introduces a self-calibrated experience replay mechanism, which selectively collects and replays visually-grounded historical experiences as stable reference anchors for policy optimization. Building upon this, we further design a vision-guided advantage allocation mechanism to allocate additional vision-aware advantages to tokens based on their local visual support and impact on downstream reasoning. Extensive experiments across diverse benchmarks demonstrate that PIVOT achieves highly competitive performance in enhancing the multimodal reasoning capabilities of LVLMs.
cs.AI / 22 / 2609.18063
The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction
Abstract
Mixture-of-experts (MoE) inference on consumer hardware is bounded by weight memory: a 35B-class model is 19.5GB at 4-bit, and sparsity shrinks the compute per token, not the bytes that must be held. Naive offloading to SSD does not help on its own, because layer N+1's experts must be chosen before layer N's output exists, so the reads cannot start early enough to hide behind compute. We present Edge0, a streaming MoE inference engine that closes the gap with a prerouter: a per-layer head predicts the next layer's routing one token ahead, and the prediction is consumed as the routing itself, so the staged expert set equals the routed set and nothing is dropped. An unmerged recovery LoRA, trained on the student path, pays back the quality lost to int4 quantization and routing replacement. On a single 24GB machine, Edge0 serves a 35B MoE at 20tok/s inside 3GiB of peak active memory, within a few points of its fp16 teacher on average across five public benchmarks. An 8B tier runs on the same framework, and the framework, checkpoints, and adapters are open source.
cs.AI / 23 / 2609.18072
Teaching AI, Robotics, & Community: A Hubs-Based K-12 Education Framework for Reaching Rural Schools
Abstract
K-12 robotics and AI education remains difficult to scale, especially in rural regions lacking sustained technical mentorship. Programs like FIRST provide competition pathways and instructional opportunities, but they do not eliminate the need for local programming and robotics expertise. We introduce AI, Robotics, & Community (ARC), a hubs-based framework where colleges train undergraduate mentors and host workshops for nearby K-12 teams. Mature school programs can become secondary hubs that support additional schools, creating a self-reinforcing education loop where mentorship reach propagates geographically and can even grow super-linearly. We first evaluate ARC through a trial deployment at one university. The trial created three rural robotics teams. On five-point Likert surveys, mean increases in K-12 programming knowledge, resource access, and practice opportunities were 2.00, 2.25, and 1.25 points. Likewise, undergraduate confidence teaching technical concepts, adapting explanations, managing groups, and finding mentoring enjoyable and meaningful increased by 1.29, 1.14, 1.00, and 1.14 points. Additionally, we create a spatial Markov model of ARC's growth and simulate it using the state of Indiana as a testbed. Under moderate conditions, we find that ARC reaches 74% of Indiana's 1,925 public K-12 schools and produces 992 robotics programs after 40 years, compared with 161 projected under natural growth alone. Together, these results show ARC can create and support rural robotics programs, train undergraduate AI and robotics mentors, and potentially scale mentorship across a region.
cs.AI / 24 / 2609.18080
Decodability is Not Causality: Dissociating Probe Readouts from Behavioral Drivers via SAE Decomposition
Abstract
Linear probes can decode safety-relevant concepts such as truthfulness from language-model activations, but probe accuracy may show only decodability, not that the features the probe weights causally drive model behavior. We demonstrate that this gap cannot be closed from the geometry of probe weights alone: the features geometrically aligned with probe direction need not be the ones the model uses, so causal relevance requires intervention. We introduce a feature-level diagnostic that decomposes a deployed True/False probe into sparse-autoencoder (SAE) features, ranks those features by both probe alignment and by gradient sensitivity of the model's behavior, and ablates the resulting shared, probe-only, and random feature sets under a coherence gate. On the truth probe of Buerger et al. (2024) (TTPD), applied in the instructed truth/deception setting of Long et al. (2025) for Gemma2-9B-Instruct, the two rankings overlap only weakly (about 12%, Spearman rho = 0.10), and ablation dissociates them sharply: features the probe shares with the model flip the output far more (up to 27%) than equally sized probe-only (6%) or random (1%) features at full coherence, while probe-only features instead perturb the probe's own readout. The dissociation holds across five seeds and a held-out split, and an activation-aware selection of features flips behavior nearly three times as often as the probe's geometric top features (17.6% vs. 6.1%). In this setting, therefore, the geometric projection of a probe's weight vector alone does not identify the features the model causally uses; however, combining probe information with feature activation statistics recovers substantially more behaviorally causal features, and coherence-gated SAE intervention is needed to separate them from probe readouts.
cs.AI / 25 / 2609.18099
When Is Graph Structure Worth Its Cost? The Case for Structure Pricing in Retrieval-Augmented Generation
Abstract
Graph-based retrieval-augmented generation (RAG) can help answer questions that require information from many documents. However, building a graph often requires many language-model calls during ingestion. It is therefore important to ask whether its quality gains justify the additional cost. We present EffiRAG, a graph-based RAG system designed to reduce this cost. It uses the graph to locate relevant passages and generates answers from the original text. This design preserves source information while keeping graph construction and query processing lightweight. We evaluate EffiRAG on UltraDomain, which contains 120 open-ended questions from four domains. Compared with LightRAG-hybrid, EffiRAG produces the preferred answer on 93 questions. LightRAG is preferred on 7, and the remaining 20 are splits. EffiRAG also reduces total system cost by 57 percent, from USD 0.952 to USD 0.408. The cost includes language-model calls during ingestion and querying. The advantage remains as the corpus grows. At 10 and 20 documents per domain, EffiRAG uses a lightweight, non-LLM filter to skip low-salience chunks. It remains preferred over LightRAG-hybrid. It costs 4.2 times and 4.5 times less, respectively. The comparisons identify different quality-cost trade-offs. Graph-based RAG systems should therefore be evaluated by both answer quality and cost. The results favor graph structure that locates and preserves source evidence.
cs.AI / 26 / 2609.18126
Designing Agentic AI Workflow Portfolios under Imperfect Selection and Compute Cost
Abstract
Agentic AI systems often approach the same task through multiple workflows that differ in reasoning strategy, verification structure, and compute cost. A natural deployment policy is to use the workflow with the highest average performance, but this can be suboptimal because different workflows may succeed on different instances. We study a portfolio-and-selector paradigm in which a firm runs multiple workflow executions and selects the final answer after observing their outputs. Additional executions may uncover correct answers that the best standalone workflow misses, but they consume compute and introduce plausible distractors that complicate final selection. We formulate this as a workflow portfolio problem in which the firm jointly chooses run size and allocation across workflow types. We summarize selector quality through an odds-lift index and derive sharp bounds on the value of workflow variety. For finite workflow pools, we develop exact formulations, linear programming relaxations, randomized rounding procedures, and computable performance certificates. For large implicit workflow classes, we derive a finite-dimensional dual and an ellipsoid method using a pricing oracle to identify workflows with high weighted accuracy net of recurring compute cost. Under a weak condition, the method obtains a near-optimal solution to the relaxation with polynomially many oracle calls. We evaluate the framework on three datasets: ABCD, Schema-Guided Dialogue, and HotpotQA. Relative to the best standalone workflow, portfolio optimization improves held-out selector accuracy by 3.1, 7.5, and 0.9 percentage points, respectively. Dual-guided workflow generation adds 3.5 points on ABCD and 24.1 on HotpotQA, with no additional gain on Schema-Guided Dialogue.
cs.AI / 27 / 2609.18163
Time-Aligned Evolving Concept Graphs for Scientific Relation Forecasting
Abstract
Forecasting scientific relations can guide discovery by identifying promising connections before they emerge. Existing approaches often model concept semantics and graph structure separately or summarize semantics over coarse historical snapshots, leaving semantic representations potentially misaligned with rapidly evolving graph evidence. We propose a time-aligned evolving concept graph framework that jointly models semantic and structural evolution. Its core idea is to treat dated papers as shared update events, reconstructing semantic and structural states from the same publication history through each prediction time. Pair-level fusion combines these states to forecast first co-occurrence, relation formation, and conditional relation type. Holding architecture and training fixed, refreshing context alongside graph updates improves mean relation AUPRC by 16.6% over frozen context. On a graph built from 187,848 papers with 270,687 concepts and 7.45 million co-occurrence links, the complete framework improves mean relation AUROC from 0.9290 for the strongest evaluated baseline to 0.9722, with mean population-weighted AUPRC 0.005778.
cs.AI / 28 / 2609.18182
WFM: Wiki Foundation Model for Complex Agentic Reasoning
Abstract
Real-world agents fundamentally require persistent non-parametric knowledge for dynamic reasoning, i.e., long-term memory and retrieval-augmented generation. While graphs have shown reliable advantages in providing structured evidence, the sparse graph representations naturally restrict machine readability and semantic density required for complex agentic workflows. Driven by this limitation, the entire industry is witnessing a paradigm shift from traditional sparse graphs to LLM Wiki, an agent-native knowledge representation that couples dense document contexts with markdown files containing multi-layered topological linkages. However, parameterizing such rich semantics is challenging to encode dense textual contexts using traditional sparse graph embeddings. Moreover, learning LLM Wiki with existing graph encoders could overwhelm distributed system overheads that hinder deployment in large-scale commercial scenarios. To this end, we propose a novel paradigm Wiki Foundation Model, i.e., WFM, tailored for scalable, agent-native representation and retrieval. Specifically, (i) we formalize a Wiki Graph schema that seamlessly bridges fine-grained structures with dense contexts, maintaining explicit topologies alongside continuous semantics; (ii) A query-conditioned attentive aggregation is tailored for rich wiki message passing and explicit attention variance regularization; (iii) We engineer an infrastructural NCCL boundary exchange protocol that hoists static partition indices and leverages fixed-shape GPU-to-GPU collectives, bypassing CPU serialization and memory copy overheads. Extensive evaluations across five long-term agent memory and multi-hop reasoning benchmarks demonstrate the remarkable performance of WFM, while achieving a 10.5 times training acceleration on distributed clusters.
cs.AI / 29 / 2609.18249
Re2A: Situated Conversational Recommendation via Rubric-based Preference Reasoning and Alignment
Abstract
Real-world recommendation scenarios are commonly grounded in shared physical environments during user-recommender interactions. This motivates situated conversational recommendation (SCR), a complex task requiring recommender assistants to jointly reason over dialogue history, co-observed scenes, and in-scene item attributes. However, current approaches struggle with this setting due to two intertwined challenges: accurately understanding situated user preferences throughout the conversation and generating responses that simultaneously satisfy user needs and grounded situations. To this end, we propose Re2A, a framework that formulates SCR as a structured reason-then-align process. We introduce rubric-based preference reasoning, which uses automated rubrics to guide the model toward producing explicit preference states. Based on these states, we propose a preference-conditioned optimization to align response generation with dual objectives: user preference satisfaction and situation consistency. Extensive experiments on two SCR datasets demonstrate that Re2A consistently outperforms state-of-the-art methods, delivering more precise, context-aware conversational recommendations. Our code is available at https://github.com/DongdingLin/Re2A.
cs.AI / 30 / 2609.18262
REPAIR: Resolving Long-Tail Confusion in Scientific Retrievers via Fact-Verified Iterative Refinement
Abstract
Precise retrieval of scientific information is fundamentally constrained by long-tailed concepts and high fact-sensitivity of scientific corpora. These challenges often limit the effectiveness of dense retrievers and hallucination-prone LLM augmentation. To address this, we present REPAIR, a self-evolving data augmentation framework for scientific dense retrievers. REPAIR iteratively synthesizes training data to address knowledge gaps by cycling through diagnosis of long-tail concepts, API-guided evidence expansion, and differentiation via hard negative mining. This process effectively grounds retrieval in factual reality to resolve fine-grained distinctions. Extensive experiments demonstrate that REPAIR significantly outperforms 19 strong baselines on nine materials science and biomedical benchmarks. Our work highlights that diagnosing and factually augmenting data to long-tail deficits is essential for robust scientific retrieval.
cs.AI / 31 / 2609.18272
Who Audits Whom, on What Substrate, with What Evidence? An Independence-Graded Audit Protocol for Agentic AI
Abstract
Agentic AI systems plan, invoke tools and act with limited supervision; they are now both the subject of audits and, increasingly, the auditor. Independence, the foundation of assurance,is still applied to them as a binary. We argue that it must be graded along three orthogonal axes: principal independence (who controls the auditor), substrate independence (an auditor sharing the auditee's foundation-model family, toolchain or guardrails fails with it) and evidence independence (whether evidence is attestable rather than self-reported). Each axis has precedent; the contribution is to grade all three on a single audit, aggregate them by the weakest link, and apply the same rubric when the auditor is itself an agent. We give the model a formal basis by transplanting the beta-factor model of common-cause failure from reliability engineering, a seven-step protocol whose outputs a third party can verify, a structural detectability analysis of a procurement-controls agent audited at three grades, and a Monte Carlo study of the model in which a conventional internal audit of an agent-a real audit team, a second agent, provider logsp-surfaces 5.9% of the faults it could in principle see and none at all in half the fault classes. We map the triple to the EU AI Act as amended, ISO/IEC 42006, UK public-sector risk-management guidance and audit-regulator practice.
cs.AI / 32 / 2609.18278
Building Trust in Artificial Intelligence: A Necessity for Railway Applications
Abstract
Artificial Intelligence (AI) is currently only applied to non-safety critical applications due to the strict standards and regulations for railway industries. We propose to review the three main fields necessary to increase trust in data science and AI algorithms and reach compliance: robustness, Operational Design Domain (ODD), and explainability. Robustness is the ability of an AI system to maintain its level of performance under any circumstances (ISO24029). ODDs allow the explicit definition of operating conditions under which a system is intended to operate, according to the recently published DIN DKE SPEC 99004. Explainability is the property of an AI system to express important factors influencing the AI system results in a way that humans can understand. Those 3 domains of research are already well investigated by nonrailway actors, with algorithms and methods ready to use for railway applications. A system view is necessary to ensure all trustworthy requirements interact continuously in a safe MLOps environment thereby fostering acceptance from regulators, operators and the public. Beyond safeguarding safety-critical applications, we aim to show that fostering deep trust in AI, as now required by regulatory frameworks worldwide, will unlock its full potential and transform the pace of adoption across mission-critical domains.
cs.AI / 33 / 2609.18328
Visual Compliance via Executable Safety Rule Entailment
Abstract
Recent advances in LLMs and VLMs have enabled safety systems to reason beyond simple risk patterns toward more contextual and semantic safety concerns. However, as risk patterns continue to evolve and safety rules become more complex, existing training-based end-to-end safeguards face persistent challenges in adaptability and explainable reasoning over complex safety rules. To address these challenges, we propose GuardEn (Guarding by Safety Rule Entailment), an executable safeguard framework that decomposes safety policies into atomic propositions through Safety-Rule Compilation, modeling their composition as executable code. At test time, Scene-Grounded Execution instantiates these atomic propositions with contextual visual information derived from scene graphs, enabling rule-grounded and interpretable safety reasoning. Experiments on SafetyVisionBench demonstrate the effectiveness of programmable safeguard for complex visual safety assessment, achieving an average improvement of 9.8 F1 points over the strongest baseline.
cs.AI / 34 / 2609.18366
Bad Genius: Counterfactual-Guided Harness Evolution Beyond Task-Specific Shortcuts
Abstract
Reliable agent evaluation is complicated by automatic harness optimization, which repeatedly uses a released benchmark $B_{\mathrm{rel}}$ to guide a Proposer that edits prompts, memory, retrieval, tools, and control code around a fixed target agent. Task holdout varies semantic tasks but leaves the benchmark protocol fixed, so a "bad genius" Proposer can produce a cheating harness whose released-benchmark gain depends on a benchmark-wide shortcut. We introduce Counterfactual Harness Search and Evolution (CHASE), which casts harness evolution as constraint generation over validity-preserving benchmark counterfactuals. After each Proposer update, a Challenger searches for an executable protocol transformation with large gain destruction. A validity firewall checks that task semantics are preserved, while a confirmation set determines whether the counterfactual enters a finite archive. We formalize an exact shortcut-neutralized benchmark $B_0$ and establish statistical guarantees linking finite counterfactual archives to $B_0$ and characterizing sequential Challenger search. We evaluate CHASE on a synthetic benchmark and on OfficeQA, where CHASE retains strong released-benchmark gains while substantially reducing gain destruction under valid protocol changes.
cs.AI / 35 / 2609.18431
HPOQuest: A Rare-Disease Diagnostic Agent Using Active Phenotype Acquisition
Abstract
More than 300 million people worldwide are affected by one of over 7,000 known rare diseases, yet diagnosis remains difficult because patients initially present with incomplete and heterogeneous phenotypes. We present HPOQuest, a training-free framework for sequential phenotype acquisition in rare-disease diagnosis. Starting from a small set of observed patient phenotypes, HPOQuest maintains a probabilistic disease ranking and iteratively selects informative follow-up questions to support clinicians during patient assessment. Confirmed phenotypes update the disease ranking, while all responses update the candidate question set. Across four benchmark cohorts, HPOQuest substantially improves diagnosis from sparse initial phenotypes, with gains of up to 30% points at Recall@1 and 45% points at Recall@5. These results demonstrate that sequential phenotype acquisition can substantially improve rare-disease diagnosis from limited initial clinical evidence.
cs.AI / 36 / 2609.18435
WetRobo: A Reproducible Robot Kit for Coding Agents in Biological Laboratories
Abstract
Automating biological research requires general-purpose, reproducible robot systems that allow individual wet-lab researchers to delegate robot tasks without performing teleoperation or neural-network training. Vision-language-action policies have been proposed for general-purpose arms, but can lose performance when their operating environment changes. We therefore built WetRobo, a robot kit that can readily transfer between laboratories. It consists of one robot arm, laboratory equipment (an incubator, a reagent bottle with a cap, and a Petri dish), the existing code that moves the arm, teleoperation demonstrations of each task that we recorded, and a general AGENTS.md skill file. A biological experimentalist provides natural-language tasks without collecting local teleoperation training data or training a neural network. The coding agent observes the local laboratory and writes and executes programs, using external tools as needed for adaptation. We demonstrate use of WetRobo with OpenAI Codex (gpt-5.6-sol) on three successful tasks: lifting a Petri dish lid, removing a bottle cap, and opening the incubator door, all in real-world laboratories. The coding agent achieved the cap task in both laboratories, Lab X and Lab Y, whereas a VLA fine-tuned on Lab X demonstrations succeeded there but failed to transfer to Lab Y. These results point to a practical route for laboratory robotics: instead of training a policy for each laboratory, distribute a kit and let a coding agent adapt it in each laboratory. Code, demonstrations, and the evolved programs are available at https://github.com/tsudalab/WetRobo.
cs.AI / 37 / 2609.18442
Risk-Aware World Modeling with Flow-Guided Occupancy Evolution for Selective Trajectory Planning in Automated Driving
Abstract
Safe motion planning in automated driving requires anticipating evolving traffic risks and deciding when to revise the current planned trajectory. We introduce RiskWorld, a risk-aware world modeling framework for shared occupancy forecasting and selective trajectory replacement. Spatial risk fields and temporal actor context are fused with visual bird's-eye-view features. Flow-guided evolution transports occupancy and scene features, while signed residuals correct occupancy after transport. One forecast is generated per planning step and reused across candidates. Each candidate is compared with a current-state persistence reference, yielding a nonnegative collision-score correction. The trajectory selected by current-world evaluation serves as the planning anchor and is replaced only when additional predicted risk triggers intervention and an alternative satisfies component-wise constraints on predicted risk and trajectory error. Candidate geometries remain unchanged. We evaluate RiskWorld for open-loop planning on nuScenes using camera features, annotation-derived current and historical actor states, and dataset-provided map context. RiskWorld achieves the lowest collision rate at a long evaluation horizon of 3 s, and the second-best average L2 error among various state-of-the-art baselines, while running at 11.5 FPS on a single NVIDIA RTX 4090 with 90.81 M parameters. Within-setting ablations show that RiskWorld achieves lower collision rates than the current-state rescoring baseline, while forecast reuse enables additional candidates to be evaluated at low marginal computational cost.
cs.AI / 38 / 2609.18453
The Mirage of Calibrated Confidence: Trajectory-Independence of Verbalized Confidence in Vision-Language Models
Abstract
A calibrated Vision-Language Model (VLM) can repeatedly self-correct, say "Wait, I should recheck," arrive at the wrong answer, and still report high confidence. We find that this occurs because verbalized confidence is largely trajectory-independent in the VLMs and calibration methods we evaluate. We examine this through three complementary lenses: content variation, token masking, and the model's own hesitation markers. We show that confidence is insufficiently sensitive to what the reasoning trajectory actually contains, and that calibration training can paradoxically worsen this disconnect. Since existing metrics like ECE and AUROC cannot detect this problem, we propose the Trajectory-Grounding Score (TGS) in two complementary forms: TGS-self, which compares confidence with and without access to the model's own trajectory, and TGS-pair, which tests whether the model assigns higher confidence to correct trajectories than to flawed ones along the vision, reasoning, and answer axes. We propose TGS-Bench, a model-agnostic suite spanning 10 benchmarks with controlled good/bad trajectory pairs, and show that conventional calibration rankings diverge from trajectory-grounding rankings, exposing a blind spot in current evaluation practice.
cs.AI / 39 / 2609.18460
Collective Loss of Control in LLM Agent Systems: An Epidemic Account of Mutation, Contagion, and Recovery
Abstract
How does a multi-agent system evolve from a local deviation into collective loss of control? We propose an epidemic explanation organized around accidental mutation, contagion, and recovery. A spontaneous deviation creates a seed; communication enables other agents to adopt and retransmit its unsafe strategy; collective failure can emerge when propagation outpaces correction and containment. Thus, rare individual deviations can coexist with substantial collective risk. Motivated by reported OpenAI agent coordination incidents, we examine two ingredients of this mechanism. A deployment audit identifies implicit communication paths between nominally independent evaluation runs and verifies transport through a default Docker backend. RogueHandoff-20, a benchmark of 20 executable scenarios, tests recipient susceptibility by injecting unsafe trajectories generated by a modified Qwen-27B route. Across four native-pending routes, executed harm is 0-5% on normal tasks and 40-95% after injection, exceeding paired direct malicious requests by 5-45 percentage points. These results support low observed baseline harm alongside high conditional susceptibility; they do not establish natural rare-event rates or demonstrate an autonomous cascade. The account motivates complementary defenses: strengthen resistance and recovery alongside prevention of spontaneous deviations, and audit and restrict unintended communication paths that can turn local failures into collective loss of control.
cs.AI / 40 / 2609.18461
Disentangling Long-Term Memory via Latent Neuro-Symbolic Reasoning
Abstract
Personalized agents are required to reason over long-term history interactions to infer both explicit preferences and implicit behavioral evidence. While early flat retrieval methods score memory fragments independently and neglect the distributed information, current structured memory frameworks rely on query-agnostic static graphs that fail to capture the context-dependent relations. Crucially, raw textual memories are inherently entangled and noisy, making fine-grained personalization and cross-session reasoning computationally prohibitive. To this end, we present LGM, a novel neuro-symbolic framework that shifts long-term memory disentanglement into a continuous latent space. Specifically, (i) instead of persisting fixed graphs, we design a tailored latent graph construction with a sparse autoencoder. Subject to each query, it maps historical interactions into latent memory nodes and disentangles the memory traces into sparse concept activations, dynamically synthesizing query-aware relational edge weights. (ii) A graph encoder then treats the query embedding as a conditioning preference to direct non-linear message passing across the task-specific latent subgraph. This yields a highly expressive memory representation for effective activations. Extensive experiments on long-term personalization benchmarks demonstrate that LGM significantly outperforms state-of-the-art baselines in capturing both explicit and implicit preferences while enabling personalized responses.
cs.AI / 41 / 2609.18471
First Token Matters: Understanding Safety Collapse in Large Reasoning Models
Abstract
Large Reasoning Models (LRMs) exhibit strong problem-solving abilities, yet their safety alignment often degrades when handling harmful queries. Existing approaches to improving safety largely rely on additional training or preference optimization, while offering limited understanding of the internal mechanisms behind safety failures. In this work, we investigate this failure through a token-level positional analysis of refusal dynamics and identify a localized vulnerability at the onset of reasoning, which we term Onset Refusal Collapse (ORC). We find that the refusal-related signal of LRMs drops sharply at the first generated token under harmful queries, which is associated with unsafe response generation. Motivated by this finding, we propose SafeToken, a lightweight inference-time intervention that injects a learned continuous safety anchor precisely at reasoning onset. Despite updating only a single token embedding, SafeToken effectively mitigates ORC, improves safety on harmful-query benchmarks, and largely preserves reasoning utility. These results suggest that safety failures in LRMs can arise from a transient breakdown at the critical transition from understanding to generation.
cs.AI / 42 / 2609.18481
Hyperbolic Graph Representation Learning for Differential Diagnosis on Biomedical Knowledge Graphs
Abstract
Biomedical knowledge graphs combine ontology-derived hierarchies with transversal associations among heterogeneous entities such as phenotypes, diseases, genes, proteins, and patients. This hybrid structure raises the question of whether hyperbolic embeddings, which naturally capture tree-like organization, remain useful beyond purely hierarchical graphs. We present a preliminary study of hyperbolic graph representation learning for Mendelian-disease differential diagnosis on a patient-integrated biomedical graph. Experiments on isolated ontology subgraphs show that hyperbolic models achieve strong performance in substantially lower dimensions than Euclidean baselines. We then evaluate the models on a link-prediction task that ranks candidate diseases for each patient. Results suggest that hyperbolic embeddings can exploit biomedical hierarchical structure while supporting diagnostic reasoning over heterogeneous patient-level graphs.
cs.AI / 43 / 2609.18525
TRIPROBE: Probing Task Separability Beyond Classification for XAI
Abstract
Modern evaluation of learning pipelines often reduces to downstream accuracy, leaving open the question of why tasks succeed or fail. TriProbe addresses this gap with a multi-level probing framework for explainable diagnosis of task separability. Rather than treating models as black boxes, TriProbe traces how separability evolves across inputs, learned features, and final classifiers. It decomposes multi-task problems into binary subtasks and applies three complementary probes: a Foundational Probe on input spaces, a Latent Probe on feature representations, and a Final Probe on classifier outputs. Using Maximum Fisher's Discriminant Ratio as a principled separability metric, TriProbe identifies bottlenecks and affected task pairs. Experiments on the Roshambo sEMG benchmark show how TriProbe reveals hidden breakdowns, guiding data collection, validation, and architecture design.
cs.AI / 44 / 2609.18676
The Uneven Impact of Generative AI on Student Learning: Examining the Roles of Reliance, Evaluation Literacy, and Course Policy in AI-related Courses
Abstract
Generative artificial intelligence (GenAI) is changing how students learn, yet the roles of course context, cognitive reliance, evaluation literacy, and early reliance remain underexplored. Using survey responses from 118 students across 12 AI-related courses at our institution, we examined differences in GenAI use and perceived learning experiences. We identified four user clusters: high-use students reporting many benefits, light users reporting less reliance and fewer benefits, and two moderate-use groups reporting different levels of benefit. We also found significant differences between free- and premium-version users, single- and multiple-tool users, and students experiencing different instructor policies. In multivariable regression models, academic benefit was associated with early reliance and academic task support; positive impact was associated with cognitive reliance, academic task support, confidence in GenAI reliability, and instructor policy; and negative impact was associated with early reliance and attitudinal change. The association between early reliance and negative impact became stronger as evaluation literacy increased. Finally, perceptions of GenAI-enhanced learning appear to reflect cognitive, performance, and self-efficacy benefits, while concerns about stress and diminished critical thinking are associated with lower perceived learning benefits. These findings suggest that institutions need better policies to address such inequities so that institutions can enable students to benefit from increasingly capable AI systems.
cs.AI / 45 / 2609.18769
Version- and Scope-Aware Question Answering over Normative Documents: A Deployed System and an End-to-End Evaluation at Production Scale
Abstract
Correctly answering a question grounded in normative documents often depends on information outside any single passage: whether the retrieved document is the version currently in force; whether it applies to the jurisdiction, subject (such as an institution or applicant), and date at issue; and whether each normative claim can be traced to its supporting source text. Hosted retrieval services have substantially lowered the engineering cost of building an initial system over such corpora, making "upload the documents and ask" a common default. We evaluate this default on approximately 73,000 candidate normative documents supplied to a production deployment. The evaluation uses a stratified sample of 200 questions from our published benchmark, with a gold source document for every question; the released sampling rule reads no system outputs or scores. We compare the hosted service with a governed system that resolves version and scope through explicit rules before generation. The governed system scored 97.7 overall, while the hosted service scored 88.1, a gap of 9.6 points computed from unrounded means. The question set, the answer text evaluated for both systems, the scores, and the scripts used to reproduce the reported benchmark statistics are public. The governed configuration has operated as a commercial product since January 2026 and serves 1,126 registered users; named customer organizations include Zhipu AI and Lecheng Health. By mid-April 2026, it had reached roughly 100,000 calls per workday.
cs.AI / 46 / 2609.18779
CERA-MoA: Co-Evolving Routing Mechanisms with Continually Learning LLM Agents
Abstract
Current Mixture-of-Agents (MoA) paradigms generally treat query routing and agent fine-tuning as separate processes, limiting their ability to respond to evolving agent capabilities. This disconnect prevents routing strategies from adapting to evolving agent capabilities during post-training and prevents agents from achieving synergistic data-driven specialization. To resolve this, we introduce CERA-MoA (Co-Evolving Router with continually learning Agents for Mixture-of-Agents), an iterative reinforcement learning framework where the dynamic router and independent agent policies co-evolve. We design a predictive familiarity estimator that leverages mid-layer hidden states to evaluate semantic competence among agents, avoiding the overhead of full rollouts. Based on these familiarity scores, a cumulative-threshold adaptive routing mechanism dynamically activates a tailored minimal agent subset, achieving a trade-off between task performance and efficiency. By proactively allocating targeted training samples to agents based on their evolving competence, CERA-MoA promotes capability differentiation. Extensive experiments across various domains demonstrate that CERA-MoA outperforms state-of-the-art static-agent routing and fix-workflow fine-tuning baselines.
cs.AI / 47 / 2609.18820
Compositional Policy Violations: When Step-Level Compliance Fails In Agentic AI Workflows
Abstract
Agentic workflows now make consequential decisions in regulated settings, and the governance placed around them is almost entirely step-scoped: input-output classifiers, per turn rails, and span-level evaluators. The policies organizations actually hold, such as referral thresholds, authority limits, and review requirements, are properties of the whole execution rather than of any one step. This mismatch admits a failure mode we call a Compositional Policy Violation (CPV): every individual step passes its own check while the composed execution violates the governing policy. A predicate over a single step cannot evaluate a property that step does not determine, so no improvement in the accuracy of the step-scoped monitors detects this class. We define CPVs as the failure of step-level compliance to compose, and present a taxonomy of four types: Authority Creep, Threshold Laundering, Cumulative Sum Violation, and Context Collapse. We show that the correct repair for each class is dictated by where the guarded quantity mutates. We then introduce a provenance-aware runtime architecture that evaluates policies over complete execution traces, recomputing guarded quantities from raw provenance rather than the pipeline's derived representation.
cs.AI / 48 / 2609.18842
Infinite-Parameter LLMs: Generating and Adapting Weights from Live Data
Abstract
The scaling laws hold that a language model grows more capable with more parameters and more training data, and Mixture-of-Experts (MoE) architectures have ridden these laws to remarkable results, activating only a fraction of an enormous stored parameter bank for each token. That success is built on static pretraining data. A deployed model faces a different world, where much of the data that would make it more useful is not in its training set but in the live interaction it is currently handling, such as the facts a user supplies or the corrections they give. A conventional model cannot learn from this data, because its weights are frozen after training. Instead, the knowledge and behaviour supplied at run time are placed in the prompt, by retrieval or instruction, and re-read on every request only to be discarded once the request ends. We ask how an architecture could learn from live interaction by writing it into its weights. Taking inspiration from MoE, we propose the \textbf{Infinite-Parameter LLM}. A compact hypernetwork turns the data given at run time into a low-rank modulation of a shared base network, so the feed-forward weights are generated from live data rather than stored in a fixed bank. Where prior weight generators read the context once and freeze, we carry a Bayesian belief over the generator's latent code and update it online, so the effective weight is re-derived from that evolving belief as the session proceeds rather than fixed after one read. The stored footprint stays fixed, yet the weights the model can compile are effectively infinite. For the knowledge and behaviour supplied at run time, carrying them in the weights rather than the prompt is amortized in compute, frees the context window, persists across turns, and can generalise better than in-context use. We specify an evaluation protocol that tests exactly this against in-context learning and retrieval.
cs.AI / 49 / 2609.19088
MUSE: Benchmarking Large Vision-Language Models on Multi-Modal Understanding in Situated Education
Abstract
Large vision-language models have achieved remarkable progress in multi-modal understanding, yet their capabilities in educational settings remain insufficiently evaluated. In AI-assisted language learning, models must interpret artistic imagery, understand its semantic, affective, and cultural content, and reason about visual context to support meaningful interaction. However, existing benchmarks primarily focus on real-world images or domain-specific educational reasoning, providing limited coverage of artistic educational content. To address this gap, we introduce MUSE, a benchmark for evaluating large vision-language models on artistic image understanding in situated educational applications. MUSE decouples image annotation from question generation, enabling diverse tasks with controllable difficulty while reducing annotation effort. It comprises twelve tasks spanning visual perception, semantic and affective interpretation, culture understanding, and compositional reasoning, together with diverse artistic images deliberately curated to center Singaporean and Southeast Asian multicultural contexts alongside Western art traditions, covering multiple themes and difficulty levels. Evaluation of open-source and proprietary models reveals substantial disparities across capability dimensions, particularly in affective interpretation and compositional reasoning. Our analysis further identifies common failure modes and key challenges for developing trustworthy multi-modal models for education. We hope MUSE will serve as a standardized benchmark for advancing multi-modal understanding in situated educational applications.
cs.AI / 50 / 2609.19124
Flag Game: A Toy Model for Mechanistic Swarm Interpretability
Abstract
Emergent coordinated behaviors of AI agents are starting to present critical safety risks. A key phenomenon driving these behaviors is the rapid formation and spread of beliefs about the world, and mechanistic understanding is crucial for collective alignment. To this end, we introduce the Flag Game, a toy model for studying the mechanisms of collective belief formation. Concretely, a hidden country flag defines the ground truth, and each bounded agent directly observes only a private crop but can exchange beliefs and weigh social evidence from peers. Despite its simplicity, the Flag Game reproduces rich collective phenomenology: non-monotonic scaling of performance with population size, accuracy gains from social-awareness prompting and team diversity, and strong effects of organizational structure. In particular, we identify that collective belief collapse at small population sizes turns into collective belief polarization as the population grows. This polarization causes the performance decline at large population sizes, but creates diversity in collective beliefs. Finally, we dissect the mechanisms underlying collective belief collapse and polarization with two complementary approaches. We first introduce social circuit attribution, a technique to predict which agent, and what view, matters most to collective dynamics, and verify its predictions by causal interventions on agents, tracing how agent patching changes collective outcomes. However, the efficacy of causal interventions on agents decreases as the population grows. We therefore develop a statistical mechanical theory for larger populations and verify that it matches the empirical phase diagram. Together, these results take a first step toward mechanistic swarm interpretability, a science of how the properties of individual agents and their communication give rise to emergent collective behavior.
cs.AI / 51 / 2609.19128
Cognitive Extensions for Dual-Process Language Agents: Memory and Self-Reflection in Interactive Environments
Abstract
Language agents remain brittle in interactive environments, where success requires long-horizon state tracking, valid action execution, and recovery from failed steps. We extend SwiftSage, a dual-process agent that combines a fast action proposer with a slower planner, using two modular cognitive extensions: an Adaptive Memory Module (AMM) for salience-gated episodic storage and trigger-driven retrieval, and a Self-Reflection Module (SRM) for bounded execution-time validation and corrective intervention. Both modules are implemented as feature-flagged extensions over the same execution substrate, enabling controlled ablations on ScienceWorld. Across four configurations---baseline, baseline+AMM, baseline+SRM, and the full system---the full system achieves the best mean final score (64.62), success rate (43.17%), and successful-step efficiency (19.33 steps), while SRM is the strongest standalone contributor. The results suggest that execution-time control is the dominant bottleneck in this setting, while episodic memory becomes most useful once the runtime loop is stabilized.
cs.AI / 52 / 2609.17843
RoboVAD: A Large Cross-Domain Evaluation Benchmark for Anomaly Detection in Robotic Arm Manipulation Videos
Abstract
Video anomaly detection (VAD) is an actively studied task, having wide applications in typical scenarios such as public surveillance and road traffic safety. The task is also relevant for robotic arm interactions, where it has several downstream applications, including learning better interaction and manipulation abilities, triggering recovery procedures when anomalies occur, etc. Despite its relevance, the exploration of anomaly detection in robotic arm manipulation videos is limited by the low number of available resources. To this end, we introduce RoboVAD, a large-scale benchmark for video anomaly detection that comprises challenging cross-domain evaluation scenarios, where certain actions (tasks executed by a robotic arm) and anomaly types (mistakes that occur while performing certain tasks) remain unseen during training. RoboVAD is designed to benchmark VAD methods in realistic scenarios, where robotic arms can perform unforeseen tasks, and thereby encounter new anomaly types. We train and evaluate several state-of-the-art VAD methods, including a novel method specifically adapted for robotic arm manipulation. While the proposed method outperforms many state-of-the-art competitors, all methods remain below a micro-averaged frame-level AUC threshold of 70% in the most challenging evaluation setup, confirming the difficulty of the proposed benchmark. We publicly release our dataset and code at https://zenodo.org/records/22754659.
cs.AI / 53 / 2609.17953
EDCT-Bench: Uncovering Faithfulness Gaps in VLMs via Explanation-Driven Counterfactual Testing
Abstract
Vision-Language Models (VLMs) can produce Natural Language Explanations (NLEs) that sound plausible yet remain inconsistent with the visual evidence they cite. We present Explanation-Driven Counterfactual Testing (EDCT), an intervention-based protocol that extracts visual concepts cited in a model's explanation, applies verified minimal edits to them, and tests whether the resulting answer and explanation remain consistent with the edited image. Using this protocol, we create EDCT-Bench, a comprehensive benchmark spanning three complementary domains: knowledge-intensive visual question answering (OK-VQA), safety-critical driving (DriveLM), and 3D spatial reasoning (3DSRBench). Across the evaluated VLMs, EDCT reveals substantial faithfulness gaps, with models frequently producing responses inconsistent with verified visual changes. Finally, our fine-tuning study suggests that EDCT-generated counterfactuals provide high-impact training signals.
cs.AI / 54 / 2609.18088
Mask 2D-3D: Adaptive Dual-Masked Autoencoder Network for Image-to-Point Cloud Registration
Abstract
Detection-free methods for image-to-point cloud registration are prone to erroneous correspondences caused by domain and modality discrepancies, limited sensitivity of feature extractors, and the presence of non-overlapping regions. The Masked Autoencoder (MAE) has shown strong performance in visual representation for images and point clouds. It may be helpful to apply this approach to image-to-point cloud registration, a task that requires unified feature extraction and accurate cross-modal correspondences. Standard MAE's random masking may overlook key regions due to limited camera views, reducing registration effectiveness. To address this, we propose the Intermodal Dual-MAE Framework (ID-MAE) with a Similarity-based RL Masking Strategy (SRLM), which adaptively masks informative positions by leveraging cross-modal similarity and reinforcement learning, thus narrowing the modality gap. Our method enhances cross-modal representation learning by enforcing representation consistency during feature extraction, thereby enabling more reliable 2D-3D correspondence estimation. Experiments on RGB-D Scenes v2 and 7-Scenes benchmarks show that our method achieves state-of-the-art performance in image-to-point cloud registration.
cs.AI / 55 / 2609.18206
CapMap-MS-TTA: 3rd Place Solution for the MUMU Track of the 8th LSVOS Challenge at ECCV 2026
Abstract
The MUMU track of the 8th Large-scale Video Object Segmentation (LSVOS) Challenge requires a single unified multimodal model to jointly solve image tagging (Task A), open-vocabulary object detection (Task B), and English captioning (Task C) under strict resource constraints (<=0.5B parameters and <=8 GB peak GPU memory). We present CapMap-MS-TTA, a training-free submission built on Microsoft Florence-2-base (~231M parameters), combining caption keyword mapping with multi-scale flip test-time augmentation. Task C uses the native <DETAILED_CAPTION> pathway with length/token sanitization. Task A maps the same detailed caption into the official quality/scene/event vocabularies via an expanded keyword lexicon with whole-word matching and a lightweight expand-hints stage. Task B runs Florence-2 open detection (<OD>) with multi-scale and horizontal-flip test-time augmentation (TTA), followed by label-aware non-maximum suppression (NMS). Without fine-tuning, the system improves our reproduced Florence-2 baseline from 15.16 to a best public score of 16.4815, and ranks 3rd on the final MUMU leaderboard.
cs.AI / 56 / 2609.18399
A Non-Linear Neuron Based Detection of Isolated Pixels in Binary and Grayscale Images using Contrast Sensitive Receptive Fields
Abstract
Identifying isolated points is important in image processing applications such as medical imaging, astronomy and quality control management. Other domains, such as cybersecurity, also present challenges that can be framed as image processing problems. One example of particular interest is the identification of anomalous single nodes in spatially organised networks where groups of nodes in different regions share similar feature values. This task can involve both binary and more complex grayscale images. However, existing methods face limitations: template matching is infeasible for grayscale images, while 2nd order derivative based methods are highly sensitive to noise and require user-specified thresholds. To overcome these issues, a novel method is proposed for detecting meaningful single-pixel deviations in images. This approach modifies and extends a neuron model, originally designed for anomaly detection, to operate on spatially diameter limited receptive fields that incorporate excitatory and inhibitory regions. The result is a method that is free from user-specified thresholds and parameters, and can be applied to both binary and grayscale images, providing an effective, robust and efficient solution.
cs.AI / 57 / 2609.18462
CSWAM: Better Causal Semantic Representations for Out-of-Distribution Generalization in World Action Models
Abstract
FastWAM-style world action models enable efficient action-only inference, but generalize poorly under visual distribution shifts. Their reconstruction-oriented representations emphasize appearance-specific details, limiting generalization to unseen scenes and objects. Without observation history, the model also lacks temporal evidence for robustly identifying task-relevant state changes and motion in unfamiliar visual conditions. To address these limitations, we present the Causal Semantic World Action Model (CSWAM), which augments FastWAM with a causal semantic expert built on V-JEPA 2.1. V-JEPA provides temporally grounded representations of semantic state changes and motion with less dependence on appearance-specific details. The expert learns their future evolution from a sparse history of current and past observations and shares the history-derived context with both the video and action streams through causal attention. At inference, CSWAM conditions action denoising on the current video state and observed semantic history, retaining efficient action-only inference. We conduct simulation and real-robot experiments to evaluate generalization under distribution shifts. With embodied pretraining, CSWAM raises Randomized success on RoboTwin 2.0 Clean-to-Randomized transfer from 10.16% to 45.18%, a gain of 35.02 percentage points over FastWAM. Across two real-robot tasks and three OOD difficulty levels, CSWAM improves average success over FastWAM by 42.5 percentage points, from 27.5% to 70.0%.
cs.AI / 58 / 2609.18582
On-the-Fly Homographies Calibration for Multi-Camera Tracking
Abstract
Precise multi-camera tracking traditionally relies on rigorous 3D site calibration, yet this requirement is often operationally impossible in large-scale deployments. Privacy regulations frequently prohibit recording video for offline calibration; limited bandwidth precludes synchronizing high-resolution streams from hundreds of cameras; and covering immense physical sites with calibration targets is logistically infeasible. We present a multi-camera homography calibration system designed to overcome these barriers through "on-the-fly" geometric refinement. Starting from coarse manual homographies, we introduce a centroid-based projection optimization (PO) that continuously aligns the ground-plane geometry using live detection streams. Because PO operates asynchronously on already-transmitted, lightweight metadata, it adds zero computational latency to the real-time tracker. This allows the system to adapt automatically to camera movements or environmental changes without human intervention. This optimized geometry feeds a multi-camera bird's-eye-view (BEV) tracker that fuses detections and unifies trajectories across zones. Crucially, by operating strictly on live anonymous metadata, our solution ensures a privacy-safe, zero-overhead, and resilient tracking pipeline that maintains global consistency in dynamic environments where static, recorded-video calibration is impossible.
cs.AI / 59 / 2609.18634
GenStream: Semantic Streaming Framework for Generative Reconstruction of Human-centric Media
Abstract
Video streaming dominates global internet traffic, yet conventional pipelines remain inefficient for structured, human-centric content such as sports, performance, or interactive media. Standard codecs re-encode entire frames, foreground and background alike, treating all pixels uniformly and ignoring the semantic structure of the scene. This leads to significant bandwidth waste, particularly in scenarios where backgrounds are static and motion is constrained to a few salient actors. We introduce GenStream, a semantic streaming framework that replaces dense video frames with compact, structured metadata. Instead of transmitting pixels, GenStream encodes each scene as a combination of skeletal keypoints, camera viewpoint parameters, and a static 3D background model. These elements are transmitted to the client, where a generative model reconstructs photorealistic human figures and composites them into the 3D scene from the original viewpoint. This paradigm enables extreme compression, achieving over 99.9% bandwidth reduction compared to HEVC for the continuous data stream. We partially validate GenStream on Olympic figure skating footage and demonstrate potential for high perceptual fidelity under minimal data. While acknowledging the significant computational costs shifted to the client and challenges in generalization, GenStream opens new directions in volumetric avatar synthesis, canonical 3D actor fusion across views, and personalized viewing experiences, laying the groundwork for scalable, intelligent streaming in the post-codec era.
cs.AI / 60 / 2609.18688
Generalist-Specialist Mixture-of-Experts for Rare Pathology Detection in Multimodal Imaging
Abstract
AI models for multimodal medical imaging must balance modality-specific specialization with cross-modal shared representations, a trade-off that pure Mixture-of-Experts (MoE) architectures currently fail to satisfy. Expert-based routing improves in-domain learning but may sacrifice cross-modal signals, which appear particularly important for rare (low-prevalence) pathologies in our experiments. To resolve this, we introduce Generalist-Specialist-MoE (GS-MoE), a two-branch (MoE) architecture that couples a cross-modal generalist model with distinct modality-specific specialists (experts) via domain-constrained feature fusion. On RadImageNet (1.35M images, 165 pathologies, three modalities), GS-MoE recovers detection of six low-prevalence pathologies on which every baseline scores F1 $=$ 0, with per-class gains up to +0.60 F1. It attains this while even slightly exceeding dense and specialist-only MoE aggregate baselines (MCC 0.770), while using ${\sim}53\%$ fewer active parameters at inference than the strongest investigated dense model.
cs.AI / 61 / 2609.18823
Using OCR Heads to Verbalize Image Semantics
Abstract
How do VLMs map from pixels to semantics? To understand this general question, we focus on a narrow one: studying how VLMs perform optical character recognition (OCR). Across four models, we identify attention heads causally necessary for OCR, and discover that these are in fact general-purpose heads that output interpretable semantic features across all image tokens. For example, pointing these heads at an image token containing the word "bike" causes Qwen3-VL-8B to output "bike," but pointing them at a bird wing causes the model to output the token "feathers." We collapse these heads' attention weights into a single verbalization lens transformation that reveals interpretable semantic features in hidden states across all layers. When combined with projection to vocabulary space, we can obtain interpretable labels starting from layer 0, showing that image representations are in fact aligned with language in early layers. We find that we can also use the inverse of this transformation to edit non-word concepts, e.g., replacing a tractor with a revolver in a naturalistic image, providing causal evidence that this subspace is useful for more than just OCR. Our results are an example of how the study of specific mechanisms can shed light on broader interpretability problems.
cs.AI / 62 / 2609.18860
Decodable but Misrouted: Sparse Features Uncover a Readout Gap in Vision-Language Models for Harmful Meme Detection
Abstract
When a large vision-language model misclassifies a harmful meme, the failure may reflect missing internal evidence or an inability to route represented evidence to its output. We distinguish these cases in Gemma-3 and Qwen3.5 using sparse autoencoders, role-conditioned probes, causal interventions, and recovery experiments across six harmful content benchmarks, with additional Spanish and Hindi-English code-mixed evaluations. Sparse readouts outperform native prediction on all six primary binary tasks: Qwen averages $0.740$ versus $0.432$ native macro-F1, while residual reconstruction reaches $0.486$, whereas Gemma improves from $0.532$ to $0.714$. These differences reflect supervised accessibility rather than a pre-existing, native decision rule, and the most influential token role depends on the task. Under the evaluated score scales, Qwen silent-feature ablation is $24-63$ times more probe-sensitive, whereas routed-feature patching on literal yes/no tasks is $16-140$ times more output-sensitive. Calibration-only routing recovers $93.3$% of the mean gap, and probe-distilled LoRA improves native predictions, although shared multi-task adaptation causes negative transfer. A case study of Gemma-3-12B on Facebook Hateful Memes finds a distributed rank-32 image-prompt interaction, reaching $0.756$ versus $0.685$ native macro-F1. Robustness controls show that the signal extends beyond English, is not explained solely by accompanying OCR, and depends on paired visual evidence. Thus, routing, rather than representation alone, is a recurring bottleneck in harmful meme classification.
cs.AI / 63 / 2609.17779
AI and Human Approaches to Mathematical Problem Solving
Abstract
AI systems have begun to report solutions, disproofs, and substantive advances on long-standing mathematical problems, raising questions about whether they approach research in the same way as mathematicians. This study compares public AI research accounts with the human literature on 11 such problems. The human corpus contains 58 papers that directly addressed the same mathematical targets later reported by AI sources as resolved, disproved, or substantially advanced; 31 within-problem comparisons were constructed from these materials. Six validated text-based measures capture problem resolution, method articulation, uncertainty and boundary specification, successor-question generation, generality, and cross-disciplinary integration. AI accounts place greater emphasis on resolving the focal problem and connecting ideas across fields. Human papers devote significantly more attention to explaining methods, specifying assumptions and limitations, and identifying questions for subsequent research. No precise difference is detected in generality. The estimated directions remain unchanged when each mathematical problem is removed in turn. The findings reveal two distinct research profiles: AI accounts concentrate on closing and recombining problems, whereas mathematical papers more extensively document the procedures, limits, and research opportunities through which results become cumulative knowledge. Evaluating research AI therefore requires attention to the organization of inquiry, not only whether a target is solved.
cs.AI / 64 / 2609.19096
Prepared Or Unprepared? Evaluating Healthcare Workforce Readiness for Clinical Adoption of Artificial Intelligence in Nigeria
Abstract
Artificial intelligence (AI) is increasingly integrated into healthcare systems worldwide, yet its successful clinical adoption depends critically on workforce readiness, particularly in low- and middle-income countries (LMICs) where infrastructural and training gaps persist. This cross-sectional study evaluated awareness, attitudes, preparedness, and barriers to AI adoption among 761 healthcare professionals across multiple disciplines and practice settings in Nigeria. Data were collected between December 2025 and March 2026 using a structured, validated questionnaire. Overall awareness of AI in healthcare was high (92.6%); however, objective knowledge and self-reported preparedness remained limited, with 40.9% reporting low or very low knowledge and only 63.0% feeling adequately prepared. Willingness to adopt AI was high: 92.5% expressed interest in training, and 78.7% supported inclusion of AI education in undergraduate curricula. Key barriers included lack of training (84.7%), poor infrastructure (71.1%), high cost of AI tools (61.0%), fear of job displacement (60.6%), ethical concerns (52.9%), and data privacy concerns (52.7%). Significant differences in preparedness were observed across geopolitical zones (chi-square (5) = 24.28, p < 0.001), and awareness differed across professional groups (chi-square (6) = 68.38, p < 0.001). Attitudes toward AI differed significantly across professional groups (F = 3.32, p = 0.003), with professionals who felt prepared demonstrating more positive attitudes (mean = 3.74) compared to those who did not (mean = 3.46). These findings reveal a critical disconnect between high awareness and actual readiness, underscoring the need for targeted training, infrastructure investment, and clear implementation frameworks to bridge the gap between AI technological potential and clinical reality in resource-constrained settings.
cs.AI / 65 / 2609.18849
Ask the Tool, Don't Guess: Agent Tool Calls Hold Their Progress, and the Serving System Should Read It
Abstract
An agentic request spends substantial wall-clock time waiting for tools, and its KV cache holds GPU memory the whole time. Serving systems decide whether that cache stays, leaves, or comes back by guessing how long the tool will run, from the tool's name, its history, a duration declared before the call, or the engine's own occupancy. We show that no estimate fixed before a call starts can know its duration, and such estimates may not even rank the calls. Meanwhile, the running tool already holds the answer, but the agent stack together with the tool silences it. We propose that tool calls report their progress explicitly while they run, and we measure what that takes. A census of four public agent corpora finds a readable signal in most tool time once it is revealed, in two strengths: a fraction of the work remaining, or an accurate signal that the end is near. A harness recovers it without changing what the agent sees, at no measurable cost to the agent's benchmark score. At the points where a KV cache decision is made, the reported progress is between several times and an order of magnitude more accurate than the best published predictors, and it stays accurate when the environment changes. Plugged into a production engine through a few small hints, it cuts the p90 time to first token (TTFT) after a tool call by 20.7% (HBM only) and 20.8% (HBM + DRAM) against LRU, close to an oracle. A serving system should not guess what its tools can tell it.
cs.AI / 66 / 2609.17883
Does AI Assistance Leave a Temporal Fingerprint? Detecting Overreliance in AI-Assisted Writing and Programming
Abstract
The rapid adoption of generative AI has made final artifacts unreliable evidence of student learning, and AI detectors that examine only the finished product are inaccurate and ethically contentious. Process data offers an alternative, but prior work covers only English essay writing. We ask whether AI assistance carries a temporal signature, whether it generalizes from writing to programming, and whether it distinguishes ordinary collaboration from wholesale delegation. We analyze three public corpora: CoAuthor (1,447 keystroke-level co-writing sessions), RealHumanEval (editor telemetry from 243 programmer records), and a pre-LLM CS1 corpus (5.1 million keystrokes) as a human-only baseline, comparing minimal-AI work, collaborative AI use, and simulated wholesale delegation. Three findings emerge. First, the signature generalizes: AI contributions arrive in bursts far outside the author's own baseline in both mediums (paired d_z = 1.13 and 3.54). Second, engagement diverges by medium: 93% of AI-inserted characters survived to writers' final documents, while only 14% of accepted code suggestions survived intact. Third, classifiers using only observable temporal features separate simulated delegation from authentic work nearly perfectly (F1 $\geq$ 0.997; at most 0.5% of real work misclassified), while ordinary collaboration remains hard to distinguish from unassisted work. Temporal evidence flags wholesale delegation rather than assistance, positioning process visibility as a candidate evidentiary basis for academic integrity, pending validation in authentic coursework.
cs.AI / 67 / 2609.19125
Affora: A Design System for Agent-Friendly Interfaces
Abstract
Computer-use agents increasingly operate software designed for people, but interfaces often leave actions or task state unclear to machine readers. We present Affora, a design system that supports both readers while preserving visual freedom and familiar human workflows. Three controlled studies examine component implementations, visual variation, and interaction-design principles. Their findings inform guidance from individual components to complete sites, supported by reusable implementations and executable checks. Agent performance depends on the interaction meaning available through its interface representation; substantial visual variation remains possible when that meaning is preserved. Evaluation on independently authored interfaces shows gains where Affora addresses existing deficits, but limited effects where those deficits are absent or outside its coverage. A workflow case provides preliminary evidence of reduced interaction cost. Affora connects user experience and agent experience through a shared interface rather than a separate agent-only surface.
cs.AI / 68 / 2609.17709
One Size Does Not Fit All! Dynamic Retriever and Generator Selection for RAG
Abstract
Retrieval-Augmented Generation (RAG) systems typically employ fixed retriever and generator configurations across queries, despite substantial differences in query complexity and information needs, leading to inefficient allocation of computational resources. While retrieval and generation adaptivity have been studied independently, their joint effect on end-to-end RAG performance remains underexplored. We systematically analyze how retriever and generator complexity interacts across factoid and multi-hop question answering (QA), including bridge and composition reasoning tasks. Our analysis shows that stronger retrieval generally yields larger gains than increased generation effort, but both exhibit diminishing and non-monotonic returns, indicating that higher-complexity configurations are not uniformly better across queries. Motivated by these findings, we introduce DRAG, a query-adaptive framework for selecting retriever-generator configurations. We first propose DRAG$_\text{QPP}$, a training-free routing approach that uses Query Performance Prediction (QPP) signals to guide retriever selection and perplexity-based measures over retrieved context to guide generator selection. We further introduce DRAG$_\text{SFT}$, a supervised routing approach that fine-tunes an LLM to jointly predict retriever-generator configurations. Across three LLM families and four QA benchmarks, \qpprag~achieves performance comparable to strong static RAG baselines while substantially reducing inference latency, whereas DRAG$_\text{SFT}$ consistently improves effectiveness over static and training-free adaptive baselines. Overall, DRAG demonstrates that jointly adapting retrieval and generation achieves a more favorable effectiveness-efficiency trade-off than static RAG pipelines.
cs.AI / 69 / 2609.18248
Quanta: A Self-Contained Python Library for Hybrid Retrieval over Quantised Embeddings, Lexical Indexes, and Knowledge Graphs
Abstract
An advanced retrieval-augmented generation pipeline is typically assembled from three or four independently operated systems: an approximate nearest-neighbour index, a full-text search engine, a graph database, and a relational document store. Each contributes its own deployment surface, configuration model, and failure modes, and the integration logic that binds them is written anew in every project. In this work, we present \textsc{Quanta}, an open-source Python library, which unifies dense vector search over 4-bit quantised embeddings, BM25 full-text retrieval, and knowledge-graph traversal behind a single retrieval API. Quanta makes two design commitments, which distinguish it from existing hybrid retrieval stacks. First, signals are combined by \emph{weighted reciprocal rank fusion} rather than by normalising heterogeneous scores onto a shared range, which we argue is ill-posed because such normalisations are query-dependent. Second, the graph is a \emph{candidate expander and not a relevance scorer}: traversal widens the candidate pool, and the newly admitted documents are re-scored by the dense indexes under an identifier allowlist, so structural adjacency determines what is considered while content evidence determines how it ranks.
cs.AI / 70 / 2609.18857
Taming the Agentic RAN: Stability-Guaranteed Arbitration of Autonomous AI Agents in O-RAN
Abstract
The O-RAN control plane is becoming agentic: autonomous AI agents, deployed as rApps by different vendors, independently close control loops over shared radio resources. We demonstrate on a live O-RAN system that this independence is unsafe. Two agents with individually correct objectives, one protecting a latency SLA and one maximizing utilization for energy efficiency, jointly drive recurring opposing excursions of the shared resource partition that neither produces alone. Existing conflict-mitigation mechanisms presume a statically known application population and cannot govern agents whose behavior emerges at run time. We present AURA, a lightweight arbitration layer that admits agent actions only when they satisfy feasibility invariants, per-variable dwell times, and a deadband, and we prove the arbitrated system converges to a feasible operating point. Implemented on an OpenAirInterface (OAI) testbed with measured one-way latency and throughput, AURA reduces recurring shared-state excursions by more than an order of magnitude (from 8.4 to 0.4 PRB amplitude) and virtually eliminates cross-slice throughput starvation (from 40-55% to 0.3%), while leaving the protected slice's own latency compliance unchanged, a trade-off the convergence guarantee makes explicit.
cs.AI / 71 / 2609.17758
CALOS: Control-Affine Lyapunov On-manifold Safety Layer for Safe Deep Reinforcement Learning for Quadrotors
Abstract
Deep Reinforcement Learning has demonstrated remarkable capability in quadrotor control, yet learned policies offer no guarantee of respecting safety constraints during training or deployment. We present CALOS (Control-Affine Lyapunov On-manifold Safety), a runtime safety layer that enforces attitude constraints on a quadrotor without modifying the underlying learning algorithm. CALOS formulates four tilt-angle inequalities and a Lyapunov descent condition as a single quadratic program whose solution is the minimum-norm correction to the nominal torque output of the policy. The quadratic program is solved exactly via active-set enumeration over the three-dimensional torque space, with a computational cost low enough to enforce constraints in real time across thousands of parallel simulation environments, as required by modern massively parallel Deep Reinforcement Learning training. Evaluated on trajectory-tracking tasks in NVIDIA Isaac Lab, CALOS reduces lateral tracking error by 55-60% relative to an unconstrained Proximal Policy Optimization baseline while achieving zero attitude-constraint violations on the training trajectory. By restricting exploration to safe regions of the state space, the safety layer also accelerates training convergence and improves data efficiency without producing suboptimal policies.
cs.AI / 72 / 2609.17771
HINT-Plan: Human Intention-Aware Robot Task Planning in Context-Rich Environments using Vision Language Models
Abstract
Approaches to incorporating human awareness into mobile robot decision-making mainly focus on collision avoidance in low-level motion planning, often overlooking the challenges posed by human presence and high-level behavior. To address this vacancy, we present HINT-Plan, a novel approach to integrate human intention prediction into robot task planning. HINT-Plan employs Vision Language Models (VLMs) to anticipate high-level human intentions from third-person image observations, convert them into goal states, and solve joint task-planning problems. To effectively enable scene awareness in context-rich environments, we use hierarchical Scene Graphs (SGs) as high-level representations of the environment, and translate environmental topology and actionable knowledge into formal planning language to ensure executable plans. Evaluated in a photorealistic simulation, HINT-Plan achieves an overall success rate of 69.71% in joint human-robot task planning, substantially outperforming the baselines by up to 35.29%, while also reducing functional conflicts. The results show the effectiveness of explicitly incorporating inferred human intentions into formal multi-agent task planning for proactive human-aware robot decision-making.
cs.AI / 73 / 2609.17824
Learning Multi-Humanoid Pickup and Transport via Decentralized Object-Centric Control
Abstract
We study cooperative multi-humanoid pickup and transport of objects with varying size, weight, and geometry, requiring robot teams of different sizes. Our approach uses decentralized object-centric control, where each humanoid is assigned a local attachment region on the shared object and learns to realize pickup and transport through gripperless bimanual pinching. This attachment-based interface provides a common control abstraction spanning single-robot pickup, cooperative multi-robot transport, and robot-to-robot handover, without per-task redesign. We find that policies trained only on single-robot pickup already transfer nontrivially to cooperative settings, suggesting that this abstraction captures much of the structure needed for coordination. At the same time, explicit multi-robot training further improves performance, showing that shared-object coupling introduces coordination dynamics that are beneficial to learn directly. We validate the approach in simulation across varying team sizes and object geometries, and demonstrate sim-to-real transfer on hardware, where the learned controllers enable real humanoids to perform cooperative manipulation tasks.
cs.AI / 74 / 2609.18259
${M}^2$Tok: Multi-head Multi-codebook Discrete Action Tokenization for Vision-Language-Action Models
Abstract
Recent advancements have successfully adapted autoregressive language models to process multimodal signals, such as images and actions. Since raw action signals are continuous, effective tokenization is essential to map high-dimensional inputs into compact discrete tokens for autoregressive processing. However, existing discrete action tokenizers often suffer from high reconstruction loss, failing to preserve the fine-grained dynamics required for precise control. This ``discretization bottleneck'' significantly limits the performance ceiling of downstream Vision-Language-Action (VLA) models. To address this, we propose $\mathcal{M}^2$Tok, a Multi-head Multi-codebook Action Tokenizer designed to minimize reconstruction error and enhance policy performance. Our approach introduces two key structural innovations: (1) we decompose the latent action features into multiple heads, enabling the model to implicitly align specific heads with distinct action dimensions; (2) we assign independent codebooks to each head for quantization. By leveraging the combinatorial nature of multiple codebooks, we significantly expand the representational expressivity of the tokenizer, leading to substantially lower reconstruction loss compared to previous methods. We evaluate the $\mathcal{M}^2$Tok-based VLA on the RoboTwin, Simpler-Env, and 3 zero-shot real-world tasks. Experimental results demonstrate our method not only achieves superior reconstruction fidelity but also significantly boosts the success rate of VLA models. Comprehensive ablation studies further confirm the effectiveness of the multi-head and multi-codebook mechanisms. Code is available at \href{https://github.com/cpaaax/M2Tok}{https://github.com/cpaaax/M2Tok}.
cs.AI / 75 / 2609.18487
ActionPiece: Rethinking Action Tokenization for Autoregressive Vision-Language-Action Models
Abstract
Action tokenizers play a central role in autoregressive vision-language-action (VLA) models, determining both the targets for policy training and the executable commands recovered from predicted tokens. Their fidelity is commonly evaluated using pointwise reconstruction metrics such as mean squared error (MSE), yet small individual errors do not fully characterize how faithfully action adjustments across demonstrations are preserved. After compression, similar actions may still cluster around a representative motion, while the adjustments needed for different contexts are diminished, distorted, or even reversed. We introduce physical rank consistency (PRC) to measure how well tokenization preserves local physical distance rankings after reconstruction. Evaluating decoded actions provides a common reference across token vocabularies and decoder architectures, complementing pointwise accuracy with a measure of relational fidelity. We further present ActionPiece, which preserves physical action relationships through joint supervision of representation learning and quantization. Physical rank preservation supervises near-far ordering in encoder and quantized feature distances, while quantization regularization applies the same ordering to codeword assignment distributions. Both objectives augment reconstruction, producing discrete action tokens for standard autoregressive policy learning and execution through a frozen decoder. Under the same Qwen3-VL-4B policy training setup, ActionPiece achieves 94.8% on LIBERO and 68.8% on unseen LIBERO-Plus, with additional evaluations reaching 71.9% on SimplerEnv and 51.5% across VLA-Arena L0-L2. Component ablations show that the two objectives jointly improve PRC and policy success, demonstrating the value of physical relationship supervision for action tokenization.
cs.AI / 76 / 2609.19104
rMuscle: Robotic Muscle Memory for Efficient Vision-Language-Action Model Inference
Abstract
Factory work is a promising early scenario for embodied AI: assigning repetitive manual jobs to robots has clear economic payoff, and a structured station keeps the jobs tractable for current policies. Vision-Language-Action (VLA) models now dominate as the policy paradigm for these robots. The inference latency of VLA models directly affects robot responsiveness and motion smoothness. However, existing VLA inference frameworks do not fully exploit the characteristics of embodied workloads or account for the distinct bottlenecks across different stages of VLA inference. In this paper, we first characterize embodied workloads and identify substantial task similarity across repeated robot executions. We further find that such similarity extends beyond observations and action trajectories to internal model states. Drawing on these observations, we present rMuscle, a real-time VLA inference framework inspired by human muscle memory. It exploits cross-execution similarity through a dual-phase muscle-memory cache. The Context Cache reuses visual-token outputs to reduce computation, while the Action Cache reuses neuron activation patterns to reduce weight accesses. We keep both the cache memory footprint and access overhead low through online cache recomputation, sliding-window cache retrieval, and mask sharing across consecutive denoising steps. rMuscle achieves 1.29-1.42X speedup on RTX 4090 and Jetson Thor across LIBERO, RoboTwin, and physical manipulation tasks, while maintaining the original success rates on real-world robots.
cs.AI / 77 / 2609.19137
Dreaming the Sound of Contact: Leveraging Video and Audio Generation for Zero-Shot Force-Aware Manipulation and Data Generation
Abstract
Recent advances in video generation allow robots to learn manipulation trajectories from generated videos. However, these approaches produce purely kinematic trajectories that lack force information, causing failures in contact-rich tasks where appropriate contact forces are essential for success. In this work, we explore augmenting generated video with audio to shape a bounded, time-varying desired-force profile using the loudness of generated contact sounds. We present a pipeline that jointly leverages generated video and audio to derive motion trajectories and corresponding desired-force profiles from a structured natural-language task prompt. We execute these force-aware trajectories on a Franka Panda robot using a closed-loop force regulator that tracks the audio-shaped force profile during contact. We evaluate our pipeline on multiple tasks that require making contact and demonstrate successful manipulation where a kinematic-only baseline fails. We also use the pipeline as a data generation engine to train policies that achieve the tasks in a closed-loop manner. Project website, videos, and dataset: https://dreamingcontactsound.github.io/
cs.AI / 78 / 2609.18333
Look Less, Hear Better: Jointly Rewarded GRPO for Streaming ASR
Abstract
Streaming automatic speech recognition (ASR) must be judged jointly on what it transcribes and on how quickly it commits each word. Delayed streams modeling (DSM) has become the dominant paradigm for streaming large audio-language models, exposing a structural delay $τ$ that bounds the decoder's lookahead. We show that $τ$ is a poor proxy for user-perceived latency, and that the alignment-based supervision of DSM leaves latency on the table: the same forced-aligned transcript is used at every $τ$, forcing the model to withhold words it could already commit. We introduce AWED, a word-level emission-delay metric defined relative to the acoustic end of each word, and post-train a DSM recognizer with GRPO under a reward that scores transcription accuracy and measured delay jointly. Trained at a single operating point ($τ=6$ frames), our model dominates both its supervised fine-tuning initialization and the Voxtral Realtime backbone across all evaluated lookahead budgets: it cuts WER by 30.8\% relative at an 80\,ms structural delay, and by 5.7\% relative at 480\,ms while lowering median AWED from 1.17\,s to 1.04\,s. Latency-rewarded post-training thus advances the accuracy--latency Pareto frontier of streaming ASR without architectural change.
cs.AI / 79 / 2609.18521
VoiceTrace: A Benchmark and Retrieval Framework for Who-Said-What Speech Retrieval
Abstract
Speech retrieval has become increasingly important as spoken content continues to grow across meetings, lectures, podcasts, and videos. Existing benchmarks and models have advanced semantic search over spoken content, but largely focus on \emph{what} is said while overlooking \emph{who} says it. In many real-world scenarios, however, users need to retrieve speech based jointly on semantic content and a target speaker, where the speaker may be specified naturally through a reference speech utterance rather than a predefined identity. To address this gap, we introduce \textbf{VoiceTrace-Bench}, a benchmark for hybrid speech retrieval in which each query combines text specifying \emph{what} to retrieve with reference speech specifying \emph{who} to retrieve. This setting requires models to integrate complementary semantic and speaker information directly from heterogeneous query inputs. Motivated by the joint audio-text modeling capabilities of audio-language models (ALMs), we develop \textbf{VoiceTrace}, a two-stage retrieval framework consisting of \textbf{VoiceTrace-Emb}, an embedding model that learns unified representations for efficient large-scale retrieval, and \textbf{VoiceTrace-Reranker}, a reranking model that jointly examines each query--candidate pair for fine-grained relevance estimation. Experiments show that VoiceTrace achieves state-of-the-art performance on established semantic speech retrieval benchmarks, while substantially outperforming cascade-based approaches on VoiceTrace-Bench, demonstrating its effectiveness for both conventional semantic retrieval and the new hybrid retrieval setting.
cs.AI / 80 / 2609.18673
Beyond EER: Multi-Dimensional Evaluation of Information Leakage in Speaker De-Identification
Abstract
Speaker de-identification (SDID) aims to preserve privacy by concealing speaker identity while maintaining speech utility. However, current evaluations often reduce privacy to a single dimension - biometric verification performance - typically measured by Equal Error Rate (EER). This narrow focus ignores critical leakage channels, such as soft biometric inference, embedding-level re-identification, and structural template similarity, which threaten the unlinkability and irreversibility of biometric references. We propose a holistic evaluation framework across five complementary metrics: (i) EER, (ii) soft biometric leakage score , (iii) cumulative match characteristic re-identification analysis, (iv) canonical correlation analysis and Procrustes embedding alignment, and (v) intelligibility via word error rate and semantic similarity. Evaluating five SDID systems from the IARPA ARTS program, we demonstrate that these metrics capture independent dimensions of information leakage. Our results indicate that reliance on a single metric can misrepresent the privacy properties of an SDID system.
cs.AI / 81 / 2609.18441
Multitask Reinforcement Learning for Assisting Choice Model Specification
Abstract
Discrete choice model specification is a time-consuming task in which modellers often specify and estimate multiple models while balancing goodness-of-fit, parsimony, and behavioural plausibility. We present Delphos, a multitask reinforcement learning framework that learns transferable specification strategies across transport choice datasets. Delphos frames model specification as a sequential decision-making problem in which it applies a sequence of modelling actions and receives feedback from an estimation environment based on model performance and convergence. To transfer modelling decisions across datasets with different sets of variables, Delphos represents utility specifications as sets of modelling terms using a DeepSet-Q architecture, allowing a shared specification policy to learn across multiple datasets. Trained on nine transport choice datasets, Delphos consistently outperforms independently trained single-task agents, indicating that sharing modelling experience improves learning efficiency and helps identify promising sequences of modelling decisions with fewer unsuccessful estimation attempts. When applied without further training to the unseen Swissmetro and Decisions datasets, the same agent identifies competitive specifications in less than 20 minutes on a standard CPU. It achieves a higher log-likelihood per observation than the VNS metaheuristic on Swissmetro and performance comparable to a published MNL specification developed by expert modellers on Decisions. These findings show that accumulating and reusing modelling experience enables Delphos to function as an intelligent assistant for discrete choice model specification. It reduces manual trial-and-error while allowing modellers to retain control over model diagnosis, refinement, and final selection.
cs.AI / 82 / 2609.18856
GrainSpeech: Less Context, More Detail for Compact Speech Synthesis
Abstract
Compact acoustic models face a challenging quality-capacity trade-off. We investigate two factors in this regime: encoder context and Mel-spectrogram supervision. A receptive-field-scaling study shows that expanding self-attention beyond 15 phonemes provides no consistent gains in pitch, energy, or duration prediction. Guided by this finding, we introduce a fixed-receptive-field convolutional encoder that reduces the respective prediction errors by 36.0%, 17.3%, and 3.4%. We further show that directly transferring image-domain gradient-variance supervision restores fine-scale variation but degrades predicted quality, motivating a Mel-specific formulation with axis-specific gradients, overlapping local statistics, and log-domain variance matching. GrainSpeech contains only 264.8K parameters and achieves 17.9x real-time Mel generation on a microcontroller (MCU), while attaining UTMOS scores comparable to substantially larger models with less than 1.5% of their parameters. Source code and demos are available at https://github.com/lab-emi/GrainSpeech.
cs.AI / 83 / 2609.18368
Semantic CSI Feedback for Beam Selection: When Task-Aware Embeddings from Sparse Pilots Outperform Full-Bandwidth Reconstruction
Abstract
Classical CSI feedback in FDD massive MIMO transmits a compressed reconstruction of the channel, optimizing fidelity to the original signal regardless of the downstream task. We propose a semantic communication perspective: instead of reconstructing the channel, the UE transmits a learned \emph{semantic embedding} optimized end-to-end for beam selection at the gNB. Comparing reconstruction-oriented feedback (CsiNet) against task-aware semantic feedback across two input domains and three observation scenarios, we show that a semantic embedding of just $d=8$ real values from only 43 NR CSI-RS pilots in the angular-delay domain achieves the highest beam prediction accuracy, outperforming every method with access to the full 512-subcarrier channel. The key insight is that beam-relevant information is intrinsically low-dimensional: the semantic encoder learns to discard reconstruction-irrelevant structure and retain only a compact representation that is relevant to beam selection, realizing the core principle of semantic communication: transmit the intent, not the signal.
cs.AI / 84 / 2609.18224
Remembering Solomon Marcus
Abstract
From the manifest of Andre Breton, through the transdisciplinary understanding, we arrive at a post-modern manifest. A talk by Laura De Marco (Harvard) will provide scientific background to approach an AMS poetry. The next section will be a qualitative analysis of some new operations on the real numbers. The conclusions will be given in the last section, and an appendix will recall some previous work with some new comments.
cs.AI / 85 / 2609.17838
Learning Nuclear Structure with AI: Radii and Collectivity
Abstract
Low-energy nuclear structure is encoded in a broad body of experimental information across the chart of nuclides. Learning how this information is organized across observables and nuclei can provide a data-driven empirical baseline for theoretical extrapolations and experimental design. Here, we develop held-out ensembles based on NuCLR (Nuclear Co-Learned Representations), a multi-task model of nuclear data, to study charge radii and electric-quadrupole transition strengths. Out-of-fold (OOF) validation shows that shared representation improves performance over single-task learning, yielding a charge-radius $\mathrm{RMS}$ deviation of $0.0147~{\rm fm}$ and a $\mathrm{B(E2)}$ $\mathrm{RMS}$ deviation of $0.192~e^2{\rm b}^2$ across hundreds of nuclides, competitive with state-of-the-art nuclear models. Our error bars estimate the expected prediction accuracy across the nuclear chart, highlighting regions where new data would encode information beyond the learned patterns. NuCLR thus serves as a data-driven surveyor of nuclear structure and a step toward a shared, multi-observable foundation model of the nuclear chart.
cs.AI / 86 / 2609.17788
SAiFE-gym: Model-based Environments for Automated Market Making with Concentrated Liquidity
Abstract
We present SAiFE_gym, a Python module that provides a collection of simulation environments for studying trading problems in Constant Product Markets (CPMs) with Concentrated Liquidity (CL). These markets give Liquidity Providers (LPs) granular control over how their capital is allocated and enable them to adjust their range of liquidity provision dynamically based on market conditions, which in turn, dictates how they earn fees. We decompose the microstructure of CPMs with CL in interactive components that allow researchers and practitioners to capture various economic settings. We employ a vectorized approach to optimize our environments, making them scalable for high dimensional Reinforcement Learning (RL) workflows that best describe sequential decision problems. We demonstrate the benefits of our environments by evaluating the performance of RL agents in CPMs with CL under uncertainty in market parameters.
cs.AI / 87 / 2609.17789
QiT: Quantum-Inspired Transformer for Visual Recognition Task
Abstract
Quantum machine learning offers a compelling representational perspective: angle-encoded states inhabit Hilbert spaces in which periodic similarities and interactions can be expressed naturally. Realizing this perspective for visual recognition remains difficult, however, because present quantum neural networks are constrained by limited qubit counts, costly circuit simulation and measurement, noise, and unstable optimization on noisy intermediate-scale quantum devices. We investigate whether useful structural ideas from quantum models can instead be realized as scalable classical Transformer operations. We introduce QiT, a Quantum-inspired Transformer for vision tasks with three components: (i) angle-inspired encoding that maps image tokens to learned trigonometric Hilbert-space features analogous to quantum rotation-based state encoding; (ii) self-attention over these periodic features, inducing a classical cosine kernel approximated to quantum fidelity kernels; and (iii) gated multiplicative emulation, a trainable classical surrogate for interaction terms found in variational circuits. All components are differentiable tensor operations, so QiT claims neither quantum computation nor quantum speedup and retains the $\mathcal{O}(N^2D)$ attention complexity of a standard Vision Transformer. Across image-classification benchmarks, QiT is competitive with a matched classical Transformer while avoiding the severe runtime cost observed for a small simulated quantum Transformer. QiT-B reaches 78.3\% ImageNet-1K top-1 accuracy with 45.7M parameters and 11.5 GFLOPs. These results position QiT as a scalable baseline for isolating and evaluating quantum-motivated inductive biases in visual recognition.
cs.AI / 88 / 2609.19090
Securing quantum error correction against misleading advice from AI agents
Abstract
Can an attacker turn influence over an artificial intelligence (AI) adviser into a harmful quantum error-correction update? We identify an ambiguity in passive syndrome records that obstructs recovery selection, then show how additional calibration measurements support certified recovery updates under uncertainty and drift. In an odd-distance square toric code with error-free preparation, syndrome measurements, and recovery operations, opposite coherent $X$ rotations produce identical passive syndrome-history distributions. Yet a fixed phase correction can help at one sign and harm at the other. A terminal logical measurement on known encoded calibration states supplies the missing sign information. A separate evaluator accepts an update only when calibration uncertainty and a justified drift bound certify improvement over the current recovery, without assuming that the adviser recommends correctly. In simulated advice attacks, calibration-confidence checks reject harmful proposals while retaining beneficial updates under honest advice. We derive sufficient limits on calibration age that require improvement through deployment. In matched simulations, a validated channel-specific bound retains more beneficial updates than the general bound after accounting for evaluation time, while preventing the tested harmful activations under the stated drift assumption. A separate surface-code experiment includes stochastic circuit faults and noise changing during acquisition. Deterministic controllers achieve at least as many beneficial updates with the same observations. Violating the drift assumption permits harmful acceptance in the toric experiment. The results identify information required for recovery selection, establish conditional guarantees against harmful updates, and quantify the recovery improvements forgone through conservative acceptance.
cs.AI / 89 / 2609.17772
When AI Generates Covariates: Causal Typing and Estimand Drift in Sequential Experiments
Abstract
AI-generated covariates from notes, conversations, images, and wearable streams can change the causal question when their roles are left unspecified. A generated feature may represent a treatment version, pre-action state, history, design variable, mediator, outcome proxy, observation process, or intercurrent event; these roles are not interchangeable. We formulate a causal type discipline for sequential experiments: a versioned representation map, a causal role classifier, a claim-status filter, and an estimand lock. The lock fixes a standardized proximal effect before generated covariates enter the analysis. Under audit correctness and standard identification assumptions, admissible role assignments preserve this estimand. We apply the established conditional-covariance characterization of compression bias to substitution of generated representations for design-relevant states. A standardized decomposition separates compression, conditional-law, and standardization drift. Further results cover mediator adjustment, post-action leakage, marker-intervention conflation, outcome-guided discovery, and state-measurement error. Cluster-level orthogonal estimators distinguish empirical and superpopulation targets under repeated sessions and missing outcomes. Simulations show that refinement helps when it retains design-relevant information, whereas design erasure, leakage, and same-data marker selection can produce bias or undercoverage. The framework places causal semantics and claim status before confirmatory inference with generated representations.
机器学习 (cs.LG)
114
cs.LG / 1 / 2609.17882
Can VLMs Reliably Assess Sidewalk Accessibility Attributes from Pedestrian-Level Imagery?
Abstract
An important component of urban accessibility, particularly for wheelchair users and people with reduced mobility, is sidewalk compliance with measurable requirements. We test whether effective width, longitudinal slope, cross slope, and pavement condition can be assessed reliably from pedestrian-level imagery using vision-language models (VLMs). We present the first application of sampling-based conformal prediction (CP) for VLM-based accessibility assessment. We evaluate four VLMs on 514 sidewalk images from Seoul, South Korea, with field-measured ground truth. Conformal calibration attains the nominal 90% coverage for all models and attributes, but the calibrated regions differ in informativeness. Effective width yields the most informative estimates, with a mean interval half-width of about 1.0 m for the best model. Since every model overestimates width, asymmetric calibration shortens the intervals by up to 33% at unchanged coverage. Longitudinal slope is marginally informative, cross-slope intervals are too wide to resolve regulatory thresholds, and pavement-condition sets degenerate to all five grades (A-E) for three of the four models. Uncalibrated intervals from raw sampling dispersion cover only 17-47% of field-measured values at a nominal 90% level. Among the images with the most self-consistent responses, these intervals miss the field-measured value in up to 96% of cases. Response self-consistency is therefore not evidence of accuracy, and sampling dispersion cannot be interpreted as uncertainty until it has been calibrated against field-measured ground truth. No quantitative attribute reaches the precision required for general compliance assessment, but CP identifies from calibration data alone which attributes can support screening of segments far from the thresholds. We release the annotated pedestrian-level images and their corresponding field-measured attribute values.
cs.LG / 2 / 2609.17909
Zing-0.5: Toward Playable Worlds with Real-Time Joint Action and Text Control
Abstract
We introduce Zing-0.5, a 5B autoregressive world model designed for playability: users can explore generated worlds, influence unfolding events, and respond to the resulting feedback through joint keyboard and online text control. Our approach brings together three technical contributions: (1) Unified action and text conditioning, combining magnitude-aware keyboard inputs with temporally aligned text instructions and jointly annotated videos to learn navigation and event control within the same sequence; (2) Event-scale supervision for incremental generation, using a segment-level teacher trained on connected multi-prompt videos to supervise a block-level causal student through distribution-matching distillation; and (3) Low-cost real-time interaction, combining four-step generation with context-preserving streaming to support 832 x 480 inference at 24 FPS at an estimated server rental cost of approximately USD 0.009 per stream-minute. Zing-0.5 achieves an overall score of 81.0 and a consistency score of 88.5 across 158 WBench Navigation cases. A joint-control demonstration shows a text-directed event change during continued navigation without restarting generation. We release the model weights, inference code, and Zing-SGLang serving implementation to support further work on playable generated worlds.
cs.LG / 3 / 2609.18129
MCLC-NET: Multimodal Continual Learning for Leaf Counting
Abstract
Leaf counting is an important task in plant phenotyping for monitoring plant growth and estimating crop yield. Most existing methods rely on RGB images, but their performance is often affected by occlusion, lighting variations, and other real-world challenges. Additional modalities, such as depth and thermal images, can provide useful complementary information. However, multimodal leaf counting remains underexplored. Also, many existing methods assume that all training data are available simultaneously, which is impractical in real agricultural settings, where data is collected over time from multiple sources. To address these challenges, we propose MCLC-NET, a multimodal continual learning framework for leaf counting. It learns tasks sequentially using a memory-based strategy with a memory buffer to retain important samples from previous tasks. We also introduce MMLC, a real-world multimodal leaf-counting dataset designed for a domain incremental scenario (DIS) in CL. It contains RGB, depth, and thermal images collected across different crop types under varying environmental conditions, arranged in three orderings: crop-wise, time-wise, and mixed. Experimental results, averaged over three random seeds, demonstrate that MCLC-NET consistently outperforms existing methods across all three task orderings, achieving the lowest AMSE of 0.675$\pm$0.027, 0.542$\pm$0.069, and 0.745$\pm$0.057, respectively.
cs.LG / 4 / 2609.18511
Learning from Distributed Eyes: Leveraging Collaborative Perception for Automated Model Adaptation
Abstract
In autonomous driving, perception models often struggle to generalize to new environments due to domain shifts. While unsupervised model adaptation offers a feasible solution without labor-intensive manual labeling, existing methods that rely solely on the ego-vehicle's data often lead to inferior pseudo-labeling performance. To address this critical issue, we propose LDE, Learning from Distributed ``Eyes", a novel framework that transforms collaborative perception (CP) into a source of high-quality supervision for model adaptation. This pseudo-labeling approach is hyperparameter-insensitive and relatively reliable, assuming CP often outperforms single-agent's perception. However, naively implementing this approach encounters (1) the communication bottleneck of sharing rich features under time and bandwidth constraints, (2) the view discrepancy between the CP view and the learner's Field of View (FoV), and (3) the unreliability even in CP-generated labels. To address these issues, we design an adaptation-oriented feature sharing mechanism that selectively transmits the most critical information for adaptation, an FoV filtering method that meticulously eliminates mismatched labels, and a curriculum learning strategy to progressively exploit pseudo labels. Extensive experiments on 3D object detection tasks demonstrate that LDE consistently outperforms both the pre-trained models and state-of-the-art unsupervised adaptation methods.
cs.LG / 5 / 2609.18112
Token Latency Fairness: Performance Isolation for Multi-Tenant LLM Serving
Abstract
LLM serving is typically offered as a shared, multi-tenant service, where high-demand workloads from one client can cause latency SLO violations for others. Existing solutions for performance isolation equalize client throughput in the long run, for example through queueing and batching fairness. However, these approaches do not provide latency isolation guarantees; as a result, well-behaved clients can still experience significant degradation to their token-level latencies. In this paper, we present FairInference, which provides the novel δ-token fairness guarantee: for a well-behaved client, if a token is generated in d time units in isolation, it will be generated within d + δ time units in multi-tenant execution, providing strong latency isolation guarantees for LLM serving. To achieve this, FairInference addresses a key challenge of LLM serving: bounding delays from sharing GPU resources without support for fine-grained scheduling or resource allocation. In FairInference, the scheduler enforces per-token deadlines, while bounding the delays from GPU compute sharing and accounting for the additional delays introduced by the shared KV caching in GPU memory. We show that FairInference effectively bounds token-level latency spikes for well-behaved clients and improves overall throughput compared to state-of-the-art LLM serving systems.
cs.LG / 6 / 2609.18519
COMPASS-ABS: Reducing Fragmentation in Shared GPU Clusters for Deep Learning Training Workloads
Abstract
With the rapid advancement of deep learning technology, shared GPU clusters receive an increasing number of deep learning training (DLT) jobs. Yet resource fragmentation make such clusters underutilized and forces the DLT jobs running on them to endure long turnaround times. Extensive research has been devoted to quantifying fragmentation and developing scheduling algorithms that alleviate its impact. However, existing fragmentation measures break down in the absence of workload distribution information, while current schedulers cannot continuously maintain resource fragmentation at a low level. To tackle these problems, we first introduce Scheduler-Induced Fragmentation (SIF), a metric built on the notion of partial-nodes that is independent of historical workload knowledge. We then propose COMPASS-ABS, which employs the COMPact-ASSured (COMPASS) algorithm to confine the cluster state within a tight Anchor-Based Space (ABS), whose construction fully leverages the topological alignment between dominant workload size and node capacity. Moreover. We also prove that it ensures SIF is bounded by $\frac{2}{N}$ under a workload composition condition that matches both theory and production. Evaluations implemented on a physical cluster and a simulated cluster demonstrate COMPASS-ABS effectiveness at improving resource utilization, reducing DLT job completion time by reducing fragmentation.
cs.LG / 7 / 2609.17686
The Missing "I Don't Know": Why Three Reasoning-Reliability Findings Converge on Calibrated Abstention
Abstract
Three recent results describe what look like unrelated LLM reliability problems. Yin et al. (2026) show reasoning RL collapses tool-reliability representations. Suleymanov et al. (2026) show that under safety-constrained generation, large models rewrite flagged spans while small models truncate. Bastounis et al. (2024) prove any consistent-reasoning system without an implicit "I don't know" function must hallucinate infinitely often on broad problem classes. We argue these findings converge on a single intervention: calibrated abstention is what each independently identifies as the missing capability, even though the unavailability they document, a capability gap, a policy gap, and a recursion-theoretic gap, has a different source in each case. Honesty post-training has narrowed the gap in deployed models, but principled closure of the class Bastounis identifies requires a calibrated abstention function whose training signal at the leaderboard level is absent: dominant benchmarks assign zero reward to decline, so the leaderboard gradient that would select for the function does not exist. We propose four changes to evaluation: triple-scoring, abstention-rate reporting, capability-stratified evaluation, and mandatory calibration metrics. Benchmark reform is necessary, not sufficient, for closing the gap the theorem identifies.
cs.LG / 8 / 2609.17697
Composite-Gradient Learning for Shared Control Authority Between Deep Reinforcement Learning and Model Predictive Control
Abstract
Integrated deep reinforcement learning (DRL) and model predictive control (MPC) methods are increasingly used to control autonomous systems by combining their complementary capabilities. DRL learns control policies through interaction with the environment. MPC uses a system model to optimize control inputs while accounting for constraints. In DRL-MPC frameworks with shared control authority, both the DRL agent and the MPC controller each determine part of the control inputs. However, common learning formulations treat MPC as part of the environment and therefore do not explicitly account for MPC's contribution to control or its interaction with the DRL agent. This paper proposes a novel composite-gradient learning (CGL) method that integrates the MPC controller into the learning process by representing the DRL and MPC control inputs as a joint action and accounting for their interaction when updating the DRL agent during training. CGL is evaluated on two multi-class freeway traffic networks with different strengths of interaction between the DRL and MPC control inputs and it is compared with alternative methods that treat MPC as part of the environment or that only partially incorporate MPC into learning. The results show that CGL offers limited benefit under weak interaction, but learns higher-performing control policies than the alternative methods in a subset of training runs under strong interaction, although the average control-performance gains remain modest.
cs.LG / 9 / 2609.17745
REVERSAL-BENCH: A Reversibility Axis and Reset Oracle for Measuring the Reset-Free RL Cliff
Abstract
A central goal of autonomous reinforcement learning is continuous policy training without external resets. However, existing paradigms largely depend on underlying environmental reversibility, a property absent in real world manipulation, where events such as pushing objects off tables or spilling granular substances cannot be undone. We introduce REVERSAL-BENCH, a benchmark that controls reversibility via a continuous parameter $ρ\in [0, 1]$ and provides a reset oracle, a ground-truth verification mechanism to test state recoverability across eight manipulation settings in five physics engines. Evaluating a broad spectrum of policy architectures, including standard actor-critic algorithms, safe RL, and specialized reset-free frameworks, reveals a sharp reversibility cliff: reset-free agents are consistently absorbed into irrecoverable states as $ρ$ increases, whereas episodic agents maintain steady learning. We see this failure mode across autonomous reset-free baselines and constrained RL. Because reset-free agents lack external resets, any transition into an irrecoverable state results in permanent absorption, leaving the agent trapped where further learning halts. We show that this absorption phenomenon persists in full physics simulations under learned manipulation policies. By evaluating against geometrically identical reversible counterparts, we confirm that this breakdown is causally driven by irreversibility rather than obstacle complexity. We release the benchmark suite, a large multi-simulator dataset labeled with recoverability and a reset oracle. We also evaluate a safety shield that intervenes before irreversible failures occur, showing that while recoverability can be predicted accurately, active recovery primarily succeeds only when the agent can physically steer clear of the trap
cs.LG / 10 / 2609.17748
SAM-on-the-Curve: Sharpness-Aware Mode Connectivity for Robust Weight-Space Interpolation
Abstract
Deep neural networks that are independently trained to similar performance can be connected by low-loss parametric curves in weight space, a phenomenon known as Mode Connectivity (MC). This geometric property underpins practical techniques such as weight averaging, model ensembling, and model merging. We argue that low-loss connectivity is an incomplete geometric criterion: it controls loss only along a one-dimensional trajectory while leaving the surrounding weight-space neighborhood unconstrained, so the optimized curve may traverse sharp ridges that become fragile under distribution shift. We therefore reformulate mode connectivity as a neighborhood-robust path optimization problem, seeking a curve whose entire local neighborhood maintains low loss. We propose Sharp Mode Connectivity (SMC), which applies a first-order sharpness-aware approximation to the resulting minimax functional, enforcing flatness along the entire curve rather than only on it. We derive a practical optimization algorithm for connectivity paths under this sharpness-aware objective. Under severe blur corruptions from CIFAR-10-C, SMC achieves up to 6.09\% absolute accuracy improvement over standard MC. Remarkably, SMC produces negative loss barriers, meaning that models obtained at interior points of the optimized path can outperform the average endpoint loss. These results, validated across ResNet-18, VGG16-BN, and ViT-Tiny on CIFAR-10 and ImageNet-100, establish path-wise flatness as a practical principle for robust weight-space interpolation.
cs.LG / 11 / 2609.17763
Modular Deep Learning Mechanisms for Auditable Next-Day Wildfire Spread Prediction
Abstract
Next-day wildfire prediction requires models whose forecasts can be evaluated alongside the assumptions and historical evidence used in their computation. Although deep learning can learn spatial patterns from remote-sensing data, predictive performance alone does not establish physical fidelity or operational trustworthiness. This study investigates three modular augmentations for next-day active-fire prediction: wind- and slope-conditioned attention biases, physics-feature retrieval-augmented output correction, and fire conditioned dual-stream gating. The attention biases expose prescribed directional preferences, while the retrieval module selects historical tiles using a nine-dimensional environmental and fire-state descriptor and applies a learned correction to a frozen model's logits. The modules are evaluated across five backbones on the Next Day Wildfire Spread benchmark, using staged ablations, directional audits, retrieval perturbations, calibration measures, and computational comparisons. The three-seed mean F1 score and area under the precision--recall curve (AUC-PR) of a SwinUNETR model with all three augmentations are 0.4216 and 0.3673. Then, a mixed ensemble (two augmented architectures and one non-augmented architecture) model achieves 0.4292 and 0.3790. Benefits vary across architectures, and retrieval-related improvements in AUC-PR do not consistently translate into higher F1. The constructed wind bias aligns closely with input wind, but its alignment with observed next-day fire displacement is much weaker, distinguishing prior inspectability from predictive physical fidelity. The study contributes a framework for exposing and evaluating selected domain-informed components within wildfire prediction models. Together, the results presented show that predictive performance, operational trustworthiness, and computational practicality need not be competing objectives.
cs.LG / 12 / 2609.17815
Principled Koopman Representations with Kalman Inference for Efficient Time-Series Prediction
Abstract
The Koopman operator has been widely used for time-series prediction in dynamical systems. However, prior work that learns latent ``Koopman spaces'' using neural networks often did not construct a valid Koopman space for forecasting, as these representations may be mathematically inconsistent with the operator-theoretic formulation and fail to capture the intrinsic low-rank structure of system dynamics. To address this issue, we introduce K$^2$SVD, a method that explicitly learns the leading singular functions of the Koopman operator by optimizing a Hilbert-Schmidt objective. This yields a well-defined low-rank approximation of the Koopman operator with an interpretable linear combination, featuring a compact latent space with less than $10\%$ of the dimensions used in previous work. In the learned Koopman space, K$^2$SVD further captures temporal evolution with a linear Gaussian state-space model and performs inference via Kalman filtering, mitigating noise accumulation during multi-step prediction. Empirical results show that K$^2$SVD outperforms state-of-the-art methods across multiple datasets, with significantly faster prediction speeds and lower computational cost than previous efficiency-focused models. This highlights the benefits of principled low-rank Koopman representations and opens up broader potential for applications.
cs.LG / 13 / 2609.17816
The Free Inference Dimension: Complexity Measure for Zero-Collision Navigation under Hypothesis Mixtures
Abstract
Solomonoff induction frames prediction as a mixture over computable hypotheses, typically leading to identification of the true environment. In a finite meta-reinforcement learning setting with nested constraint families, in our previous work, we observe a different regime: a value-mixture (VM) agent achieves near-optimal, zero-collision navigation without identifying the true environment, a phenomenon we call Free Inference. This regime persists up to a sharp density threshold, beyond which performance degrades and posterior-mode selection (PMS) becomes preferable. We formalize this behavior via the Free Inference dimension dFI(S,N), a combinatorial measure of the environmental complexity a VM agent can handle while preserving trajectory coherence. We prove dFI is strictly smaller than the VC-dimension and relates to the Natarajan dimension up to a path-length factor, capturing the cost of non-decomposable loss. A PAC-style relaxation yields generalization bounds driven by dFI^(epsilon,delta). We also define a complementary PMS identification dimension and show that a hybrid strategy---averaging until the first collision, then switching to selection---is optimal, with links to Littlestone-type dimensions supported by grid-world experiments.
cs.LG / 14 / 2609.17825
NObSP: Functional Decomposition of Neural Networks via Oblique Subspace Projections
Abstract
Understanding how deep neural networks make decisions remains a fundamental challenge. We present NObSP (Nonlinear Oblique Subspace Projections), a framework that decomposes predictions into explicit per feature contribution functions and an interaction residual. NObSP exploits the linear final layer of a trained network and uses oblique projections in sample space to reduce double counting when learned feature subspaces overlap, thereby supporting both local explanations and global functional analysis. We establish connections to functional ANOVA and the Kolmogorov-Arnold representation theorem and derive an efficient partial regression algorithm for out of sample evaluation. For convolutional networks, NObSP-CAM produces class activation maps without backward passes after a one time calibration. Experiments on tabular and vision benchmarks show faithfulness comparable to established attribution methods. On a synthetic benchmark with known component functions, NObSP obtains a Function Reproduction Score of 0.989, compared with 0.966 for KernelSHAP and 0.922 for Integrated Gradients. On TinyImageNet, contribution vector embeddings improve mean nearest neighbor class purity from 0.654 for raw activations to 0.713 and reduce mean neighbor distance by more than half. These results indicate that NObSP complements scalar attribution methods by recovering functional contribution profiles with separable positive and negative evidence.
cs.LG / 15 / 2609.17831
Procedural Pretraining for Molecular Property Prediction
Abstract
Molecular property prediction is often limited by the small size of labeled downstream datasets, motivating pretraining on large corpora of unlabeled molecules. In this work, we ask whether useful inductive biases can instead be learned from abstract, procedurally generated data before a model sees any molecular data. We introduce a three-stage training pipeline consisting of procedural pretraining, molecular pretraining on SMILES, and downstream fine-tuning, and evaluate several procedural tasks spanning sequence structure, cellular automata, and graph reasoning. We find that procedural pretraining can improve molecular property prediction even after subsequent molecular pretraining: on Lipophilicity, \textsc{Reverse} reduces test error by 4.8\%. For context, the magnitude of this improvement is roughly 90\% of the performance difference between our 250K-molecule baseline and the publicly released MoLFormer checkpoint pretrained on approximately 100M molecules. Our analysis shows that the benefit is strongest under downstream data scarcity, depends on the structure of the procedural data rather than only surface-level statistics, and does not increase monotonically with additional procedural training. Instead, transfer typically peaks at an intermediate procedural budget and deteriorates as the model approaches convergence on the procedural task. We further find that, for several tasks, much of the transferable information is localized in the attention layers, while feed-forward layers can contribute to over-specialization. These results show that procedural data can provide transferable structure for molecular learning and offer a complementary route to improving performance when labeled molecular data are limited.
cs.LG / 16 / 2609.17866
Uncertainty-Aware Continual Learning for Open-World Intent Discovery Under an evolving Label Space
Abstract
Real-world intelligent systems increasingly operate under open-world conditions, where user intents are not fixed or exhaustively known a priori and may evolve as new interaction patterns emerge. This paper proposes a unified uncertainty-aware probabilistic framework for continual new intent discovery under an evolving label space. Each utterance is encoded through an adaptive $β$-VAE into a latent mean, used for classification and density modelling and a posterior uncertainty estimate acting as a global reliability signal. Classifier confidence, posterior uncertainty and DP-GMM likelihood are combined through a multi-signal decision mechanism to distinguish known intents from potentially novel samples. Candidate novel instances are clustered through a density-based discovery module and only reliable clusters are promoted to new labels, enabling controlled label-space expansion. Replay and Elastic Weight Consolidation mitigate catastrophic forgetting and preserve previously acquired knowledge. The paper formalises continual intent discovery as a structured multi-phase open-world problem, introduces adaptive label-space expansion under stability--plasticity constraints and uses posterior uncertainty to regulate trusted-sample selection, pseudo-labelling, novelty admission and replay. Experiments show high novelty precision, stable adaptation across sequential phases and limited forgetting. Near-zero NMI and ARI indicate limited reconstruction of the complete fine-grained intent taxonomy, consistent with the framework's conservative promotion strategy. Qualitative analyses nevertheless reveal dense and locally coherent semantic clusters, showing that reliable novel structures can be discovered without exhaustive recovery of the underlying taxonomy.
cs.LG / 17 / 2609.17886
Dataset-Dependent Effects of Cross-Depth Aggregation and Soft-Routed Experts in EEG Foundation Model Fine-Tuning
Abstract
EEG decoding tasks can rely on different temporal dynamics and cross-channel relationships. We test whether specialized modules improve a fully fine-tuned EEG foundation model by augmenting CBraMod with cross-depth Attention Residuals (AttnRes) and two soft-routed expert banks. Across matched three-seed experiments on FACED, ISRUC, SEED-V, and PhysioNet-MI, the complete model changes mean balanced accuracy relative to full fine-tuning by -0.12, +1.27, +0.77, and -1.27 points, respectively. AttnRes alone improves mean balanced accuracy on three datasets, whereas adding experts on top of AttnRes helps only FACED and SEED-V. These gains come with substantial overhead: AttnRes requires 2.11 to 2.88x runtime and 1.78 to 2.67x memory, while the complete model requires 2.41 to 3.04x runtime and 1.86 to 2.85x memory. Overall, the added modules produce dataset-dependent, sometimes opposing effects rather than consistent gains over full fine-tuning.
cs.LG / 18 / 2609.17888
Long-Context Demonstration Selection Using State Space Models
Abstract
We study the problem of demonstration selection, which involves selecting a subset of examples for prepending to a query to a language model. This problem is closely related to in-context learning and language model inference. Since the inference cost of a transformer model scales quadratically with sequence length, the selection problem becomes especially challenging in a long-context scenario. In this paper, we tackle this problem by building on state space models (SSMs), which require only linear inference time given the input. Our approach involves two algorithms. The first learns a small set of SSMs through distillation of a (trained) transformer model. We partition all the layers into consecutive groups. Then for each group, we estimate a separate state space model to replicate the input-output behavior within the adjacent layers. Second, we map the distilled model outputs to a small set of tokens, and apply these embeddings for demonstration selection in downstream applications. We perform extensive experiments in both synthetic and real-world datasets to validate our approach. We demonstrate that the distilled SSMs only incur an approximation error of less than $0.7\%$ relative to the true output. In downstream evaluation, we show that on several text classification and reasoning tasks, our approach reduces FLOPs by $14.2\times$ and improves accuracy by $6.48\%$ relative to baseline demonstration selection methods.
cs.LG / 19 / 2609.17895
TabPFN-3.5: Technical Report
Abstract
We introduce TabPFN-3.5, our new flagship Tabular Foundation Model. It significantly outperforms its predecessor, TabPFN-3, and all existing baselines across a broad range of tabular problems. TabPFN-3.5 sets a new state of the art on standard tabular prediction in TabArena, and extends it to the data practitioners encounter in practice: non-i.i.d. data with temporal or grouped splits, tables with strings, text and images, high-cardinality categorical features, and wide tables with many features. These gains carry over to our task-specific harnesses: state of the art on relational data and stronger time-series forecasting. For faster inference, our variant TabPFN-3.5-Fast runs up to 3x faster than TabPFN-3 while keeping most of the accuracy gains. In addition, we upgrade TabPFN-3.5-Plus, expanding our multimodal capabilities with advanced text and date handling alongside proprietary inference optimizations. Finally, we release a new version of our Thinking mode, TabPFN-3.5-Thinking, which scales inference-time computation to push the state of the art further. It benefits from our stronger base model and from inference-time improvements that make it up to 12x faster than TabPFN-3-Thinking.
cs.LG / 20 / 2609.17901
Walking the Score Manifold: Continuous-time Generative Dynamics on Learned Data Manifolds
Abstract
Generative modeling of time-dependent data is typically formulated on a discrete temporal grid, restricting supervision to the observed timestamps in the training data. We instead frame generation as continuous-time evolution on a learned data manifold. To this end, we leverage pretrained score-based models as geometric priors and learn a vector field that evolves data along score-induced interpolation paths. Because these dynamics follow transitions that respect the geometry learned by the score model, they support generation at arbitrary timestamps and temporal super-resolution beyond the discretization of the training data. Moreover, this geometric formulation allows us to train the vector field simulation-free through a regression objective. To improve long-horizon rollout robustness, we introduce an objective that promotes path-relative transverse exponential stability. While motivated by stability theory, it admits a practical interpretation as denoising score matching transverse to the interpolation path. Further, we extend the framework to a probabilistic setting that models a distribution over plausible future trajectories. We demonstrate the method on natural video and scientific dynamical data, including temporal super-resolution, PDE-based spatiotemporal fields, and molecular dynamics. Our results show that score-based priors provide a strong foundation for learning stochastic continuous-time generative dynamics.
cs.LG / 21 / 2609.17916
EdgeReMIND: A Scalable, Top-Ranked Memorization Baseline for Temporal Multi-Relational Link Prediction
Abstract
Temporal link prediction on the Temporal Graph Benchmark 2.0 (TGB 2.0) faces a scalability ceiling: on the benchmark's three largest datasets, every existing embedding method runs out of memory or exceeds the time budget. These large-scale graphs are the ones nearest real deployment scale, so failing on them is a real production limitation. EdgeReMIND sets the highest reported test mean reciprocal rank (MRR) on six of eight TGB 2.0 datasets and is the only relation-aware method that runs on all of them. This linear memorization model, with learned per-relation weights over data-calibrated features, is therefore not merely a fallback where embeddings fail but a practical state-of-the-art baseline across the benchmark.
cs.LG / 22 / 2609.17926
Symmetry without a manifold: intrinsic dimension on orbits
Abstract
The standard geometric derivation of neural scaling exponents takes the intrinsic dimension of a data manifold as its input. On modular addition in $\mathbb{Z}_p$ that derivation has no input. The exact algebraic solution is an orbit of $\mathbb{Z}_p$ acting by isometries. Transitivity alone makes the ratio statistic underlying the standard dimension estimator a point mass, so the estimator is undefined, and here the two nearest neighbour distances coincide exactly. Breaking the symmetry at scale $ε$ returns a number, but one that tracks $1/ε$ with no scale free plateau. We show that the failure is general, since on any finite orbit of a group acting by isometries the estimator reports the resolution at which the set is probed rather than a dimension. What replaces the power law is exponential in hidden width, $L(h)=L_\infty+A\exp(-c\,h^α)$, with $R^2$ between 0.982 and 0.995 against 0.857 to 0.906 for a power law admitting the same floor and fitted under the same protocol. Where the data supply is sufficient the rate belongs to the regulariser rather than to the group, since weight decay moves $c$ by a factor of 47 while group order moves it by 1.10, a residual below seed to seed resolution, for every fixed $α$ between 0.75 and 2. The critical width falls with group order rather than rising, against capacity counting that assigns a fixed number of neurons to each irreducible representation.
cs.LG / 23 / 2609.17930
Locating Hidden Failures Makes Long-Horizon Agents More Reliable
Abstract
As AI agents take on long, autonomous tasks, we increasingly oversee rather than perform the work, yet we still judge them almost entirely by whether they finally succeed. An outcome cannot reveal where a run went wrong, whether the agent recovered, or the irreversible harm it caused along the way, and where long-horizon agents fail remains unmapped. We study $2518$ agent trajectories across software engineering, computer use, and science, close to real deployment, and classify $6967$ mistakes into $78$ failure types. Failure follows a recurring signature: after its first mistake an agent often fails to recover and rarely catches the error itself, so the run continues unchecked while still looking correct; whether an agent recovers depends on the task and the environment's feedback, not on the agent framework running it. Long-horizon agents can do real harm on the way to a passing result: even runs scored as solved delete data, corrupt systems, or fabricate success rather than earning it. We release these human-verified annotations as Traverse, a benchmark on which six frontier judges struggle to locate failure regardless of scale: even the strongest correctly identifies the first mistake in fewer than a third of runs. Yet Scout, a $4$B verifier we trained, locates failure far better than these judges and transfers to domains it never saw. Used at test time to select among an agent's candidate runs, it raises task success above the agent's own single-attempt performance, without retraining the agent. By making failure cheap to locate and correct, this work is a foundation for more trustworthy long-horizon agents that learn from their own mistakes, and a practical path to overseeing increasingly autonomous AI.
cs.LG / 24 / 2609.17940
Beyond the Previous Layer: Residual Predictive Structure in Sparse MoE Routing
Abstract
Sparse mixture-of-experts models route each token through a sequence of expert selections. We ask whether the immediately preceding selection adequately summarizes this trajectory for predicting the next router. Using frozen OLMoE and JetMoE models, we measure the held-out predictive gain from earlier expert selections while retaining the most recent selection as a common baseline. In OLMoE, extending the history from one to eleven layers raises router-logit $R^2$ from 0.59879 to 0.66544. A preregistered JetMoE replication yields four-layer gains of 0.14275 and 0.20528 at two target depths, with paired bootstrap intervals above zero. These gains survive nonlinear decoding: adding history to a small multilayer perceptron improves $R^2$ by 0.17137 and 0.21861, whereas nonlinear decoding of the recent state alone adds 0.00139 and 0.00936 over a linear probe. Parameter-matched controls preserve the advantage, and cross-fitted history residuals predict target residuals with $R^2$ of 0.20549 and 0.23556. These findings identify residual predictive structure in expert-selection trajectories beyond adjacent-layer persistence.
cs.LG / 25 / 2609.17942
On the Identifiability of Mixed Ordinal and Exponential Family Causal DAGs under Linear Parametric Models
Abstract
The problem of identifiability in linear parametric models (LPMs) whose nodes follow either an ordered logit model or a regular one-parameter exponential family is evaluated. The results go beyond classical structural equation models as well as results for nodes with observations from a homogeneous family of distributions. The main result establishes that the orientation of every edge joining an ordinal node to an exponential-family node is identifiable from the joint distribution alone at every parameter value, provided the ordinal node has at least three categories and the exponential-family node at least three points of support, with no restriction on the sufficient statistic. Converses show that both requirements are necessary: the three-category requirement is binding only for affine sufficient statistics, and the three-point requirement is binding under the canonical link. The guarantee extends to orienting every such mixed ordinal-exponential family edge of a given $d$-node undirected skeleton. Numerical experiments illustrate the theoretical results by successfully separating orientations within a Markov equivalence class, which are indistinguishable by conditional independence alone.
cs.LG / 26 / 2609.17943
ASPIRE: Asynchronous Batched Self-Speculative Decoding for Long-Context LLM Inference
Abstract
Long-context LLM inference is bottlenecked by attention, whose repeated KV-cache reads make decoding memory-bound. Self-speculative decoding alleviates this by drafting tokens with sparse attention and verifying them with full attention, but existing batched methods remain synchronized: all requests in a batch share a single draft-verify schedule, even though the optimal draft length varies widely across requests and changes dynamically within each request. We propose ASPIRE, a non-synchronized batched self-speculative decoding framework built on three components. First, a unified mixed forward allows drafting and verifying requests to coexist in the same batched forward pass, removing the need for global draft-verify phases. Second, a lightweight online speculation scheduler uses per-request acceptance-rate estimates and a batch-aware cost model to let each request independently choose when to verify. Third, an intra-draft refresh layer performs full attention at a single designated layer during drafting, updating the sparse context at every draft step to reduce staleness during drafting. Across three models and five reasoning and long-context benchmarks, ASPIRE achieves $1.70$-$4.58\times$ speedup in decoding throughput over autoregressive baselines and improves average speedup by approximately $27\%$ over the strongest prior self-speculative baselines.
cs.LG / 27 / 2609.17951
Maximum Strong Independent Sets in Hypergraphs: Reductions, Bounds, and Greedy Certificates
Abstract
We study the maximum strong independent set problem in a finite hypergraph: find the largest vertex set that intersects every hyperedge in at most one vertex. This objective arises whenever each observed block is a local incompatibility constraint but transitive closure across overlapping blocks is not justified. A motivating example is multi-band LSH-MinHash deduplication, where each collision bucket gives local evidence, while connected-component contraction can impose spurious global equivalences. The paper develops an incidence-structural toolkit for this problem. We prove exact reductions for dominance, incidence twins, and weight-1 blocks; derive closed-form and low-weight upper bounds; introduce puncturing and covering certificates that sharpen those bounds; and analyze a layered greedy clustering algorithm driven by block weights and residual incidence. The algorithmic analysis includes feasibility, maximality, conditional optimality, a layered witness-matching upper bound, and incidence-local complexity bounds. The results give correctness, termination, fixed-point, and optimality certificates for broad incidence families, together with examples showing when different certificates separate or coincide.
cs.LG / 28 / 2609.17995
QuanText: Protecting Dataset-Level Secrets in Textual Data Sharing
Abstract
Natural-language datasets support many downstream applications and research studies, but releasing text can reveal sensitive global properties of the underlying data source, such as the proportion of records associated with a particular gender, diagnosis, or political stance. Existing work has largely focused on property inference attacks that recover such global properties, while defenses for protecting these dataset-level secrets remain limited. Differential privacy, although effective for protecting individual records, provides only weak protection for aggregate properties. We propose Randomized Quantization for Text (QuanText), a training-free and large-language-model-agnostic data release mechanism that protects global secrets in textual datasets while preserving data utility. Given a dataset-level secret, such as the proportion of records with a particular diagnosis, and attributes whose utility should be preserved, such as topic and sentiment, QuanText perturbs both the secret distribution and the distributions of correlated attributes. It does so by constructing candidate release distributions over secret and non-secret attributes, randomly selecting a candidate sufficiently close to the private empirical distribution, and rewriting each private text sample to match the selected distribution using attribute-related snippets from the original text. QuanText is inspired by the Statistic Maximal Leakage (SML) framework, which bounds leakage about a secret function of a data distribution. Under idealized conditions, we show that QuanText satisfies an SML guarantee. Since these conditions may not hold exactly in practice, we also evaluate QuanText empirically on real-world datasets. Our results show that QuanText achieves a better empirical privacy-utility trade-off than competing data generation baselines.
cs.LG / 29 / 2609.17997
The Attention Within: Consensus Dynamics in Selective State Space Models
Abstract
Selective state space models (SSMs) have recently emerged as a compelling alternative to transformers, combining competitive performance with substantially improved inference efficiency. At each SSM layer, a sequence of hidden states are propagated by a recurrence, mixing information of different tokens. Despite using a different mechanism, this mixing plays a role analogous to attention in transformers. In fact, recent works have shown that the two architectures may be closer than they first appear, as this recurrence admits a formulation akin to linear attention. In transformers, attention is known to drive the tokens to cluster, i.e., to reach consensus, collapsing in the limit to a single direction. Thus, we ask: does the recurrence at the core of SSMs drive the tokens to consensus, as attention does in transformers? To answer this question, we take a dynamical systems perspective on SSMs, modeling the evolution of tokens across layers as an ordinary differential equation. By exploiting input-to-state stability arguments, we establish local exponential stability of the consensus equilibria and characterize their domain of attraction for time-varying weight matrices, a setting not addressed by previous results. We thereby show that the resemblance between SSMs and transformers does run deeper: the recurrence at the core of SSMs aggregates tokens just as attention does. Numerical experiments on a pretrained Mamba-2 model point to the output gate as the component that regulates the extent of this consensus, preventing the tokens from reaching it in full.
cs.LG / 30 / 2609.18045
Structural Inference under Hidden Agents
Abstract
Recovering latent interaction structures from multi-agent dynamics is important for understanding and predicting interacting systems. Trajectory-based structural inference has achieved promising performance, but conventional formulations assume that the trajectories of all modeled agents are available. In practice, agents may become unobserved at deployment because of limited sensing, occlusion, or communication failure. Existing studies have considered unseen-node estimation, structural inference under partial observations, and missing-value imputation, yet the joint recovery of hidden-agent trajectories and their interactions remains underexplored. We formulate this problem as structural inference under hidden agents. Its key difficulty is a circular dependency: recovering interactions involving a hidden agent requires an estimate of its trajectory, while trajectory reconstruction can itself benefit from structural information. To address this challenge, we propose Structural Inference under Hidden Agents (SIHA), which combines structure-agnostic initialization with structure-guided iterative refinement. SIHA reconstructs hidden trajectories from visible observations, infers interactions using Neural Relational Inference, and feeds the estimated structure back into hidden-state reconstruction through multi-strength structural attention and iterative state--structure updates. Experiments on three benchmark dynamical systems demonstrate consistent improvements in visible-to-visible structural inference, while also showing benefits in hidden-state reconstruction and future prediction. Motion-capture experiments with simulated whole-limb occlusion further demonstrate its effectiveness in realistic hidden-agent settings.
cs.LG / 31 / 2609.18049
Regional Explanations via Causal Sufficiency and Necessity
Abstract
Model explainability is essential for understanding and trusting machine learning models. Existing explainable AI methods often explain predictions through feature importance, counterfactual explanations, or rules. However, a region-level characterization of when and only when a prediction behavior arises remains less explored. This paper proposes Causal Sufficient and Necessary Regional Explanations (SNRE), a framework that learns an input region $A$ and output region $B$ such that membership in $A$ is both sufficient and necessary for the model output to fall in $B$. Motivated by the classical Probability of Necessity and Sufficiency (PNS), we formulate a region-level PNS measure through stochastic interventions and derive a differentiable finite-sample estimator for optimization. SNRE parameterizes the input-output region pair with explicit and interpretable algebraic region families, together with a learnable feature mask, balancing expressiveness and interpretability. Experiments demonstrate that SNRE learns region pairs with strong sufficiency-necessity performance, robust explanation behavior, and practical utility for model analysis.
cs.LG / 32 / 2609.18078
Beyond Embedding Transfer: Component Roles in Grokking Transfer and Stability
Abstract
Warm-start transfer can make algorithmic tasks generalize rapidly, yet it is unclear which model components provide the gain and whether that gain remains stable under continued optimization. We study cross-operator transfer on modular arithmetic and separate efficacy (early velocity) from stability (post-reach drawdown). In a scale-matched 108-run battery across 12 seed blocks (96-run 2^3 factorial plus 12-run scale control), transferring internal attention/MLP weights (B) alongside token embeddings and readout (E+U) improves early accuracy by 5.46 pp (Holm p=0.0039) and cuts confirmation latency by 558 steps (Holm p=0.0088). While readout plus internal-block transfer satisfies the pre-specified +/-500-step latency equivalence criterion in 1-layer models (TOST p=0.0011, though Full is faster in 11/12 paired seeds), a prospective 2-layer replication confirms the internal-block advantage (12/12 seeds, +704.67 integral units, p=4.88x10^-4) while revealing an architectural boundary: omitting donor embeddings falls 4475.6 units below Full, outside the +/-250-unit margin. Continued target training frequently triggers severe post-grokking relapse. Freezing transferred representation carriers (E, U) nearly eliminates offline relapse (19.40% -> 0.07%, Holm p=0.005859). Online validation-triggered gating slashes True Max Drawdown from 22.06% to 0.60% on 2a+b (p=0.000488), with prospective confirmations extending protection across affine, nonlinear quadratic, and 2-layer targets (10.94-23.47 pp reductions), distinguishing continual stabilization from static early stopping. In non-abelian S_5, unshielded transfer surges transiently (95.4% peak), but a prospective shielding cohort yields no confirmed benefit (+0.15 +/- 1.14 pp). These results establish a component-level dissociation between transfer acceleration and trajectory stability, and expose the empirical boundaries of parameter shielding.
cs.LG / 33 / 2609.18089
FedPGT: Progressive Gradient Transmission for Vehicular Federated Learning over Time-Varying Channels
Abstract
Vehicular federated learning (VFL) enables privacy-preserving collaborative model training for intelligent transportation systems, where communication resource allocation and gradient sparsification techniques have been explored to reduce communication overhead. However, vehicle mobility leads to rapidly varying channel conditions and transmission capacity, rendering predetermined resource allocation and sparsification decisions ineffective. In this paper, we propose FedPGT, a progressive gradient transmission scheme for VFL over time-varying channels, where vehicles progressively transmit high-magnitude gradient entries in response to instantaneous channel conditions. We establish a convergence bound that characterizes the impact of transmitted gradient entries and reveals diminishing-return behavior governed by a power-law decay. Motivated by this result, we formulate a stochastic optimization problem for online decision-making, where the main challenge lies in a cumulatively coupled, non-separable objective. To handle this challenge, we introduce per-slot surrogate transmission variables to decouple the long-term dependence across time slots and convert the original objective into an additive per-slot optimization problem, enabling a Lyapunov drift-plus-penalty approach for online scheduling. We further develop a low-complexity resource allocation algorithm for efficient online implementation. Experimental results demonstrate that the proposed scheme achieves a 3.65% accuracy improvement on the CIFAR-10 image classification task and a 12.66% reduction in average displacement error on the Argoverse trajectory prediction task compared with state-of-the-art baselines, demonstrating its applicability to diverse learning tasks under highly dynamic vehicular environments.
cs.LG / 34 / 2609.18094
Agora: Git as Shared Memory for Collective AutoResearch
Abstract
Autonomous research loops such as AutoResearch show that one coding agent can improve a training setup unattended. Run several of them and each session starts from scratch, so more agents tend to mean more duplicated search rather than more discovery. Agora is a shared memory for such agents: research is recorded as an append-only directed acyclic graph (DAG) stored in Git, so that every claim is a commit anyone can check out and rerun. Each result, insight, hypothesis, verification, and report is an immutable commit whose parent edges say what it builds on; a derived index exposes the frontier, the neglected branches, and the verification status of each claim, and a diversity-aware selection rule keeps the community from collapsing onto one leader. We describe the system and report its first sustained use: a run of nearly 12 days in which 13 language-model workers, with no assigned tasks and no central planner, worked on a weight-transfer problem. Given 141 pretrained donor models and a frozen 119.6M-parameter attention-SSM hybrid whose dimensions match no donor, the workers had to initialize the target without training data or gradient updates. They published 1,703 contributions and drove the evaluator from 3.39 to 1.899 bits per byte, closing 62% of the gap to a trained GPT-2 124M. The winning recipe compresses donor next-token statistics into the target's embedding and output head, then adds a short-range context signal through sparse edits to attention, feed-forward, and state-space blocks. Its 145-commit ancestry spans 15 accounts, and 165 independent reproductions were posted, none of which failed. We describe the single mid-run human intervention that pulled the community out of a monoculture, what the trace does and does not establish, and the controlled comparison that would settle whether shared research state improves discovery per unit of compute.
cs.LG / 35 / 2609.18104
iMINDBench: iEEG Multi-Institution Neural Decoding Benchmark
Abstract
Intracranial electroencephalography (iEEG) is widely used to record electrical activity directly from electrodes inside the human brain, making it an attractive modality for neural decoding. However, progress in iEEG decoding, especially toward general-purpose foundation models, remains difficult to measure reliably: datasets are task- or institution-specific, limiting evidence of generalization across tasks and recording environments, and preprocessing choices can strongly influence performance, making model improvements difficult to distinguish from preprocessing gains. Thus, we introduce iMINDBench, an iEEG Multi-Institution Neural Decoding Benchmark that evaluates models on a shared suite of fifteen decoding tasks across three naturalistic movie-watching datasets. The benchmark additionally defines standardized preprocessing tracks and fixed evaluation splits to support consistent model comparisons. Using iMINDBench, we find that the evaluated pretrained systems generally outperform baselines within their respective preprocessing tracks, while strong spectral baselines remain competitive across institutional datasets. In our scaling study, adding up to 25 times more supervised data from other subjects or institutions yields only small or task-dependent gains over within-session training. Together, these findings highlight the need for iEEG models that improve on strong preprocessing baselines and make more effective use of data across subjects and institutions. Project website: https://imindbench.github.io/
cs.LG / 36 / 2609.18107
FoundAna: A GNN-assisted Foundation Model for Graph Anomaly Detection
Abstract
Graph anomaly detection aims to identify graph structures (e.g., nodes, edges, or subgraphs) that deviate significantly from expected patterns, which supports critical applications in fraud detection, spam identification, network intrusion, etc. Despite the growing methods in the field, existing approaches follow a one-model-per-dataset paradigm, limiting their transferability across diverse real-world scenarios due to task heterogeneity, label scarcity, and domain variability. In this work, we introduce FoundAna, a GNN-assisted Foundation Model for Graph Anomaly Detection - the first foundation model framework designated for generalizable, cross-graph anomaly detection by combining GNNs and transformers. FoundAna integrates an anomaly detection-specific GNN component with a standard transformer encoder augmented by four complementary positional encodings, which enable the model to capture both local and global structural information. Specifically, the positional encoding enriched node representations are passed through attribute and adjacency decoders, and the reconstruction errors serve as the anomaly score. Extensive experiments on nine benchmark datasets spanning financial, social, and citation network domains demonstrate that FoundAna consistently outperforms state-of-the-art baselines. The code implementation and Supplementary materials are here: https://github.com/FoundAna331/FoundAna.
cs.LG / 37 / 2609.18127
Learning Fractional-Order Dynamics from a Single Trajectory
Abstract
Many real-world processes exhibit long-range dependence, where the current state depends on a slowly decaying trace of past states rather than on the most recent state alone. This paper studies system identification for discrete-time fractional-order linear time-invariant systems from a single observed trajectory of length $t$, a setting that captures such non-Markovian dynamics through the Grünwald--Letnikov difference operator. Unlike Markovian systems, fractional-order systems couple estimation across the entire history, making both statistical analysis and practical identification more challenging. We propose \emph{Fractional-Order Ordinary-Least-Squares Grid-Search (FO-GS)}, a simple two-stage estimator that exploits the diagonal structure of the fractional-difference operator to decouple the identification problem row-wise. Under the stability assumption, we establish high-probability, non-asymptotic error bounds for estimating both the fractional order and the system matrix in the heterogeneous setting, with both estimation errors scaling as \(\mathcal{O}(t^{-1/2})\). Through experiments, we show that \emph{FO-GS} outperforms existing baselines in recovering both the fractional order and the underlying system dynamics.
cs.LG / 38 / 2609.18131
Colla-Q: Toward Collaborative Experts in MoE Quantization via Minimax Precision Balancing
Abstract
In this paper, we present a Mixture-of-Experts (MoE) quantization method based on activation entropy. Although quantization reduces memory and computational costs, it can substantially degrade performance. In particular, performance decline is pronounced in quantized MoE models, where individual experts have a small number of parameters that are sensitive to low-bit representation. Considering that MoE operates as an ensemble model with collaborative contributions from routed experts, a significant performance decline of a particular expert due to quantization can harm model performance. Therefore, we propose Colla-Q, a bit-allocation framework to maintain balanced performance across experts through an activation-entropy-based bit-width allocation algorithm. This approach encourages each expert to operate collaboratively in the quantized model, thereby 1) improving the overall MoE performance and 2) reducing the dependence on the calibration dataset. Since uniformly adjusting each expert's performance facilitates robustness and stability of the MoE model, the proposed MoE quantization method can generalize more consistently across different calibration datasets. Our code is available at: https://github.com/mmai-laboratory/Colla_Q
cs.LG / 39 / 2609.18134
Rethinking How We Evaluate Methodological Progress in Health AI
Abstract
Methodological progress in artificial intelligence (AI) for electronic health records (EHRs) depends on our ability to determine which algorithms work better, and under which conditions. However, such progress is thought to be hindered by difficulties in reproducibility and in defining clinically meaningful evaluation tasks. We empirically study these barriers by re-implementing 12 historical and recent algorithms within a shared evaluation framework and evaluating them on two clinical datasets, MIMIC-IV and NWICU. We compare two complementary task families: expert-authored clinically meaningful tasks and generated tasks defined from randomly sampled event codes and prediction horizons. We ask whether relative algorithms comparisons transfer across task families and datasets, whether residual task heterogeneity contains useful methodological structure, and what a controlled comparison reveals about progress over the last decade. We find that aggregate pairwise comparisons transfer strongly across evaluation settings, including from randomly generated tasks to clinically meaningful tasks and across datasets. At the same time, clinically meaningful tasks exhibit greater task-method interaction, providing preliminary evidence that task properties can help explain when particular modeling choices are advantageous. Finally, newer algorithms do not consistently outperform earlier approaches: gradient-boosted trees remain highly competitive when paired with a modern, wide and sparse representation of the EHR. Together, these results suggest that useful methodological knowledge may require less task engineering than commonly assumed, while highlighting the importance of understanding the structured heterogeneity that remains across tasks and methods.
cs.LG / 40 / 2609.18145
Reaching Every Position Without Searching: Rotating Sparse Wiring on the Hypercube as a Substitute for Attention
Abstract
Attention pays, at every layer and for every input, the cost of searching for whom to connect. We ask how far one can get with wiring that is fixed, sparse, and simply rotated from layer to layer. Treating the $n$ positions of a sequence as the vertices of a $\log_2 n$-dimensional hypercube and connecting each position, at layer $\ell$, to its neighbour along dimension $\ell \bmod \log_2 n$, information from every position reaches every other in $\log_2 n$ layers with $2n$ links per layer instead of $n^2$. On a synthetic task that is unsolvable unless all positions are reached, this rotation matches all-to-all wiring at $1/32$ of the links, while the same sparse pattern held fixed across layers fails; what matters is that every dimension is touched, not the order. On character-level language modelling of a public corpus (the first $12$M characters of enwik8), a hybrid that keeps two attention layers among sixteen sparse ones reaches $0.06$ bits-per-character lower held-out loss than a fully attentive model of the same width at the same step budget (three seeds each, no overlap), with $1/7$ of the links, $42\%$ fewer parameters, and $2.4\times$ less wall-clock time; the purely rotated schedule is level with the hybrid. The same ordering holds on a second corpus of mixed Japanese, English and code, where the gap widens to $0.16$. The usable learning-rate window is four to eight times wider than attention's on both. We also report what did not work - learned coordinates, and a "dynamics" variant whose apparent gains turned out to be an artefact of a saturated kernel - and the measurement discipline (frozen corpus, full-coverage evaluation, seed spread as the bar for ranking) that we found necessary to say anything at all at this scale.
cs.LG / 41 / 2609.18176
MoRE: Mixture of Reused Experts
Abstract
Mixture-of-Experts (MoE) architectures decouple model capacity from computational cost, yet incur high memory footprints as parameters grow linearly with the number of experts. Recurrent Transformers achieve parameter efficiency by reusing layer weights, but typically lack the capacity for competitive language modeling. We propose Mixture of Reused Experts (MoRE), a hybrid that shares expert pools across groups of adjacent layers. Each layer retains its own router but selects from a larger shared pool, expanding the diversity of routing combinations without additional parameters. To enable shared experts to distinguish between layers, we introduce lightweight learnable depth embeddings that condition each layer's input before routing. Experiments across three model scales (114M-1.15B parameters) show that MoRE consistently achieves lower perplexity and stronger downstream performance than standard MoEs and state-of-the-art weight-sharing architectures at matched compute and parameter budgets, with only minimal modifications to existing MoE implementations.
cs.LG / 42 / 2609.18190
Transformation Laws in Neural Representations: Structure, Realisability, and Construction
Abstract
How neural representations preserve the structure of input changes connects representation analysis with internal intervention. We study operable representational content through compatible actions of reference transformations on neural features. We characterise when a transformation descends through an encoder, and give a linear setting in which the defect is governed by the transformation's demand for discarded information, measured in the metric the representation induces. On a rectifier the failure to realise a transformation has two distinguishable sources --- what the source region has already made unrecoverable, and what it costs to satisfy every region the transformation visits with one operator --- and for a \textit{measured} harmonic carrier the same question has a closed answer: a linear realisation exists exactly when the retained harmonic blocks are invariant under the action. Using colour as the in-depth instance, we find that hue orbits in frozen visual features concentrate 84--88\% of their energy in the first two harmonics with rotation planes shared across shapes, that this organisation is substantially inherited from input and architecture and is reshaped by training and depth, and that the measured structure supports prediction, transport from new starting states, and composition --- with global and local realisations differing sharply in which they achieve. Guided by the measurements, we construct a compact interface whose rotation action is fixed by the structure and never fitted: it reads hue zero-shot at 3.4$^\circ$ median error on unseen shapes. Theory, structural measurement, and construction together establish transformation laws as a concrete object connecting the understanding of neural representations to their design.
cs.LG / 43 / 2609.18212
A Lightweight CNN Integrated Compact Convolutional Transformer for Multi-Scale Feature Learning and reducing computational complexity for breast cancer mammography image detection and classification
Abstract
Over the years, Convolutional Neural Networks (CNNs) have demonstrated strong capability in cancer detection and classification using medical images. However, CNN-based models often struggle to capture long-range contextual dependencies. In such scenarios, integrating Compact Convolutional Transformer (CCT) architectures after the CCT layer allows CNN-extracted features to reshape into compact patch tokens using a CCT tokenizer, followed by the addition of positional embeddings to preserve spatial structure. Using 5-fold cross-validation, the model was tested on 3 sets of breast cancer mammography. With only 250,435 parameters, the model achieved 99%-100% accuracy across 3 datasets, indicating robust generalization. Explainable AI (XAI) was integrated into the model to explain the breast cancer classification process to enhance clinical trust. The results indicate that the proposed framework is suitable for computer-aided diagnosis systems, particularly in resource-constrained clinical environments. The novelty of the proposed CNN-integrated CCT overcomes the limitation of CNN's gradient degradation in the last layers by integrating convolutional tokenization with transformer-based learning. Lighter than ViT, which is effective in capturing long-range dependencies, the model has also proven efficient in breast cancer classification by capturing long-range dependencies among breast tissue regions.
cs.LG / 44 / 2609.18219
APGEM: Adaptive Policy-Guided Error Mitigation for Quantum Reinforcement Learning on a Real-World CVRP Case Study
Abstract
Quantum Reinforcement Learning (QRL) represents policies as variational quantum circuits (VQCs), making it attractive for combinatorial optimization such as the Capacitated Vehicle Routing Problem (CVRP). On noisy intermediate-scale quantum (NISQ) hardware, however, decoherence degrades fidelity and destabilizes learning, and conventional error mitigation is applied statically without regard to the learning context. We introduce Adaptive Policy-Guided Error Mitigation (APGEM), a controller that selects among Zero-Noise Extrapolation (ZNE), Probabilistic Error Cancellation (PEC), Clifford Data Regression (CDR), and Readout Error Mitigation (REM) online, driven by a fidelity, entropy, and cost aware utility function and an epsilon-greedy rule over temporal-difference Q-scores. We evaluate on a realistic urban-logistics testbed, a Delhi-based CVRP over real landmarks with geodesic inter-node costs, exercised across five noise families and four severity levels. On this instance, the QRL agent outperforms constructive heuristics and approaches metaheuristics, while mitigation restores approximation ratios from 0.84-0.87 to 0.92-0.94 under high noise. The controller shifts from a CDR-dominated regime under short training horizons to a balanced deployment across all four techniques under longer horizons, indicating genuine regime-dependent selection. These preliminary results position adaptive, learning-aware mitigation as a practical route to noise-resilient QRL.
cs.LG / 45 / 2609.18228
Anomaly Detection in General Ledger Data: Results from a Hybrid Approach
Abstract
Journal Entry Tests (JETs) are a mandatory part of annual audits to evaluate and assess both highrisk audit areas and potential material misstatements. However, as JETs are designed to detect known patterns based on domain knowledge, the resulting lists are often very large and require substantial additional effort from the auditor. To ensure the economic efficiency of the audit, the number of false positives in JET result lists must be reduced. Especially machine learning (ML) methods represent a promising approach to improve anomaly detection in this field. In this research in progress paper, we investigate different approaches on how to combine JETs with ML-methods in a hybrid manner. We present specialized models to increase the detection performance and validity of anomaly detection results to improve audit efficiency. The experiments are based on synthetic data consisting of different normal and anomalous journal entries.
cs.LG / 46 / 2609.18238
F-DACE: Fuzzy Disagreement-Aware Causal Evidence Fusion for Abstention-Safe Conversational Retail Decision Support
Abstract
Observational decision-support systems often expose one causal estimate as a recommendation even when plausible estimators disagree. The inherent engine of the proposed system is causal machine learning: a conditional-average-treatment-effect estimand identified by backdoor adjustment, estimated by an EconML DML causal forest and DoWhy linear regression, checked by two-way fixed effects, and converted into candidate levers by constrained optimisation. F-DACE is the decision layer on that engine. It represents precision, propensity overlap, placebo-refutation stability, interval overlap, and directional agreement as fuzzy memberships. Hard vetoes force abstention after estimand mismatch, failed diagnostics, informative sign conflict, or weak evidence. In 180 panel simulations spanning six identification conditions, F-DACE made a decision in 67.2% of runs and limited false recommendations to 17.2%; the corresponding rates were 33.3% for the causal forest and 35.6% for backdoor regression, matching deterministic unanimity rather than dominating it. Nearly all (30 of 31) false recommendations occurred under shared unmeasured confounding, which no fusion rule can diagnose when every component shares the omitted variable. The retail application aggregates a public Walmart panel to 6,435 store-weeks across 45 stores. F-DACE abstains for all five markdown indicators: some estimates are imprecise, one refutation fails, and MarkDown5 has a direct sign conflict. A LangGraph conversational agent exposes impact, what-if, and lever-optimization tools while a deterministic verifier preserves causal-layer status. On 24 live questions it achieved 100.0% tool-routing accuracy, 100.0% status fidelity, and 0.983 mean groundedness. On ten adversarial questions it resisted all injected instructions.
cs.LG / 47 / 2609.18281
A GAN-Based Framework for Robust DDoS Attack Detection
Abstract
The availability and consistency of online services remain vulnerable due to Distributed Denial of Service (DDoS) attacks. These attacks are evolving by adopting more complex strategies to evade traditional network security systems. Despite the effectiveness of machine learning models in detecting DDoS traffic, targeted adversarial attacks can degrade their classification accuracy. This work proposes a robust detection framework that integrates generative adversarial modelling with advanced machine learning models. We trained Random Forests, Deep Neural Ensembles, and Transformer-based models using the CICDDoS2019 dataset to establish the frameworks baseline performance. To enhance the models defensive capacity, we generated synthetic adversarial flows that simulate potential evasion attempts and adversarial traffic using a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP). Then, we combined the generated traffic with benign and malicious traffic to construct hybrid datasets to train the models to learn more generalizable decision boundaries. The experimental results indicate that the proposed methodology significantly enhances detection accuracy and resilience, especially against unseen adversarial traffic. We also tested the designed framework using real-world generated traffic, which demonstrates its capability in practical settings. The scalable and efficient solution against adversarial DDoS attacks, introduced in this work, paves the way towards more resilient and adaptive network defense systems that combine generative adversarial augmentation with recent advances in learning models.
cs.LG / 48 / 2609.18314
Beyond Quadratic Loss: The Stability Phase Diagram of Adam
Abstract
Loss spikes are recurrent instabilities in neural-network training and can arise from multiple mechanisms. For Adam in particular, macroscopic loss spikes have been linked to optimizer dynamics, yet how its two momentum timescales govern them remains unclear. We investigate this dependence by mapping training dynamics across the $(β_1,β_2)$ plane. Across a range of model--task settings, an approximately linear boundary, $1-β_2=C(1-β_1)$, separates spiky from non-spiky dynamics, whereas a one-dimensional quadratic loss produces approximately cubic slope. A one-dimensional superquadratic loss $L(x)\propto|x|^n$ recovers the near-linear scaling and links the boundary coefficient to the effective loss exponent $n$. We further show that confident cross-entropy losses develop a core--wall landscape comprising a narrow quadratic core followed by a steep wall, which produces effective superquadratic behavior at the scale of an optimizer update. Together, these results connect Adam loss spikes to both the mismatch between momentum timescales and finite-scale superquadratic loss geometry beyond the Hessian.
cs.LG / 49 / 2609.18315
Multi-Appliance Non-Intrusive Load Monitoring via Label-Preserving Aggregate Recomposition and Prediction Consistency
Abstract
Non-intrusive load monitoring (NILM) estimates appliance power sequences from aggregate power, but models trained on source households commonly lose accuracy in unseen households. Aggregate power also contains loads from other appliances and measurement error, so predictions may depend on the residual background that co-occurs with source-household targets. Time-aligned submetered measurements and the additive decomposition of aggregate power expose a relation unused by window-wise supervision: an aggregate window can be recomposed by replacing only its residual background while preserving all modeled target-appliance power sequences pointwise. We combine label-preserving aggregate recomposition with prediction consistency. Both windows receive complete power and operating-state supervision. For each appliance, disagreement between the two power predictions is penalized only when both satisfy a fixed reliability criterion and only to the extent that it exceeds a fixed margin. The proposed method is implemented using a multi-appliance architecture with two-stage shared-to-specific mixture-of-experts routing. On REDD, UK-DALE, and REFIT, the proposed method lowers appliance-averaged mean absolute error relative to single-window training from 14.75 to 13.14 W, from 8.88 to 8.51 W, and from 15.83 to 14.55 W. Label-preserving aggregate recomposition and prediction consistency are used only during training, and add no inference-time module or parameter.
cs.LG / 50 / 2609.18321
Trajectory Learnability for Offline On-Policy Distillation with Imperfect Teachers
Abstract
Offline on-policy distillation gains efficiency by collecting student trajectories and teacher supervision once and reusing them throughout optimization. The same reuse makes imperfect supervision persistent. Since even strong teachers can fail, we ask \emph{what remains learnable from imperfect teacher supervision?} Teacher failure is only a coarse problem-level signal and does not imply that all supervision along the associated student trajectory is unhelpful. A natural alternative is to estimate teacher recoverability along the trajectory, but repeated continuations largely erase the efficiency advantage of offline distillation. We instead use teacher-successful problems to define a cheap reference for what the student can learn. We train on teacher-successful problems and measure how the likelihood of each observed token in trajectories from teacher-failed problems changes. We use these signed likelihood changes as an operational \emph{learnability signal}: larger increases indicate behavior more strongly promoted by successful-only learning. We aggregate this signal into trajectory-level weights for the original distillation loss. Unlike continuation-based estimates, our learnability requires no additional generation and can be computed once from stored trajectories and model checkpoints. Across mathematical reasoning and code generation, our method improves an offline OPD baseline by up to 2.7 percentage points and matches or outperforms online OPD variants on multiple benchmarks. Despite the additional successful-only distillation stage, it uses 2 GPUs and about 22 GPU hours, compared with 3 GPUs and 36--48 GPU hours for representative online OPD methods.
cs.LG / 51 / 2609.18381
Every Fixed Metric Has a Blind Spot: A Learned Atmospheric Critic for Scoring Forecast Realism
Abstract
Despite their high accuracy on point-wise metrics, machine learning weather forecasting models can exhibit different failure modes such as blurring, periodic irregularities, and other unphysical spatial artifacts. This has motivated a variety of metrics to detect known failure cases. Existing metrics fix a representation or transformation in advance, and that choice limits the artifacts they can detect. We propose to train a discriminator for separating reference data from the model's output, and using its output logit to obtain a divergence-like realism score. The discriminator learns whatever separates the model's fields from real weather, adapting to whichever failure mode that model exhibits. We compare our learned atmospheric critic to existing metrics using various synthetic corruptions applied to ERA5 reanalysis data. Our method successfully identifies the corruptions and ranks their severity, while existing metrics fail on at least one corruption. Additionally, we evaluate forecasts from real weather models, and find that the realism score degrades with longer lead times and the metric generally assigns higher realism to numerical models than to machine learning models.
cs.LG / 52 / 2609.18396
Reliable Virtual Sensing: A Multi-Domain Benchmark for Robustness Under Sensor Failures
Abstract
Virtual sensing, the estimation of hard-to-measure quantities from available sensor measurements, is a critical enabler for control and monitoring in cyber-physical systems. However, when sensors fail, learning-based predictors can produce physically implausible estimates that propagate to system-level failures. We argue that real-world deployment demands robustness and introduce MuViS-C, the first multi-domain benchmark of robustness against common sensor failures in learning-based virtual sensing. Building on an existing nominal-performance benchmark and established corruption taxonomies, it covers ten sensor failure modes, from subtle drifts to catastrophic signal dropouts, at multiple severities. These are paired with complementary robustness measures capturing average error under corruption, relative degradation, and worst-case fragility. Across nine datasets from six domains, we benchmark six architectures spanning gradient-boosted trees and the major inductive biases for sequence modeling: convolution, recurrence, attention, and MLP-mixing. On the attention-based architecture, we further probe three robustification strategies. We find that (i) every model degrades substantially under corruption, becoming worse than a naïve predictor on at least one corruption setting, (ii) gradient-boosted tree ensembles achieve strong robustness, and (iii) dedicated robustification closes the gap between the attention-based architecture and the most robust models, though each strategy hurts nominal performance. The benchmark's multi-domain design proves essential, as model rankings shift across datasets, and no single domain captures the full robustness picture. MuViS-C is open-source and extensible to new datasets, failure modes, measures, and models.
cs.LG / 53 / 2609.18407
TERN: A Delta-rule Memory with a Seasonal Reference and Online Adaptation for Epidemic Forecasting
Abstract
Weekly influenza surveillance counts guide vaccine distribution and public-health alerts, yet they are hard to forecast. Each region offers only a few seasons, waves shift in timing and height every year, and information that helps while a wave grows misleads after its peak, whereas last season's shape stays informative for a year. Existing epidemic graph models and general forecasters read a short fixed window and treat all past information alike, so they neither exploit earlier seasons nor discard stale associations when the epidemic phase changes. To address these limitations, we propose TERN, a forecaster built around a delta-rule fast-weight memory that decays channel-wise and erases along a learned address under gates driven by local epidemic-phase features, combined with an explicit seasonal reference and online adaptation. On three Cola-GNN influenza benchmarks, TERN outperformed epidemic graph models and general forecasters, matched or exceeded seasonal references, and a controlled comparison confirmed the contribution of the memory itself.
cs.LG / 54 / 2609.18527
Provable Guarantees for Spectral Structured Prediction
Abstract
Structured prediction is the simultaneous prediction of multiple labels, and is widely used in various fields, such as natural language processing and computer vision. In this paper, we study binary node label recovery on signed graphs with edge-flip noise, a model introduced by (Globerson et al., 2015), via a simple spectral method that decodes node labels from the signs of the principal eigenvector of the noisy signed adjacency matrix. We develop graph structure-agnostic theoretical guarantees for approximate inference of node labels as well as guarantees for maximum angle deviation with respect to the ground truth node labels. By leveraging tools from matrix concentration theory and eigenvector perturbation analysis, we derive new concentration inequalities that explicitly quantify the effect of the spectral gap of the adjacency matrix, number of nodes, degree distribution, and noise level. As a corollary, we relate our general results to the Cheeger constant and provide results for different classes of graphs. We perform several synthetic experiments to validate our theory. To the best of our knowledge, we are the first to provide theoretical guarantees for the spectral-based approach. As a byproduct of our analysis, we derive technical results that might be of independent interest and useful for other machine learning problems.
cs.LG / 55 / 2609.18535
Provable Guarantees and Efficient Learning of Structural Equation Models with Latent Confounders
Abstract
Causal discovery aims to recover causal relationships from observed data. In various fields, exploring causal relationships among variables remains an important topic, but this task becomes challenging due to the existence of latent confounders. Ignoring such confounders can lead to false associations and incorrect edge directions. In this paper, we study the linear structural equation model with latent confounders. We propose an algorithm that iteratively identifies terminal (observed) nodes and reconstructs the directed acyclic graph of the observed variables. To do this, we recover the precision matrix of the observed variables as a sparse plus low-rank matrix: a sparse matrix captures the conditional dependencies among observed variables, while a low-rank matrix captures the combined influence of a few latent confounders. We establish that for $p$ observed variables, $r$ latent confounders and $s$ edges, our procedure correctly identifies the directed causal relationship among observed variables, for $n \gtrsim \max\{s\log p,\ r p\}$ samples. Experimental results validate our theoretical contributions.
cs.LG / 56 / 2609.18545
Revisiting the Objective of Echo Chamber Detection
Abstract
In this paper, we study the detection of an echo chamber in a social network, i.e., the identification of a set of nodes that agree on a topic, while disagreeing with the rest of nodes. We argue that this problem is different from other social network analysis problems such as community detection, and from other graph problems such as maximum graph cut and maximum clique. To the best of our knowledge, we are the first to formalize the objective function of echo chamber detection, by using the theory of Fourier transforms of set functions (Stobbe and Krause, 2012). We propose scalable semidefinite relaxation, solved via an interior point method and sparse linear algebra. Experimentally, our algorithm recovers the ground truth echo chamber better than competing methods on small synthetic experiments. Our algorithm produces echo chambers with better network properties than competing methods on large real-world datasets. To independently validate our proposed objective function, we show that our algorithm finds echo chambers with more agreements with suspended users than competing methods on a small real-world dataset.
cs.LG / 57 / 2609.18555
Interpretable Patch-Based Deep Learning for Wildfire Spread Prediction from Ensemble Simulations
Abstract
Wildfire spread is traditionally predicted using physics-based simulators, which are physically interpretable but whose cost increases with each additional ensemble member. We ask how well deep learning surrogates can reproduce these simulations at a fraction of this cost, training them on 10,584 fire spread simulations at 2m resolution for the Rectoret region in Catalonia, Spain. Four architectures are compared: a patch-based U-Net, a transfer-learned ResNet-50, a physics-informed network constrained by the wind-driven advection equation and a Swin-Unet transformer. Among the terrain and vegetation variables, only surface fuel load predicts burn probability with any strength (r = 0.27) and including it lowers prediction error by 21%. The remaining variables correlate weakly and are highly duplicative. Next, an experiment with saliency, occlusion and rotation demonstrates the models' learning. Convolutional models rely primarily on distance from the current fire front, while Swin-Unet assigns more weight to fuel and terrain, a finding also noted in an unrelated wildfire dataset. When applied without retraining to the second region, Pedriza, all three convolutional models still predict fire spread, losing accuracy by a small but systematic margin.
cs.LG / 58 / 2609.18577
Accurate Trace Estimation with Fewer Random Bits via Recursive TensorSketch
Abstract
We consider the problem of estimating the trace of an implicit matrix $\mathbf{A} \in \mathbb{R}^{d^p\times d^p}$ that can only be accessed through matrix-vector products queries. The \textit{Hutchinson trace estimator}% ~\cite{Girard1987algorithme, article-hutchinson} is a classical sketching method for this problem. Their estimator, $H_{m}(\mathbf{A}) = \frac{1}{m} \sum_{i=1}^{m} {\mathbf{z}^{(i)}}^T \mathbf{A} \mathbf{z}^{(i)}, \quad \text{where } \ {\mathbf{z}^{(i)}}\in \mathbb{R}^{d^p}$, and $z^{(i)}_j \in {N}(0, 1), j\in [d^p]$, satisfies the following guarantees: (i) $\mathbb{E}[H_{m}(\mathbf{A})]=\operatorname{tr}(\mathbf{A})$, and (ii) $\mathrm{Var}[H_{m}(\mathbf{A})]=\frac{2}{m}||\mathbf{A}||_F^2$. Generating one query vector $\mathbf{z}^{(i)}$ requires $O(d^p)$ random bits; thus, $m$ queries require $O(md^p)$ random bits, which can be prohibitive in large-scale applications. Recent work by Meyer et al.~\cite{meyer2025hutchinsonsestimatorbadkroneckertraceestimation} proposes a variant of the Hutchinson trace estimator in which each query vector in $\mathbb{R}^{d^p}$ is constructed as the Kronecker product of $p$ random vectors in $\mathbb{R}^d$, requiring $O(mpd)$ random bits for $m$ query vectors. The estimator of~\cite{meyer2025hutchinsonsestimatorbadkroneckertraceestimation} is unbiased; however, its variance grows exponentially with $p$. In this work, we address this limitation by proposing a sketching-based estimator that requires $O\!\big(p (d + m)\log m\big)$ random bits, yields an unbiased estimate of the trace, and simultaneously achieves a variance bound that grows polynomially with $p$.
cs.LG / 59 / 2609.18587
Label-free steering: Compressing test-time reinforcement learning into bias-only subspaces
Abstract
Test-time reinforcement learning (TTRL) enables models to improve their reasoning without relying on labeled training data, but existing approaches typically optimize a large fraction of the model parameters. This raises a natural question: can effective test-time adaptation emerge when both the reward signal and the optimization space are severely restricted? We answer this question with label-free bias-only TTRL, which uses majority-vote pseudo-labels as rewards and optimizes only approximately 100K bias parameters while keeping the pretrained backbone frozen. On MATH-500, our approach reaches 76.67% accuracy, slightly exceeding our own labeled bias-steering reproduction while optimizing 76,000x fewer parameters than full-parameter TTRL. The same training procedure improves performance across vision-language and audio reasoning tasks, including MathVista, AI2D, LogicVista, and MMAU. We further show that the learned steering vectors transfer to 4,500 held-out MATH problems, indicating that the adaptation is not limited to the problems used during test-time optimization. Finally, we analyze why this highly restricted adaptation can work, showing that majority-vote reliability improves with rollout consensus and that bias subspaces with greater accessible gradient energy exhibit stronger downstream trainability. These results demonstrate that substantial test-time adaptation can emerge from optimizing a tiny bias-only subspace using entirely label-free rewards.
cs.LG / 60 / 2609.18588
Peak-Aware Short-Term Load Forecasting Across Distribution Grid Aggregation Levels
Abstract
For distribution system operators, short-term load forecasting (STLF) supports congestion management, voltage control, and asset protection. Most existing approaches focus on overall accuracy across all time steps and neglect performance during high-demand (HD) periods, where larger forecast errors can increase the risk of congestion and voltage violations. In this paper, we study peak-aware STLF across three operator-relevant distribution grid aggregation levels, area codes (AC), secondary substations (SUB), and low-voltage (LV) feeders, using open datasets from the United Kingdom and Switzerland. We compare statistical baselines, machine learning models (LightGBM and XGBoost), and recent time-series foundation models (Chronos Bolt and Chronos-2) under a peak-aware evaluation framework that reports both overall and HD forecasting performance using NMAE and MAPE. The results show that Chronos-2 achieves the best HD performance across all aggregation levels, with HD-NMAE and HD-MAPE of 0.039 and 4.53% at AC, 0.080 and 9.45% at SUB, and 0.138 and 16.14% at LV, while Chronos-Bolt consistently ranks second best. Compared with the gradient boosted ML models, Chronos-2 reduces mean HD-NMAE by about 20-51% across levels while remaining best or near-best on the overall metrics. A quantile analysis of the probabilistic Chronos outputs further identifies aggregation-specific operating points, and runtime measurements indicate that foundation model inference is fast enough for practical deployment. Overall, the findings highlight peak-aware evaluation and aggregation specific quantile selection as a practical pathway toward more operationally relevant STLF in distribution networks.
cs.LG / 61 / 2609.18595
ReDIL-GNN: Resynthesis Domain Incremental Learning for Circuit Graph Neural Networks
Abstract
Logic resynthesis preserves circuit functionality while changing gate vocabulary, topology, and structural statistics, creating domain shift for circuit graph neural networks (GNNs) without changing task labels. To study this setting, we introduce ReDIL-GNN, a resynthesis domain-incremental learning framework that adapts a fixed prediction or representation head as new synthesis styles arrive and evaluates retention on all previously observed domains. Because not every shift should be adapted blindly, ReDIL-GNN further introduces the Resynthesis Adaptability Index (RAI), a pre-adaptation score that combines adaptation need, source-equivalence recoverability, structural coverage, and update compatibility. We evaluate supervised hardware-security tasks and representation-learning models using task-native metrics for classifiers and source-equivalence retrieval metrics for embedding models, comparing naive fine-tuning with LwF, Online EWC, MAS, ER, A-GEM, DER++, ER+LwF, and equivalence-guided replay. Across the studied pipelines, RAI separates unsupported shifts from promising updates, ranging from 0.001 for a structurally uncovered GNN-RE ABC-rewrite shift to 0.824 for the best original-only GNN-RE adaptation case. In practice, ReDIL-GNN turns resynthesis-aware circuit learning into a deployment control loop: RAI screens each new synthesis flow before update, guiding whether to reuse the current model, apply retention-aware adaptation, or defer adaptation until the shift is better supported.
cs.LG / 62 / 2609.18599
Online Robust Reinforcement Learning Through Monte-Carlo Planning
Abstract
Monte Carlo Tree Search (MCTS) is a powerful framework for solving complex decision-making problems, yet it often relies on the assumption that the simulator and the real-world dynamics are identical. Although this assumption helps achieve the success of MCTS in games like Chess, Go, and Shogi, the real-world scenarios incur ambiguity due to their modeling mismatches in low-fidelity simulators. In this work, we present a new robust variant of MCTS that mitigates dynamical model ambiguities. Our algorithm addresses transition dynamics and reward distribution ambiguities to bridge the gap between simulation-based planning and real-world deployment. We incorporate a robust power mean backup operator and carefully designed exploration bonuses to ensure finite-sample convergence at every node in the search tree. We show that our algorithm achieves a convergence rate of $\mathcal{O}(n^{-1/2})$ for the value estimation at the root node, comparable to that of standard MCTS. Finally, we provide empirical evidence that our method achieves robust performance in planning problems even under significant ambiguity in the underlying reward distribution and transition dynamics.
cs.LG / 63 / 2609.18610
A Geometric Theory of Decision Boundaries in Structured Markov Decision Processes
Abstract
Classical dynamic programming represents optimal sequential decisions through value functions and policies. While this functional representation is natural for computing optimal decisions, it does not directly identify the mathematical object governing policy reconstruction, representation complexity, or oracle-query complexity once an optimal policy is fixed. This paper addresses this question by developing a geometric theory of structured optimal policies in which the decision-boundary geometry induced by the policy becomes the primary object of analysis. We show that, under suitable structural regularity conditions, this geometry provides the minimal representation required for policy reconstruction and determines the statistical and computational complexity of the reconstruction problem. Building upon this representation, we establish structural properties of policy-induced decision geometry, introduce intrinsic notions of boundary and decision complexity, derive information-theoretic measures of decision compression, and obtain statistical guarantees for boundary estimation and policy reconstruction from black-box policy queries. Collectively, these results demonstrate that, for the structured decision problems considered here, the complexity of policy reconstruction is governed by the geometry of the decision boundary rather than by the cardinality of the ambient state space. Controlled numerical experiments examine the principal theoretical predictions and provide empirical evidence consistent with the proposed framework.
cs.LG / 64 / 2609.18622
How Many Labels Does Model Choice Need? Certificates and Budgets for Selective Prediction
Abstract
Classifiers can make identical predictions yet require labels to compare their selective performance: confidence ranks weight the same errors differently. We quantify this requirement for the area under the generalized risk-coverage curve (AUGRC). A prelabel lower bound rules out insufficient budgets. With all labels known, a covering linear program bounds the minimum number of labels sufficient to fix the winner (the certificate size) within $K-1$ labels for $K$ candidates. For fixed $K$, independent uniform orders and identical predictions, the prelabel bound approaches one quarter of the pool. With iid Bernoulli errors independent of the orders, every exact acquisition policy reads almost all labels asymptotically, although a two-candidate certificate needs only half. Across 108 feature-panel comparisons on nine datasets, disagreement labels settle every accuracy choice but no AUGRC choice. A 20% budget is ruled out in 96 conditions; certificates need 56-57% on average. On ten conditions with pretrained image classifiers, confidence-score choice reads 68-91% of 10,000 labels for exact selection and 50-67% with AUGRC tolerance $5\times10^{-4}$. An exact stopping test works with any acquisition order. Together, these results link confidence ranks to label budgets and certified model comparison.
cs.LG / 65 / 2609.18639
CoRe-MARL: Cooperative Redistribution Under Unknown Dynamics Using Recurrent Multi-Agent Reinforcement Learning
Abstract
Emergency management assistance programs, such as relief distribution, are essential for delivering necessary supplies to affected communities. However, these programs operate in a decentralized network of local centers that face uncertain local demand and supply dynamics, resulting in inconsistent avail- ability of local services. Redistribution of supplies among these local centers reduces these imbalances, but the centers often make decisions independently, with limited information and disrupted transportation. This study develops CoRe-MARL, a cooperative multi-agent reinforcement learning (MARL) framework, by formulating a decentralized partially observable Markov decision process (Dec-POMDP). We treat each center as an agent that learns a redistribution policy to improve the service in the worst-case region and reduce the service gap across regions while protecting network-wide service. We incorporate a recurrent network that captures evolving supply and demand dynamics without direct observation, while multi-agent proximal policy optimization (MAPPO) enables centralized training and decentralized execution (CTDE). We evaluate the framework in a simulated environment with diverse trajectories, where exact dynamics are not observed by actors and the MAPPO critic. We compare the recurrent MAPPO with the recurrent independent PPO (IPPO) and a local only heuristic, and find that MAPPO reduces the service gap across local centers and enhances service for the worst-served center while maintaining competitive network-wide service. The recurrent MAPPO also shows consistent performance across diverse trajectory patterns, demonstrating its ability to adapt to evolving dynamics. The findings demonstrate the capability of cooperative learning for decentralized redistribution and improving equitable service under uncertain and evolving dynamics.
cs.LG / 66 / 2609.18655
Learning to Program Adaptive Non-Local Observables for Machine Learning
Abstract
Quantum neural networks (QNNs) are typically built from variational quantum circuits (VQCs), which are limited by local measurements. Adaptive non-local observables (ANO) address this by jointly optimizing circuit parameters and multi-qubit measurements. However, existing ANO-based VQCs learn only a single static observable that remains invariant across all inputs. We propose QFWP-ANO, a novel architecture which employs a classical hypernetwork to dynamically program VQC parameters and/or non-local observables conditioned on each input. On multivariate time-series forecasting across four ETT datasets, QFWP-ANO achieves the lowest MSE in 16 of 20 settings and second-lowest in the remaining four, surpassing ANO-based and other strong baselines. On reinforcement learning tasks, QFWP-ANO consistently surpasses ANO-VQCs. Our results establish input-conditioned ANO as an effective approach for enhancing QNNs.
cs.LG / 67 / 2609.18656
Revisiting Distributed Sign-Based Variance Reduction
Abstract
Sign-based methods reduce communication costs in distributed environments, but aggregating local signs can introduce bias when data are heterogeneous. As a result, existing sign-based variance reduction methods fail to obtain the optimal convergence rates. In this paper, we solve this problem and obtain optimal rates for both nonconvex stochastic and finite-sum optimization. We first give a counterexample showing that majority voting can fail to approach stationary points even with exact local gradients. Motivated by this limitation, we propose tracking the global gradient at the server through unbiased compression of recursive gradient increments. As a result, we can obtain the convergence rates of $O(\sqrt{d/K}+\sqrt d (a/(nK))^{1/3})$ for the $\ell_1$-norm and $O(\sqrt{a/K}+\sqrt a/(nK)^{1/3})$ for the $\ell_2$-norm. Here, $K$ is the iteration number, $n$ is the number of workers, $d$ is the dimension, and $a=1+ω$, with $ω$ denoting the compressor's relative variance. For finite-sum problems with $M$ components, we combine periodic exact gradient refreshes with compressed component-gradient differences. The resulting total sample complexities are $O(M+d\sqrt{aM}ε^{-2})$ and $O(M+a\sqrt M\ epsilon^{-2})$ for $\ell_1$ and $\ell_2$ gradient norms at most $ε$, matching the corresponding bounds in centralized settings.
cs.LG / 68 / 2609.18745
When Edit Flows are Edit Jumps: replicating Edit Flows and EvoFlows
Abstract
Antibody lead optimization calls for a small, bounded set of edits to an existing candidate: substitutions, but also insertions and deletions. Edit-based generative models are the only ones that allocate such an edit budget without fixing the edit positions, the edit count, or the output length in advance. However, the existing approaches Edit Flows and EvoFlows did not release code or complete training specifications. Here, we show that both methods follow the same underlying process -- edits firing one at a time, at learned rates, in continuous time -- the pure-jump case of generator matching over finite sequences. With EditJumps we introduce the first open implementation of this framework, with a single generalist antibody editor trained on 1.66M Observed Antibody Space homolog pairs to propose homolog-like variants of a seed sequence, editing unseen leads zero-shot, without the per-family retraining original approaches require. Replicating this system from scratch exposes why open code is essential for generative biology: reconciling published edit distributions required reverse-engineering an undocumented rate-scaling hyperparameter that dictates realized mutation counts. Moreover, we show that published evaluation metrics are highly sensitive to reference sample size, frequently flipping method rankings. We release our full codebase, automated test suite, and configurations at: https://github.com/VisiumCH/editjumps
cs.LG / 69 / 2609.18782
A Convergence Framework for Deep $V$-Learning: Error Propagation and Sharp Action-Gap Bounds
Abstract
We establish convergence bounds for deep $V$-learning with horizon $H$. The algorithm fits a scalar value function to targets from executed transitions and selects actions using a predictive model and the value function. For current observed-successor targets with fresh true-kernel outcomes, the conditional mean is $\mathcal{T}^βV$, which averages over behavior-policy actions. The Bellman optimality update is $\mathcal{T} V$. We decompose the update error into six residuals: fitting, transition reuse, target construction, replay, action selection, and exploration. Under $L^s$ concentrability, their $L^p$ norms ($p=s/(s-1)$) control expected $L^1$ policy loss. The bound explicitly weights residuals from only the last $H-1$ update blocks, plus an initialization term for shorter runs. We quantify the cost of a shared sampling distribution across horizon levels. For statistical error bounds of order $n^{-ν}$, we derive optimal continuous allocations and an integer allocation whose objective is within a factor $2^ν$ of the constrained optimum. A margin condition with exponent $α$ gives action error of order $Λ^{1+α/p}$, where $Λ$ combines network drift and score error; a one-step construction proves the exponent sharp. Bounds on the distance between frozen and optimal scores transfer an optimal-gap condition to frozen-iterate gap bounds while retaining the mass of optimal ties. Survival probabilities and coverage conditions at deployment yield bounds for policies selected with approximate scores. Separate spatial ReLU networks per horizon level give a conditional neural regression rate, and the finite-state case gives a log-free expected fit rate. These results give expected policy-loss consistency for the fixed-horizon generative-reset approximate-ERM procedure with exact action scores and provide an explicit residual-decay criterion for FIFO/interleaved SGD.
cs.LG / 70 / 2609.18812
WaveTLM: Reliable Time-Series Language Modeling through Task Compilation
Abstract
Time-series language models provide a shared natural-language interface across temporal tasks, but plausible text does not guarantee reliable task outputs. Responses may appear reasonable while hallucinating the required object: numerical sequences can violate shape, scale, channel order, or temporal alignment, and textual decisions can fall outside the legal label space. We formulate reliable time-series language modeling, separating task-object reliability from predictive quality. We introduce ExecTS-QA, a contract-grounded benchmark spanning forecasting, imputation, classification, anomaly detection, and waveform analysis. We further propose WaveTLM, a unified compiler-executor model whose task compiler transforms user requests, visible arguments, and wave-grounded evidence into typed task states, while task-native executors construct numerical tensors, legal decisions, or structured records. On ExecTS-QA, a single WaveTLM checkpoint achieves 99.40% contract-valid coverage, compared with 37.83% for the strongest evaluated string-first baseline, while retaining balanced predictive performance across all five task families. Evaluations on SciTS, TSQA, IRTS-ToolBench, and ARFBench provide additional evidence of transfer. The code, construction scripts, and ExecTS-QA dataset will be publicly released upon publication. These results show that task compilation can convert plausible language generation into reliable time-series outputs.
cs.LG / 71 / 2609.18825
Interpretable Multi-Instance Learning Enables Early Prediction of Key Molecular Alterations from Routine Flow Cytometry in Acute Myeloid Leukemia
Abstract
Background: Molecular testing for NPM1 and FLT3-ITD mutations guides critical early treatment decisions in acute myeloid leukemia (AML), but results can take weeks, long after these decisions must be made. Flow cytometry, already performed within hours of admission as part of routine care, may carry enough signal to predict these mutations directly, without added cost or delay. Methods: We developed an interpretable multi-instance learning classifier based on a decision tree, in which each patient sample is modeled as a collection of individual cells and mutation status is inferred from cell-level predictions. The model was benchmarked against a random forest trained on clinical variables and a deep convolutional neural network adapted for multitube flow cytometry data. Performance was assessed by cross-validation on a discovery cohort of 197 patients and tested on an independent cohort of 161 patients, using the area under the receiver operating characteristic curve (AUROC) and positive predictive value. Results: In cross-validation on the discovery cohort, the MIL model achieved mean AUROCs of 0.96 (SD=0.05) for NPM1 and 0.86 (SD=0.10) for FLT3-ITD, outperforming the clinical baseline and matching deep learning approaches. The model then successfully generalized to the independent test cohort of 161 patients, reaching AUROCs of 0.90 (NPM1) and 0.82 (FLT3-ITD), with positive predictive values of 0.87 and 0.68, respectively. Cell-level interpretation recovered established immunophenotypic signatures (CD33${}^{+}$ /CD34___ for NPM1-mutated cases, CD33${}^{+}$ /low side-scatter for FLT3-ITD), directly linking model predictions to known biology. Conclusions: These results show that an interpretable model applied to data already collected in routine care can predict AML molecular status within hours, offering a practical route to earlier, biology-informed treatment decisions.
cs.LG / 72 / 2609.18863
Physics-based prediction, uncertainty quantification and decision-making for IN718 crystallographic texture intensity across LPBF defocus regimes
Abstract
Reliable prediction of crystallographic texture in laser powder bed fusion is critical for linking process conditions with anisotropic response and for qualification. However, black-box models may fail under shift and cannot distinguish weak data support from loss of physical validity. This study develops a two-stage physics-based model for <001> || BD (build direction) texture in Inconel 718. Stage 1 maps process variables to melting mode and melt pool geometry. Stage 2 predicts texture by combining an empirical physics model with a random-forest residual model. A k-nearest-neighbor weight attenuates residual corrections for poorly supported queries, while a study-specific areal beam-power-density criterion withholds predictions outside the adopted conduction envelope. Conformal intervals are evaluated on the retained physics-valid set, and SHAP and Sobol analyses assess residual sensitivity. Under a controlled leave-one-defocus-out evaluation, the physics anchor achieved R^2 = 0.778, against -0.001 for the black-box model and 0.750 for the gated hybrid. Under leave-one-group-out cross-validation, the gated hybrid reached R^2 = 0.592 against 0.538 for the black-box model. Retained-set coverage was 92.9% at a mean full width of 3.65 multiples of a uniform distribution (MUD) under grouped cross-validation and 100% at a width of 3.21 MUD under transfer to a withheld +80 mm defocus regime. An illustrative mapping produced a retained BD elastic-modulus span of 127-187 GPa. On nine conditions from a separately built sample set, the framework withheld three, attenuated three, and matched the measured ordering for the rest. Separating data applicability, physics validity, and predictive uncertainty into distinct decisions lets the framework transfer where an unconstrained model does not, and withhold predictions where no model class performs adequately.
cs.LG / 73 / 2609.18916
Higher-order pruning of experts in mixture-of-experts language models
Abstract
Mixture-of-Experts (MoE) language models suffer from large parameter counts, which create a significant memory bottleneck. Expert pruning is the most direct approach for reducing this parameter count, yet existing methods make pruning decisions for each expert independently, and assume experts' contributions are purely additive. In reality, expert usage in MoEs is inherently cooperative. We derive HOPE (Higher-Order Pruning of Experts), a second-order pruning objective which provably minimizes an upper bound on the error resulting from pruning. We show that REAP (a state-of-the-art first-order pruning method) is a special case of HOPE where interaction terms are ignored. Across three frontier MoE models (up to 122B parameters), two distinct calibration sets, and multiple benchmarks (including math, instruction following, coding, and an agentic suite), we demonstrate that HOPE produces better pruning decisions than existing methods, and its advantage is most pronounced at high pruning rates and on challenging agentic workloads. At 50% pruning, HOPE outperforms all baselines and achieves an average rank of 1.58 out of 5 methods (versus 2.42 for the next-best method, REAP), with gains of up to +6.1% on agentic coding. Over all conditions, HOPE again achieves the best average rank and surpasses every other method in the majority of head-to-head comparisons. By preserving cooperative expert structure that first-order methods ignore, HOPE enables aggressive compression with minimal degradation, particularly on complex tasks where diverse expert combinations are invoked over long sequences.
cs.LG / 74 / 2609.19074
RLLBC-Lib: An Educational Code Library for Reinforcement Learning and Learning-Based Control
Abstract
Reinforcement learning (RL) is an exciting concept as well as a remarkable success story worth sharing. However, RL builds on rather complex interactions between different objects that play out over several cycles. Such dynamics are often best explained with an easily accessible implementation. We present RLLBC-Lib, a carefully crafted code library with the goal of lowering the entry barrier for students and other learners of RL in the context of learning-based control. At its heart, RLLBC-Lib comprises a comprehensive library of tabular RL approaches to enforce a clear understanding of the theoretical foundations. A deep RL library follows the same design principles, underscoring the parallels between simple tabular and state-of-the-art deep RL approaches. Additionally, RLLBC-Lib provides a collection of implementations illustrating core RL principles and contrasting RL to other learning-based control approaches. Finally, RLLBC-Lib provides an ideal basis for creating programming assignments with automated grading.
cs.LG / 75 / 2609.19076
Double descent is the principle of least action
Abstract
The test error of a model plotted against its number of parameters $d$ falls, peaks when the model can just fit the training data, and falls again, exhibiting the double descent phenomenon. We explain the phenomenon with statistical mechanics. The training trajectory of a stochastic gradient-based method is a particle wandering over the energy landscape of the training loss at an induced temperature $T$, and a run that has equilibrated visits every parameter vector of a given training loss equally often, the fundamental postulate of statistical mechanics, with probability given by the Boltzmann distribution. Because training starts at an initial point and has only finite time to diffuse, it carries an effective weight decay, which makes every parameter a quadratic degree of freedom. The equipartition theorem then distributes the energy among the $d$ degrees of freedom in shares of $T/2$, so at a fixed training loss adding parameters lowers the temperature and drives the Boltzmann distribution toward the stationary path. Finally, adding parameters can only lower the $L^2$ norm of the stationary path, so a solution sampled at fixed loss is less likely to be large with increasing $d$, effectively increasing weight regularization.
cs.LG / 76 / 2609.19077
Probabilistic Linear Explanations
Abstract
Formal explainability provides mathematically grounded justifications for individual predictions. However, abductive explanations often exceed human cognitive limits by involving too many features, while probabilistic relaxations have remained largely limited to categorical classification. We present a unified framework for probabilistic explainability based on sparse, anchored linear models, applicable to both binary classification and continuous regression. By mapping instances to the Boolean hypercube, our linear explanations strictly generalize subset-based approaches: they capture both the magnitude and direction of feature contributions while enforcing a prescribed sparsity budget $k$. We show that minimizing the relevance error for such explanations is \ClassNPPP-hard when the underlying model is a neural network, and we relate this intractable objective to a tractable surrogate---the fidelity error. For a parameterized family of local distributions, the relevance error of any $k$-sparse explanation is bounded by its fidelity error up to a multiplicative factor that remains small locally. We address the resulting empirical problem using two complementary approaches: a Mixed Integer Programming (MIP) formulation that yields provably optimal empirical solutions while maintaining polynomial sample complexity, and a polynomial-time Iterative Hard Thresholding (IHT) algorithm with provable approximation guarantees. Empirical evaluations show that, unlike state-of-the-art baselines such as LIME and MAPLE, our explanations satisfy both the anchoring and sparsity constraints by construction, while consistently achieving lower relevance error.
cs.LG / 77 / 2609.19099
Evidence-Grounded Agentic Formulation Development in an Autonomous Laboratory
Abstract
Self-emulsifying drug delivery systems (SEDDS) can improve the oral bioavailability of poorly soluble drugs, but identifying high-performing formulations remains experimentally intensive. We present Andromeda 2, an agentic system that reasons over structured in-house experimental evidence and invokes computational and experimental tools to design and execute successive formulation batches. Using a miniaturized automated laboratory at a matched budget, we benchmark it against Andromeda 1, a probabilistic optimization model deployed across dozens of live development projects, and a wet-lab design-of-experiments (DoE) campaign. For paclitaxel, Andromeda 2 achieved a 50% high-performance hit rate versus 17% for Andromeda 1 and 2% for DoE, and identified 12 formulations meeting all four target product profile (TPP) objectives versus 6 and 0, respectively. Median $AUC_{10-240}$ was 70.1, 12.0, and 3.5 mg$\cdot$min/mL, while maximum AUC was comparable between Andromeda 2 and Andromeda 1. A selected full-TPP formulation achieved an apparent effective paclitaxel loading of $19 \pm 5\%$ w/w at the first FaSSIF measurement, approximately 3.3-fold higher than the 5.7% w/w loading reported for a published paclitaxel S-SEDDS. A controlled ablation showed that access to structured in-house experimental evidence increased mean AUC by 34%.
cs.LG / 78 / 2609.19107
How Model Growth, Recursion, and Boundary Operators Influence Scaling Exponents
Abstract
Scaling laws predict how loss decreases with increases in computation. We show, contrary to conventional wisdom, that architectural interventions can modify scaling exponents in pre-training, leading to exponential improvements in performance with increases in computation. As an anchoring point, we consider the architectural formulation of looped transformers. Although not typically used in this way, looping, also known as recursive depth, provides a mechanism for model growth, by increasing the number of loops during training. Model growth, with and without shared weights, provides the biggest changes to the scaling exponents. In particular, a 7.4B model growth architecture matches GPT-3 13B on CORE with roughly $20\times$ less compute, and has compute efficiency gains that increase with scale. Moreover, simply using a boundary operator in a vanilla transformer, which normalizes and injects an earlier block, also provides increasing compute-efficiency gains, although to a lesser extent. In the data-constrained, multi-epoch setting, standard looping has a useful regularizing effect, where we find it is compute-optimal to increase the number of loops with scale. These results can be understood through the lens of computational depth: for a given computational budget, we wish to increase the usable depth of the transformer, which can lead to efficiency gains that increase with scale.
cs.LG / 79 / 2609.19135
Exponential Hardness of Off-Policy Evaluation under History-Dependent Logging
Abstract
Can a logged dataset visit every hidden state frequently and still be exponentially uninformative about a target policy's value? We show that it can when the logger depends on history. For every horizon $H \ge 3$, we construct two POMDPs with at most two latent states per stage, three actions, and a common logger with three memory states. Action coverage, belief coverage, and two behavior-marginal outcome-revealing conditions all have constants independent of $H$. Nevertheless, evaluating a known deterministic target policy to accuracy $1/8$ requires $Θ((3/2)^H \log(1/δ))$ logged episodes at confidence $1-δ$, for $0 < δ\le 1/4$, even when both candidate models are known. The mechanism is simple: a reset erases the unknown transition that determines the target value. We characterize the resulting statistical experiment exactly and obtain a matching optimal estimator. A directed two-lane gridworld realizes the construction, and trajectory simulations agree with its finite-sample prediction. The result establishes intractability for the history-dependent-logging, model-based case posed by Zhang and Jiang (2025, arXiv:2503.01134), under their behavior-marginal definition of revealing.
cs.LG / 80 / 2609.17992
The Operable Pareto Front: Distilling Offline Search into Run-Time Control for Multi-Objective UAV Edge-Computing Scheduling
Abstract
A UAV mobile edge computing (MEC) fleet trades energy against delay, and its schedules form a Pareto front; we call a scheduler operable when the fleet can be asked for any point on that front at run time. We propose PrefDT, to the best of our knowledge the first preference-conditioned Decision Transformer for the problem of joint trajectory, association and offloading scheduling. Its idea comes from language modeling: we hand the model the desired trade-off as an input, such that a single model only needs to be trained once offline to return any desired point on the curve in one rollout. The fleet's state is summarized by attention pooling with a per-user bypass, so the scheduler keeps working when user reports are lost. The energy target is a running budget decremented by what the fleet actually spends. As a result, when wind or load pushes consumption off the plan, the policy can track the difference and hold its budget. Because no corpus of preference-labeled flights exists, we design a distillation pipeline and build the corpus by ourselves. In simulation against 26 method variants, PrefDT produces the best trade-off curve of any learned method and holds its energy budget to within 0.6% when propulsion cost rises by half in mid-flight.
cs.LG / 81 / 2609.18704
Toward Composable Network Digital Twins: A Subgraph-Based Latency Prediction Study
Abstract
Modern networks must support changing topologies, configurations, and performance objectives, motivating fast and reliable performance estimation. Network digital twins (NDTs) enable what-if analysis for performance estimation in such network scenarios, however, existing machine learning-based NDT approaches often rely on entire topology representations, which are inherently monolithic and lack reusability under topological or traffic changes in the network. This paper introduces a composable NDT approach that decomposes networks into subgraphs represented by reusable unit twins that capture subgraph structure, configuration and traffic behaviours. A lightweight composer aggregates unit twin combinations to create NDTs that predict per-route end-to-end latency through an overall topology. Evaluation across controlled synthetic topologies and diverse traffic scenarios, real-world Topology Zoo topologies, and a public NDT challenge dataset demonstrates that the composable NDTs achieve high in-distribution accuracy while remaining stable under out-of-distribution scenarios. Comparison with monolithic full topology NDTs demonstrates that our composable approach achieves reusability, while achieving comparable or superior accuracy.
cs.LG / 82 / 2609.18084
Not All Layers Need Tuning: Diagnosing and Directing Adaptation in Vision-Language-Action Models
Abstract
Fine-tuning a Vision-Language-Action (VLA) model for a new deployment environment is expensive, yet most methods apply uniform-capacity adapters to every network region as if every region requires equal adjustment. This paper tests that assumption on five architecturally diverse VLAs (OpenVLA-OFT, $π_0$, SmolVLA, DTP, Octo; 93M-7B parameters). Measuring per-region adaptation cost as normalized parameter displacement under region-isolated fine-tuning reveals an adaptation spectrum in which appearance shifts concentrate cost in the vision encoder, instruction shifts in the language backbone, and novel-object shifts in the vision encoder together with the action head, across all five architectures. To exploit this structure, we introduce a pipeline that observes, diagnoses, allocates, and adapts. From ten unlabeled target observations and without fine-tuning, the diagnostic estimates per-region cost by combining reference-free gradient and Monte Carlo Dropout signals with a Centered Kernel Alignment score against a cached source reference; the allocator converts the estimates into variable-rank LoRA adapters under a parameter budget and freezes well-calibrated regions; and standard LoRA fine-tuning trains the resulting adapters. The diagnostic ranks regions within each deployment at a median Spearman of 0.91, and the allocation matches or exceeds uniform LoRA at every budget we tested on LIBERO and CALVIN. On a physical xArm-7, the pipeline matches full fine-tuning under an instruction-wording shift with 0.04% of its trainable parameters, and on five held-out scenes evaluated without retraining it leads every baseline, with 11-23 successes of 30 rollouts against 8-18 for the strongest parameter-efficient baseline at equal or larger budgets and 2-11 for full fine-tuning. These results suggest that adaptation cost in VLAs is structured enough to measure before fine-tuning begins.
cs.LG / 83 / 2609.18100
Beyond Pixel Similarity: Task-Aware Evaluation of GAN-Based Synthetic Sonar Data for Robotic Perception
Abstract
Synthetic data can reduce the cost of collecting and annotating training data for robotic perception, but generating sensor observations that preserve the characteristics relevant to downstream perception remains challenging, particularly for sonar imagery. In this work, we investigate whether conventional image-fidelity metrics adequately reflect the downstream perception performance of GAN-generated synthetic sonar data. We employ a Pix2Pix conditional generative adversarial network with four discriminator configurations characterized by different receptive fields: PixelGAN, PatchGAN-16, PatchGAN-70, and ImageGAN. The models are trained using sonar imagery from two datasets and evaluated using conventional image-fidelity metrics, including Structural Similarity Index (SSIM), Peak Signal-to-Noise Ratio (PSNR), and Mean Squared Error (MSE). To complement these pixel-level measures with task-oriented evaluation, YOLOX-S, YOLOX-L, and Faster R-CNN detectors are trained exclusively on real sonar imagery and subsequently evaluated on the GAN-generated images using identical test samples and annotations across all discriminator configurations. The results reveal a discrepancy between image-fidelity and downstream object-detection performance: the configuration achieving the best SSIM, PSNR, and MSE does not consistently yield the best detection performance. In particular, PatchGAN configurations achieve strong downstream detection results despite not achieving the highest pixel-level similarity scores. These findings suggest, for the datasets and models considered, pixel-level image-fidelity metrics alone may not consistently capture the task-relevant realism of synthetic sonar observations and motivate the use of task-aware evaluation for synthetic sensor data intended for robotic perception.
cs.LG / 84 / 2609.18167
Characterizing Replay Retention Under Dynamics Shift in Model-Based Reinforcement Learning
Abstract
Adapting to changes in robot dynamics requires learning from new data without discarding experience that may still be useful. In continual model-based reinforcement learning (RL), replay collected before a dynamics change can slow adaptation, while removing it unnecessarily reduces available training data and can be especially costly if earlier dynamics return. We study when recent transitions are preferable to the full replay history. Two quantities characterize this trade-off: change magnitude and age-staleness area under the curve (AUC), measuring how well transition age separates stale from fresh data. Forgetting stale data helps after large permanent shifts but hurts when dynamics recur and older data becomes useful again. Choosing a replay strategy therefore depends on predicting when older data will help or hurt. We test these effects across two locomotion morphologies, two model-based RL algorithms, and Real-World RL benchmark perturbations. Because ground-truth staleness labels are unavailable on deployed robots, we evaluate whether an estimator built from interaction data can still provide the quantities needed to choose a replay strategy after permanent changes. Our results show that replay retention depends on change magnitude and on how the dynamics evolve.
cs.LG / 85 / 2609.18207
Reinforcement Learning for Real-Time Vision-Language-Action Policies
Abstract
Reinforcement learning fine-tuning on top of large, pretrained Vision-Language-Action (VLA) models offers promise for highly reliable robot deployment. However, because of their scale, modern VLA models suffer from high inference latency, so the observation used to select an action is often stale by execution time, creating a distribution shift that can substantially degrade reliability and performance. Prior work has explored asynchronous policy execution to reduce the effect of latency, but these methods are mostly built on imitation learning and offer no mechanism for moving beyond the training distribution toward higher reliability. We close this gap by enabling RL fine-tuning that meets the real-time control requirements of dynamic real-world manipulation. Our approach builds on EXPO-FT, a framework for sample-efficient, reliable VLA fine-tuning with reinforcement learning, and decouples slow, expressive action generation from fast, reactive action edits: a large pretrained VLA proposes action chunks using its strong behavior prior, while a lightweight edit policy performs fast, reactive decision-making by editing actions in response to changes in state, conditioned on the latest observation. We instantiate this as Real-Time EXPO-FT, an RL framework for finetuning real-time VLA policies. On the Kinetix benchmark, Real-Time EXPO-FT enables a delayed policy to achieve the best performance among delayed and non-delayed methods in 10 out of 10 environments. On four dynamic real-world tasks, robot object passing, ball balancing, table soccer kicking, and dynamic object picking, with online robot data capped at 10 minutes, Real-Time EXPO-FT improves average policy performance from 42% to 97%, all without human intervention, demonstrating rapid, sample-efficient adaptation to challenging real-world dynamics. Website: https://pd-perry.github.io/real-time-expo-ft
cs.LG / 86 / 2609.18243
Acting in Meters: Learning Metric Interactions for Precise Robotic Manipulation
Abstract
Vision-Language-Action models and World-Action Models have advanced language-conditioned robotic manipulation, yet often leave metric relations among actions, objects, and scene geometry implicit. Human manipulation combines semantic understanding of task-relevant objects with spatial feedback that guides hand motion relative to objects and their surroundings. Inspired by this, we introduce a metric interaction framework that models object-level and scene-level interactions in physical Cartesian space at a shared metric scale. At the object level, Interaction-Centric Tokens (ICTs) explicitly represent end-effector pose trajectories relative to manipulated objects and are jointly denoised with actions, providing physically grounded interaction supervision. At the scene level, the Metric Action Interaction Field (MAIF) uses action and ICT queries to attend to metric scene point-cloud features and learns geometry-conditioned action corrections. Through two-stage adaptation, our framework improves diverse VLA and WAM baselines with a small number of additional parameters and training steps. Experiments demonstrate average success-rate gains of 0.80 and 3.59 percentage points on LIBERO and RoboTwin~2.0, respectively, alongside gains of 6.80 percentage points on real-world tasks and 7.45 percentage points on their out-of-distribution variants.
cs.LG / 87 / 2609.18359
RecMorph: Topology-Guided Spatial Recurrence for Generalized Morphology Control
Abstract
Generalized morphology control requires a single policy to transform information across limbs with different physical roles, coordinate whole-body motion, and remain efficient as body size grows. Existing communication mechanisms address these requirements only partially. We introduce RecMorph, a topology-guided spatial recurrent architecture that uses recurrent sequence computation to jointly perform cross-limb communication and representation transformation. A depth-first traversal converts the kinematic tree into a morphology-derived sequence, along which shared bidirectional transitions progressively transform limb information before action decoding. Residual preservation, RMS normalization, and input-dependent channel modulation stabilize this repeated spatial transformation, yielding linear token complexity at fixed model width and depth. Across five UNIMAL tasks, RecMorph achieves the strongest mean final training performance among the evaluated generalized morphology controllers and the highest measured inference throughput on FT, while generalizing to unseen variations and bodies with up to 30 limbs. We further migrate representative generalized controllers from UNIMAL benchmarks to a four-platform quadruped setting. RecMorph achieves the best macro-averaged performance under nominal and high friction, reduces nominal velocity RMSE by 43.5% relative to specialist MLPs, and one shared policy completes 40 physical Go1/Go2 trials without falls. These results show that topology-guided recurrent transformation provides an effective and efficient communication mechanism for Generalized Morphology Control and remains effective when transferred from procedural bodies to physical robot platforms. Code and experimental resources are publicly available at https://github.com/quanruirao/RecMorph.
cs.LG / 88 / 2609.18514
ActiveScale: Scaling Active Perception for Robots across Model, Data, and Hardware
Abstract
Active perception is essential for robotic manipulation when fixed viewpoints leave task-relevant information occluded or unobserved. However, enabling vision-language-action (VLA) models to reason across changing viewpoints and actively acquire informative observations remains challenging. We present ActiveScale, a framework that advances active perception through coordinated model, data, and hardware designs. Our model augments a VLA with historical video observations and explicit camera-pose supervision, using per-frame pose tokens and a lightweight prediction head to associate observations across viewpoints and support a coherent understanding of the scene. To learn from the camera motion naturally present in human activity, we introduce a scalable human--robot mid-training recipe using 1000 hours of egocentric and robotic data, adapting the model to temporal inputs and pose supervision. We further introduce Active-perception Mobile-manipulation Platform (AMP), a robotic platform that supports active perception and mobile manipulation through single-operator teleoperation, enabling scalable collection of demonstrations that coordinate viewpoint changes and manipulation. Experiments demonstrate improved success rates on active-perception tasks, while ablation studies validate the contributions of camera-pose-aware modeling and egocentric mid-training. Together, these components provide an integrated foundation for studying and developing active perception in robotic manipulation.
cs.LG / 89 / 2609.18663
VLA-ULAP: Interleaving Cloud VLA Calls with Ultra-Lightweight Local Action Prediction at the Edge
Abstract
Billion-parameter vision--language--action (VLA) policies demand substantial onboard power, while communication delays in remote inference hinder timely responses. We propose VLA-ULAP, which interleaves remote VLA calls with an Ultra-Lightweight Local Action Predictor (ULAP). With approximately 7.4M parameters including the frozen vision encoder, ULAP combines current views, proprioception, and executed action history to predict chunks in one pass. Trained independently, it requires no VLA hidden states, online verification, or server round trips. On Jetson Orin Nano, ULAP takes 19.9 ms and 0.183 J per inference, compared with 284.3 ms and 50.55 J for GR00T on RTX A6000. Across three simulated base-policy/benchmark pairs, selected operating points remove 48.8--76.7\% of VLA calls while retaining 95.0--97.5\% of the baseline success rate. Against local VLA-acceleration alternatives on VLA-JEPA, ULAP uses an estimated 49.2\% less inference time and 51.0\% less GPU energy per successful episode than ACT at comparable success rates, and 77.1\% less time and 79.9\% less energy than SP-VLA at equal success rates. Physical SO-101 experiments retain 95.2--100\% of the baseline success rate across seen and held-out placements while reducing inference time by an estimated 47.9--58.0\% and inference-device energy by 52.1--62.5\%, based on successful-episode call counts and measured device costs. Faster responses also improve dynamic-task success rates: in latency-aware LIBERO-Safety simulation, VLA-ULAP exceeds $π_{0.5}$ by 11.0 and 15.5 percentage points on two tasks while approximately halving VLA calls.
cs.LG / 90 / 2609.17884
The Unbearable Weight: Scaling Models and Methods for UAV Audio Classification
Abstract
As unmanned aerial vehicles (UAVs) become increasingly prevalent in consumer and defense settings, classifying them reliably from limited, modality-specific data is an urgent challenge. The dominant approach, large pretrained networks fully fine-tuned on task data, carries a substantial computational and memory weight that is hard to bear in resource-constrained UAV deployments, where edge inference and rapid retraining for emerging platforms are both required. This paper systematically scales across both model architectures and fine-tuning methods for UAV audio classification, asking when that weight is justified and when lighter alternatives prevail. Using a custom dataset of 3,100 audio clips spanning 31 drone classes, we evaluate transformer (ViT, AST) and convolutional (custom CNN, ResNet-18/152, MobileNet-V3-S/L, EfficientNet-B0/B7) backbones under full fine-tuning, classifier-only fine-tuning, and four parameter-efficient fine-tuning (PEFT) methods: SSF, IA3, OFT, and selective batch-norm tuning. All configurations are evaluated with 5-fold cross-validation across accuracy, training time, trainable-parameter share, and inference-time memory footprint. Selective batch-norm fine-tuning of EfficientNet-B7 with three-fold augmentations achieves the highest validation accuracy (97.65% +- 0.30) while updating under 0.5% of model parameters. Across the sweep, lightweight CNNs consistently outperform transformers on both accuracy and efficiency. For UAV audio classification under data scarcity, scaling the method outperforms scaling the model.
cs.LG / 91 / 2609.18585
TTM-Bench: A Framework for Text-to-Music System Performance Benchmarking
Abstract
Text-to-music (TTM) systems are increasingly used to generate musical audio from natural-language descriptions. Robust evaluation is therefore essential, yet reliable performance comparison remains challenging. This difficulty stems from differences in system architecture, supported conditioning information, and access mode, as well as heterogeneous and fragmented metrics that cannot be applied uniformly across systems. To address these challenges, we introduce TTM-Bench, a framework that defines a common protocol for systematic, reproducible performance benchmarking of contemporary TTM systems. It evaluates performance along two dimensions: musical-content alignment, quantified by interpretable semantic, genre, and musical-descriptor agreement scores against a common musical specification and summarized by an aggregate score; and computational efficiency, characterized by generation latency and real-time factor, alongside resource use for local models and cost for hosted services. We demonstrate the framework through a preliminary comparative case study, illustrating the complementary evidence captured by these dimensions. The results show that higher musical-content alignment does not systematically coincide with lower computational demands, highlighting the importance of assessing TTM performance through distinct, interpretable measures rather than a reductive overall indicator.
cs.LG / 92 / 2609.18173
Beyond Direct Sensing: Harnessing Indirect Observations from Third-Party Sensors in Vehicle Tracking
Abstract
Vehicle tracking is fundamental to applications ranging from urban mobility and public safety to security and defense. Conventional tracking relies on direct access to sensors that provide strong observations such as vehicle identity and location. In practice, however, factors such as ownership, privacy, cost, and operational constraints may limit directly accessible sensors, leaving sparse observations and long tracking gaps. Meanwhile, many additional third-party sensing assets may be present across the environment but remain inaccessible at the raw-data level, preventing their direct integration into the tracking system. In this work, we investigate whether weak, indirect observations with uncertain spatial and temporal cues can complement sparse direct sensing for vehicle tracking. Specifically, we propose GrayTrack, which fuses weak anonymous events with sparse direct observations using a road-constrained particle filter. We build a CARLA-Mininet-WiFi pipeline to evaluate the system under controlled conditions, generating direct observations from accessible cameras and indirect observations from third-party cameras. Our learning-based detector achieves an F1 score of 0.989 for anonymous vehicle passages. Further, incorporating indirect third-party observations reduces trajectory RMSE by 60.1% and catastrophic track loss from 35.8% to 0.3%. These results demonstrate that GrayTrack can effectively exploit weak indirect observations to extend tracking capabilities.
cs.LG / 93 / 2609.18616
Learning Array Signal Topologies as Conditional Neural Manifolds
Abstract
Subspace methods such as multiple signal classification (MUSIC) achieve super-resolution direction of arrival (DoA) estimation by exploiting the orthogonality between the array manifold and the noise subspace of the measurements. Their accuracy therefore depends on the assumed manifold and degrades under model mismatch, while parameters not identifiable from the spatial manifold cannot be recovered. In this work, we propose the conditional neural manifold (CNM), which replaces the fixed manifold with an observation-conditioned mapping from source parameters to steering vectors. An encoder maps the snapshots to a latent scene representation that conditions a zero-initialized neural field over the parameter space. The manifold is learned without steering-vector supervision by shaping the resulting MUSIC landscape. Since the correction acts on the manifold rather than on the estimator, it can be used by other manifold-based methods without modification. The CNM restores resolution under array imperfections, colored noise, correlated sources, and near-field propagation, and resolves the angle-frequency ambiguity inherent to the nominal spatial manifold.
cs.LG / 94 / 2609.17808
Synthetic Electric Vehicle Charging Session Generation Using a Conditional Variational Autoencoder
Abstract
The increasing adoption of electric vehicles (EVs) is expected to place significant additional demand on residential distribution networks, creating a need for realistic charging datasets for planning and simulation studies. However, access to real-world EV charging data is often limited due to privacy constraints, incomplete records, and restricted availability. This paper proposes a conditional variational autoencoder (CVAE) for the generation of synthetic EV charging sessions from real transaction-level charging data. The model is trained on engineered session features describing plug-in duration, charging duration, delivered energy, charging delay, and cyclical time-of-week, while conditioning on day of week and managed charging status. A Gaussian negative log-likelihood (NLL) reconstruction loss is employed to model feature-wise heteroscedastic uncertainty, and the latent space is regularised using a Kullback-Leibler (KL) divergence term. The statistical fidelity of the generated data is evaluated using distributional metrics and downstream task performance through the Train-on-Synthetic-Test-on-Real (TSTR) protocol. Results demonstrate that the proposed approach produces synthetic EV charging sessions that preserve key statistical properties of the original dataset while supporting predictive modelling tasks.
cs.LG / 95 / 2609.17964
Mixed-Integer Nonlinear Differentiable Predictive Control for Underground Pumped Hydro Energy Storage Systems
Abstract
This paper extends Mixed-Integer Differentiable Predictive Control (MI-DPC) to multi-modal discrete decisions and nonconvex polynomial dynamics arising in Underground Pumped Hydro Energy Storage Systems (UPHES). A neural policy mapping problem parameters to continuous setpoints and integer mode selections via a Gumbel-Softmax layer is trained in a self-supervised manner by differentiating the expectation of the finite horizon control objective through the nonlinear dynamics model. Three methodological contributions enable this extension: a parallel differentiable simulator that preserves gradient magnitude, a Transformer encoder that captures long-range temporal dependencies, and a Gumbel-Softmax temperature annealing schedule that regularizes the combinatorial search. We demonstrate the framework on day-ahead scheduling of a UPHES, a large-scale mixed-integer optimal control problem with nonlinear unit performance curves and volume-head coupling. MI-DPC achieves only 1.6% suboptimality relative to a piecewise mixed-integer quadratic programming baseline, while providing five orders of magnitude speedup in online scheduling time.
cs.LG / 96 / 2609.17726
Self-Supervised Learning for Robust Resonance Mass Regression in Cascade Decays
Abstract
Reconstructing the mass of a heavy resonance from its decay products with missing energy is one of the central tasks that directly determine the sensitivity in new physics searches at collider experiments. Supervised learning approaches to this problem often struggle to generalize well due to the presence of various systematic uncertainties and distribution shifts. Exhausting all possible variations in the labeled data can be very compute-intensive, while a failure of the model to generalize can corrupt the reconstructed resonance widths that are critical in peak-hunting analyses. In this work, following the foundation model paradigm, we use a self-supervised approach to pre-train a transformer encoder with VICReg to learn an embedding invariant to various corruptions, then fine-tune it for mass regression on a heavy resonance with masses ranging from 2.5 to 6.5 TeV and a SUSY-like cascade decay into an eleven-body final state. We show that the pre-trained model reconstructs sharper resonance peaks and has a more stable performance under various realistic corruptions, compared to a supervised model of the same architecture trained on the same augmented data from scratch.
cs.LG / 97 / 2609.17738
Similarity Pairing with Energy Mover's Distance for Self-Supervised Pre-Training at the LHC
Abstract
Many self-supervised methods for training foundation models at the Large Hadron Collider (LHC) rely on data augmentations to encourage the model to embed events into a representation space invariant to certain physical or detector symmetries. A common challenge arises from the large freedom in choosing a proper set of augmentations on which downstream performance depends. The implementation of augmentations involves either modifying existing events, potentially breaking the event fidelity, or simulating more event variants, which is computationally intensive. In this work, we present a data-driven method of pairing events by their similarity via the energy mover's distance (EMD), which measures how similar two events are in terms of the work required to transform one into the other. With this approach, distinct events are sampled and matched by their similarity to serve as views for learning invariance, keeping the physics content of each event intact without handcrafted distortions. We demonstrate this augmentation-free pairing method by pre-training on QCD jets via self-distillation and show that it can yield semantic jet embeddings with downstream discrimination power comparable to or better than an augmentation-based baseline.
cs.LG / 98 / 2609.18566
Deep learning emergent spacetime from fermionic spectral functions in holography
Abstract
We present a physics-informed machine learning framework based on Neural Ordinary Differential Equations that solves the holographic inverse problem: reconstructing the bulk spacetime and gauge field of a charged AdS black hole directly from boundary fermionic spectral functions. Encoding the UV asymptotics, horizon regularity, and zero temperature extremality as hard constraints in the neural network architecture, our framework reliably reconstructs the extremal Reissner-Nordström AdS geometry across three quantum critical regimes set by the $U(1)$ probe charge---non-Fermi liquid, marginal Fermi liquid (strange metal), and Fermi-liquid-like states---and can jointly infer the probe charge itself to sub-percent accuracy. Relaxing the near-AdS boundary constraint uncovers a geometrical degeneracy: bulk profiles that differ throughout the radial direction but share the same near-horizon $AdS_2 \times \mathbb{R}^2$ data reproduce identical spectral functions near the Fermi surface. This isospectral non-uniqueness is precisely the bulk degeneracy expected on general holographic grounds at zero temperature, and its spontaneous emergence across independent training runs shows that the network isolates the IR CFT universality rather than overfitting a single UV completion.
cs.LG / 99 / 2609.18894
Learning Lyapunov Operators for Nonlinear Systems
Abstract
Constructing Lyapunov functions for nonlinear dynamical systems is a central problem in stability analysis, yet remains challenging. Lyapunov functions are commonly characterized as solutions to first-order partial differential equations (PDEs), but these solutions are typically obtained for single systems, limiting their reuse across systems. In this paper, we study the Lyapunov solution operator that maps a vector field to the corresponding Lyapunov function defined by a dissipation-based Lyapunov PDE. We establish that, on compact subsets of the domain of attraction and under exponential stability assumptions, this operator is well-defined, unique, and continuous with respect to perturbations of both the vector field and the dissipation function. These results provide a theoretical foundation for approximating Lyapunov functions uniformly over families of nonlinear systems. Building on these theoretical foundations, we employ Fourier Neural Operators (FNOs) as a data-driven approximation of the Lyapunov solution operator. Numerical experiments demonstrate that a single trained operator can accurately approximate the numerical Lyapunov functions across parameterized families of dynamics. This illustrates the potential of neural operators for approximating Lyapunov functions.
cs.LG / 100 / 2609.18419
HiLNO: A Hierarchical Latent Neural Operator with Multi-Scale Supervision for PDEs on General Geometries
Abstract
Latent neural operators improve the efficiency of operator learning for partial differential equations (PDEs) by performing the main computation on compact latent representations. However, directly compressing the input representation to obtain such compact representations may discard solution-relevant spatial information, especially for PDE solutions with multiscale structures. To address this problem, we propose HiLNO, a hierarchical latent neural operator that constructs a fine-to-coarse-to-fine latent space and further introduces multi-scale supervision (MSS) and anisotropic Gaussian attention. The hierarchy mitigates potential information loss during compression, while MSS aligns intermediate predictions with downsampled target fields, encouraging solution-relevant structures to be captured across multiple spatial scales. Anisotropic Gaussian attention enables feature transfer across the hierarchy, making HiLNO applicable to general geometries. Experiments on representative PDE benchmarks and a large-scale automotive aerodynamics task show that HiLNO achieves competitive predictive accuracy, while reducing the parameter count by an average of 84.4% and FLOPs by an average of 69.2% compared with LinearNO. Additional experiments demonstrate effective generalization to unseen spatial resolutions. Code is available at https://github.com/JcLimath/HiLNO.
cs.LG / 101 / 2609.17759
Derivative-Free Structured Updates for Muon
Abstract
Muon updates matrix-valued neural-network parameters by orthogonalizing a gradient-based momentum matrix. Its reliance on derivatives limits its use when gradients are unavailable or unreliable. We develop a derivative-free framework that constructs Muon-style updates from structured finite differences. Four variants are considered: full entrywise recovery, random low-rank surrogates, basis-aligned rank-one probing, and direct structured search. Exhaustive basis-aligned probing is equivalent, up to positive scaling before ideal polar orthogonalization, to coordinate finite differences. Matrix-regression experiments show that random rank-one probing can reduce the number of function evaluations substantially, at the cost of less accurate updates. Controlled noisy-gradient experiments on regression and a neural network illustrate when accurate function values can compensate for an unreliable gradient oracle. A small CartPole study further examines orthogonal rank-one probes under a fixed episode budget. These results support structured probing as a practical option for selected black-box problems; they do not establish a general convergence guarantee or an advantage over accurate, inexpensive gradients.
cs.LG / 102 / 2609.17973
Matching Multi-Loop Complexities with a Single Loop: Optimal Optimization Stationarity and Best-Known Game Stationarity in Nonconvex--Concave Minimax Optimization
Abstract
We introduce a new single-loop algorithmic framework for smooth nonconvex--concave minimax optimization. The resulting projected damped extragradient method combines projected extragradient updates, dual momentum, and a moving proximal center. Under both the optimization-stationarity and game-stationarity criteria, our method achieves the best-known complexity among single-loop first-order methods. For optimization stationarity, our method achieves a gradient complexity of $O(L^2D_Y\barΔ_0\varepsilon^{-3})$, where $L$ is the gradient Lipschitz constant, $D_Y$ bounds the diameter of the dual feasible set, and $\barΔ_0$ is an initialization quantity involving the value-function gap and the initial gradients. Moreover, by incorporating a fixed-center warm-up phase, the complexity can be improved to $O(L^2D_YΔ_φ\varepsilon^{-3})$, up to an additive lower-order cost, where $Δ_φ:=φ(x_0)-\inf_xφ(x)$. We further establish a lower bound of $Ω(L^2D_YΔ_φ\varepsilon^{-3})$ for optimization stationarity over projected zero-respecting first-order methods. This lower bound proves that the warm-started version of our algorithm is optimal up to a constant factor for optimization stationarity within this oracle class. For game stationarity, our method achieves $\mathcal{O}\!(L^{3/2}D_Y^{1/2}Δ_φ\varepsilon^{-5/2})$ gradient complexity. This matches the best-known complexity of multi-loop first-order methods, thereby establishing the same complexity with a single-loop algorithmic structure. Under dual strong concavity, the proposed framework achieves $O\!(\sqrtκ\,LΔ_φ\varepsilon^{-2})$ leading complexity for both stationarity criteria, where $κ=L/μ$ is the dual condition number, up to an additive initialization cost. The $\varepsilon^{-2}$ accuracy dependence is optimal under fixed regularity and initialization bounds.
cs.LG / 103 / 2609.18416
Gradient Descent with Stochastic Subspaces via Persistence of Memory
Abstract
Stochastic subspace methods have gained popularity as gradient descent based techniques for large scale optimisation problems, especially in distributed settings. In this paper, we introduce the technique of "persistence of memory" to greatly extend and improve the random subspace methods. To this end, we leverage a vector that is only weakly correlated with the gradient in order to provide a guiding structure to the generative process of the random subspace along which the descent is going to take place. This guidance vector may be fixed for a large number of iterations, only to be refreshed at wide intervals (on whose size we can provide guarantees in terms of problem parameters). In important machine learning settings, such as optimisation problems embodying sparsity or a minibatch structure, we show that the guidance vector can be obtained in an effective and computationally inexpensive manner by leveraging the structured properties of the problem. En route, we establish to our knowledge the first theoretical analysis of classical SSD methods for sparse functions. In a local neighbourhood of the optimum, we demonstrate an alignment phenomenon of our gradient estimates with a low-lying eigenvector of the Hessian, allowing a once-for-all computation of the guidance vector which renders the method computationally favourable even in scenarios with unstructured objectives.
cs.LG / 104 / 2609.18489
Butterfly Effect and the Kinetic Energy Cascade in Probabilistic Machine Learning Weather Prediction Models
Abstract
This study analyses kinetic energy (KE) spectra, difference kinetic energy (DKE) spectra, and signatures of KE transfer across spatial scales in four state-of-the-art probabilistic machine learning weather prediction (MLWP) models: NeuralGCM-ENS, FourCastNet 3, AIFS-ENS, and GenCast. Results are compared with those from the physics-based numerical weather prediction model IFS-ENS. While NeuralGCM-ENS successfully reproduces the expected upscale transfer of KE, noise injection at its encoder stage underestimates mesoscale KE. Conversely, AIFS-ENS, GenCast, and FourCastNet 3 produce realistic KE spectral magnitudes but do not capture the expected upscale transfer of KE. In particular, AIFS-ENS and GenCast, which employ spatially uncorrelated stochastic perturbations, exhibit enhanced accumulation of KE at high wavenumbers. All examined models exhibit upscale error growth, reflected by the progressive shift of the DKE spectral peak toward larger wavelengths over time. However, the MLWP models struggle to reproduce the rapid initial growth of ensemble spread at small spatial scales associated with the butterfly effect. The results show that MLWP models can misrepresent the known scale transfer of kinetic energy despite producing skilful weather forecasts.
cs.LG / 105 / 2609.17736
Machine learning kinetics from molecular dynamics data
Abstract
Most molecular transitions occur on timescales far beyond direct molecular dynamics simulations. The committor, the probability that a configuration reaches a product state before a reactant state, is a central kinetic statistic, providing a mechanism-independent reaction coordinate and a foundation for transition path theory and the calculation of rates. This review surveys modern approaches for estimating the committor and related kinetic statistics from molecular simulations, with an emphasis on self-supervised methods that learn solutions of their defining dynamical equations rather than relying on labeled shooting data. We develop a common operator viewpoint connecting generator-based partial differential equations, variational principles, Markov state models, dynamical Galerkin approximation, and neural networks. Empirical and theoretical evidence points to the efficiency of these methods. We provide theoretical and practical guidance for realizing their full potential in applications, including strategies for treating non-Markovian effects and for sampling. We conclude by identifying opportunities for further research, including connections to reinforcement learning and generative modeling.
cs.LG / 106 / 2609.17721
Decoding Extrahepatic Targeting of Lipid Nanoparticles with Interpretable Machine Learning
Abstract
Lipid nanoparticles (LNPs) have transformed RNA medicine, yet their clinical utility remains constrained by predominant hepatic accumulation after systemic administration. Redirecting LNPs to extrahepatic tissues requires understanding of how lipid chemistry and formulation composition jointly govern in vivo biodistribution. Here, we develop an interpretable machine learning framework to predict hepatic versus extrahepatic LNP accumulation and identify molecular design rules for extrahepatic RNA delivery. A literature-derived dataset of 476 intravenous LNP formulations was curated from 81 studies, integrating formulation composition, lipid chemical structures, and IVIS-based biodistribution profiles. Standardized SMILES representations of ionizable lipids, helper lipids, sterols, PEGylated or polymer-conjugated lipids, additional lipids, and polymer repeat units were converted into RDKit Expert descriptors and combined with formulation-level variables to generate an 808-dimensional feature representation. Logistic regression, random forest, and XGBoost achieved ROC-AUC values of 0.839, 0.866, and 0.874, respectively. SHAP-based interpretation and consensus feature ranking revealed that ionizable-lipid descriptors dominate biodistribution prediction, while formulation composition, particularly ionizable lipid, sterol, and PEGylated/polymer-conjugated lipid fractions, contributes substantially. The top 20 consensus features retained nearly all predictive information in tree-based models. The most informative features implicated electrotopological surface properties, charge- and hydrophobicity-weighted surface areas, molecular topology, and amide/alkyl structural motifs as drivers of extrahepatic accumulation. This study establishes an interpretable, data-driven strategy for decoding LNP biodistribution and provides actionable design principles for engineering LNPs beyond the liver.
cs.LG / 107 / 2609.17896
QEMScore: How Much Does the Measurement Add to Learned Quantum Error Mitigation?
Abstract
How much does the noisy measurement add to learned quantum error mitigation? An accuracy table cannot say, because a model handed circuit structure can score well without reading the measurement at all. QEMScore adds the comparison that can. Each simulated circuit carries an exact ideal answer. The learned mitigator is scored beside a capacity-matched control, a model just as flexible that reads the same circuit description but never the measurement. Each method's measurement spend is accounted and not equalized. We run a controlled campaign on simulated circuits and reanalyze two published learned mitigators, Q-LEAR and QRAFT, from their released hardware data. Three findings stand out. First, under familiar within-family conditions (S0) evaluated across two spin-chain families and three seeds, continuous couplings identify the target, and the control that never reads the measurement matches 87.7 to 100.5 percent of the mitigator's gain over an affine fit to the circuit description. A plain polynomial in the coupling parameters, fitted after the campaign, beats the mitigator on all six evaluations, reflecting the selected learners' capacity. Second, for these selected learners, matching most of the gain is not matching the accuracy: on five of six evaluations the mitigator removes 19.5 to 74.5 percent of the error the capacity-matched control leaves, a learner-specific gap rather than a measurement requirement. Third, on released hardware data where descriptors only partially identify queries, the findings differ: flexible models of the descriptors show negligible mean gain over affine fits in Q-LEAR, and measurement inputs carry predictive gains in both Q-LEAR and QRAFT. These comparisons reflect representation- and protocol-specific behavior rather than an isolated cross-regime difference. A learned mitigator's accuracy should therefore be reported beside such controls.
cs.LG / 108 / 2609.17991
Fourier Analysis of Parametrized Interactive Quantum Classifiers
Abstract
Interactive Quantum Classifiers (IQCs) constitute a family of quantum machine learning models inspired by open quantum systems, in which the interaction between a target qubit and an environment is described by a Hamiltonian. Previous works introduced alternative Hamiltonian parameterizations and showed empirically that they can improve classification performance, but the role of these parameters in the resulting classifier remains poorly understood. In this work, we derive a closed-form expression for the reduced quantum channel generated by a parametrized IQC with a single target qubit. The analytical solution explicitly reveals how the Hamiltonian parameters control the constant, sine, and cosine components of the classifier output, establishing a Fourier interpretation of the induced feature map. This analysis motivates a generalized family of Hamiltonian encodings, including matrix-parameterized environmental Hamiltonians whose Fourier components depend on linear combinations of input features, thereby enabling non-separable Fourier structures. Numerical experiments on synthetic and real-world datasets show that the proposed models can improve classification performance on several nonlinear benchmarks. The generalized matrix encoding achieves the strongest aggregate performance in the evaluated benchmark, while a simpler four-parameter extension often attains comparable performance with substantially fewer trainable parameters. We additionally characterize the generated state ensembles using the standard fidelity-based expressibility measure, finding that global expressibility does not directly predict classification performance. Our results provide an analytical characterization of parametrized Hamiltonians in Interactive Quantum Classifiers and establish Fourier analysis as a useful framework for understanding and designing open-system-inspired quantum learning models.
cs.LG / 109 / 2609.17845
Sharp margin-based generalization bounds for realizable SVM
Abstract
Let the exact homogeneous hard-margin support vector machine be trained on \(m\) independent observations from a Borel probability law on a real Hilbert space. We prove that, with score zero counted as an error, there is a universal numerical constant \(C\) such that \[ \Pp\left( γ_m>0,\quad \Risk(u_m)> \frac{C}{m} \left( K_m+\log\frac1δ \right) \right) \le δ. \] Here \(γ_m\) is the empirical homogeneous margin, \(u_m\) is the exact minimum-norm unit-margin separator, \(r_m\) is the largest training radius, and \(K_m:=r_m^2\norm{u_m}^2=r_m^2/γ_m^2\) on \(\{γ_m>0\}\). The proof is driven by a deterministic deletion problem. Given vectors \(x_1,\ldots,x_n\) in the unit ball, delete a set \(B\) of constraints and let \(u_B\) be the closest point to the origin that satisfies every retained unit-margin constraint. Suppose that \(\norm{u_B}^2\le k\) and that every deleted vector has nonpositive score under \(u_B\). We prove that a family of such deletion sets of cardinality \(q\) has size at most \(\exp(8k+2q)\). The conceptual step is an exact identity obtained from the KKT representation of \(u_B\). For a random deletion set, the identity converts the mean squared spread of the separators into a weighted sum of score deficits. It therefore forces a coordinate whose deletion status separates the two conditional means by a quantitatively large amount. Revealing that coordinate decreases the conditional separator variance enough to control the binary entropy of the split. An entropy induction gives the deletion count, and an exact factorial ghost-sample identity converts that count into the stated high-probability SVM bound.
cs.LG / 110 / 2609.17892
Bracketing Uncertainty in Clustering Under the Manifold Hypothesis
Abstract
The manifold hypothesis suggests a natural criterion for clustering: partition data according to the manifold component from which each point is drawn. Whether two components are separable depends on a geometric tradeoff: the ambient separation between components versus the largest gap in sampling. In practice, this tradeoff is rarely assessed explicitly, leading standard methods to over-commit to a single clustering assignment even when the data do not support a unique answer. We formalize this tradeoff by combining intrinsic manifold geometry (volume growth and reach) with sample-level quantities (fill distance and density), yielding a threshold phenomenon for mutual-$k$-nearest-neighbor graphs: when the offset-to-fill ratio exceeds a conservative upper threshold, component separation is preserved; below a lower threshold, components fuse. The gap between these thresholds defines a geometric uncertainty zone in which the number of clusters is not identifiable from the data. Nevertheless, conventional approaches still seek one: sweeping parameters (an engineering approach) or fitting a generative mixture model (a model-based approach). Rather than forcing a single estimate of the number of clusters, we propose Manifold-Based Clustering (MBC), which returns an explicit bracket interval to quantify the underlying data uncertainty. This bracket acts as an empirically calibrated diagnostic: it narrows when a single resolution is supported, widens when multiple resolutions coexist, and collapses to one when no separated structure is detectable. Empirically, we find that many real datasets lie within the uncertainty zone rather than admitting one clear answer. Our results suggest that ambiguity in cluster number is often intrinsic, and should be quantified rather than resolved.
cs.LG / 111 / 2609.18118
Preservation of Log-Concavity and Convergence of Wasserstein-Fisher-Rao Gradient Flows
Abstract
We study the convergence of Wasserstein-Fisher-Rao (WFR) gradient flows for sampling from probability distributions known up to a normalisation constant. By combining Wasserstein transport with Fisher-Rao birth-death dynamics, WFR flows balance exploration and selection. These flows have been recognised as a promising mechanism to accelerate convergence beyond Langevin dynamics. We show that for a class of strongly log-concave target distributions satisfying additional curvature conditions, WFR flows preserve strong log-concavity, in contrast to Wasserstein flows which enjoy this property only in the Gaussian setting. Exploiting this result, we derive explicit non-asymptotic convergence rates for the symmetrised Kullback-Leibler divergence, without requiring a warm-start as required in current estimates. In particular, we show that the convergence rate decomposes additively into Wasserstein and Fisher-Rao contributions, thereby confirming a recent conjecture within this setting. These results provide refined convergence guarantees and further develop the theoretical foundations of WFR gradient flows for sampling and Bayesian inference.
cs.LG / 112 / 2609.18682
Rank and computation of the pathlifting Jacobian of a DAG ReLU network
Abstract
This paper provides a self-contained proof of the rank of the pathlifting Jacobian of a DAG ReLU network by performing an induction on the network's number of hidden nodes. In fact, the induction is elementary, and the key recipe is to consider the skeleton matrix of the network, a sparse matrix encoding the network paths, and transform the representation of one of its hidden neurons into an output node. The proof relies on intermediate propositions which link the pathlifting, its Jacobian, the network parameters, and its skeleton matrix, which, on top of permitting to conclude on the rank of the pathlifting Jacobian, also provide a way to compute it without backpropagation and whose computation cost is super efficient in practice compare to usual backpropagation. The paper is provided with a Python module that implements the different propositions of the paper for feed forward networks and is used to experimentally quantifies the computational gain of computing the pathlifting Jacobian with the proposed theory.
cs.LG / 113 / 2609.18901
Fast Learning Rates for Physics-Informed Kernel Methods
Abstract
In physics-informed machine learning, a target function $u^*$ is learned from noisy value observations $y_i=u^*(x_i)+ \varepsilon_i$, together with differential information, given either by noisy observations $d_j=(Du^*)(z_j)+ξ_j$ or by a known physical constraint $Du^*=v$. We consider the setting where $D$ is a linear differential operator and analyze a physics-informed kernel estimator $\hat u$ combining $n$ value observations and $m$ differential observations. In this context, we ask how much can differential information improve predictions, and how does this improvement depend quantitatively on $n$, $m$, and $D$. We prove finite-sample bounds, supported by numerical simulations, revealing a two-regime structure for the prediction error. When $m$ is limited, the rate depends jointly on $n$ and $m$; when $m$ exceeds a problem-dependent threshold, the rate saturates and matches the oracle rate obtained when the perfect constraint $D \hat u = Du^*$ is imposed. Examples are discussed for Sobolev spaces which are reproducing kernel Hilbert spaces and include partial Laplacian constraints on the torus and gradient observations on bounded domains. These examples illustrate the range of possible learning rate improvements --- from the standard nonparametric $n^{-1/4}$ to the parametric rate $n^{-1/2}$. Finally, we derive physically consistent rates in a stronger norm that jointly controls the errors in $\hat u$ and $D\hat u$.
cs.LG / 114 / 2609.19083
A General Kernel Framework for Non-CND Distance Measures Using |D|-Dimensional Sparse Landmark Embeddings
Abstract
Kernel methods, and Gaussian Processes (GPs) in particular, require a Hilbertian distance measure---one whose square is conditionally negative definite (CND)---to guarantee positive semi-definiteness (PSD) of the kernel matrix; a condition that fails for many natural input spaces, including smooth manifolds and spaces of probability distributions. We propose the Sparse Landmark Embedding (SLE) kernel, which eliminates this requirement entirely. Each input is embedded into a sparse feature vector via compactly supported bump functions centered at all |D| training points; applying any standard PSD kernel in this embedding space yields a kernel that is provably PSD for arbitrary distance measures. The compact support automatically controls embedding sparsity, keeping kernel matrices well-conditioned and computationally tractable despite the high ambient dimension. We provide theoretical guarantees on PSD, sparsity, stability, and universal approximation, and demonstrate, using geodesic and Wasserstein distances, that the SLE kernel matches or substantially exceeds domain-specific baselines in both predictive accuracy and uncertainty quantification.
神经与进化计算 (cs.NE)
4
cs.NE / 1 / 2609.17894
Graph neural networks for exoplanet atmospheres
Abstract
Calculating disequilibrium chemistry in exoplanet atmospheres remains a significant computational bottleneck in atmospheric retrievals. The increasing observational precision from facilities such as JWST and the Ariel mission requires including disequilibrium chemistry in these analyses. Previous studies have demonstrated that neural networks can emulate kinetic chemistry, although their spatial inductive bias does not align with the topology of chemical reaction networks. This study introduces a graph neural network surrogate that represents chemical species as nodes and temperature-dependent reaction rates as edges, thereby enabling information propagation along physically meaningful chemical pathways. The model is trained on atmospheres generated using the Venot+2020 chemical scheme and Guillot temperature-pressure profiles. The GNN accurately reconstructs disequilibrium abundances across the sampled parameter space and reduces the mean abundance error by a factor of approximately 3 compared to the previous U-Net model. When applied to transmission spectra, most predictions fall within the observational precision expected for JWST and Ariel, with only about 7% of test atmospheres exceeding a 20 ppm mean spectral error. Performance variations are primarily observed in chemically transitional regimes near a carbon-to-oxygen ratio of one and at low temperatures. An evaluation of the boundary-case planet WASP-39b demonstrates effective performance under a moderate domain shift. Perturbation analysis indicates that disturbances propagate along chemical connectivity rather than spatial adjacency, confirming that the architecture captures the structure of reaction networks. These results suggest that GNN surrogates provide accurate, computationally efficient predictions of disequilibrium chemistry, facilitating integration into the atmospheric retrieval pipeline.
cs.NE / 2 / 2609.18130
Benchmarking Tabular Foundation Models as Surrogates in Expensive Evolutionary Optimization
Abstract
Surrogate-assisted evolutionary algorithms (SAEAs) are effective methods for solving expensive optimization problems (EOPs), where surrogate models replace most expensive evaluations and critically influence the final optimization results. In recent years, tabular foundation models have advanced rapidly, and the Tabular Prior-data Fitted Network (TabPFN) has been adopted as a surrogate model for EOPs due to its strong predictive capability, demonstrating promising performance. Motivated by its potential as a surrogate model in SAEAs, this work conducts a comprehensive study that combines extensive experiments with in-depth theoretical analysis to investigate the effectiveness of TabPFN. Specifically, we perform experiments across both offline and online SAEA settings, covering diverse problem scenarios such as single-objective, multi-objective, constrained, combinatorial, mixed-variable, and engineering optimization problems. In addition, we further analyze the advantages and limitations of TabPFN within SAEAs and provide practical guidelines for its application in different optimization settings. Results show that the effectiveness of TabPFN is highly problem dependent, and it cannot replace conventional surrogates universally. Overall, TabPFN should be adopted selectively according to data availability, landscape complexity, search space characteristics, and its role within the algorithm. Customized model management strategies and role-specific algorithm design are necessary to fully exploit its advantages and avoid its pitfalls.
cs.NE / 3 / 2609.18702
A Metaheuristic Optimization Framework for Discrete Optimization under Strict Time Limits
Abstract
Real-time applications often rely on optimization approaches that can find high-quality solutions to hard problems on the order of milliseconds. Metaheuristic optimization frameworks (MOFs) are useful tools for such tasks, as they provide large sets of general-purpose search mechanisms that can return solutions under different computational budgets. However, existing work largely overlooks the available computation time as an explicit dimension of analysis. In this work, we introduce STILO, a MOF specifically designed for optimization under strict time limits. STILO integrates fine-grained configuration spaces for ant colony optimization (ACO), genetic algorithm (GA), and simulated annealing (SA), combining existing and novel operators. We performed experiments using both synthetic and benchmark instances of various discrete optimization problems. The results indicate that the proposed discrete distance calculation mechanism for SA is useful for optimization under strict time limits. They also show that the relative effectiveness of the proposed problem-independent graph structures for ACO can vary across time limits, even for the same instance characteristics. More generally, the results demonstrate that the effectiveness of algorithm families and operators depends not only on the problem type, but also on the characteristics of the instance and the available computational budget.
cs.NE / 4 / 2609.18033
Neural noise enables accurate internal simulation of rare events
Abstract
The brain needs an accurate internal model of the world to generate predictions and guide behavior. However, it must estimate the statistical structure of the environment from limited experience. This is particularly difficult for rare events, whose observed frequencies in a limited sample may substantially under- or overestimate their true frequencies. How the brain constructs an accurate internal model despite this sampling problem remains unclear. We address this problem using a Bayesian Confidence Propagation Neural Network (BCPNN) trained on event sequences from a Markov-chain random walk with controlled event frequencies. Treating the underlying Markov structure as the ground truth, we train the network on limited sample of event sequences and then allow it to generate autonomous replay based on the learned structure. We evaluate replay fidelity at the levels of both marginal event frequencies and conditional transition structure. We find that moderate neural noise, modeled as temporally correlated random fluctuations in unit activity during replay, is critical for faithful internal simulation. Without this variability, deterministic replay systematically under- or overrepresents rare events, whereas moderate noise restores both their marginal and conditional occurrence. Moderate noise also broadens the range of parameter values that produce accurate replay, making the model more robust to parameter variation. Together, these results support noise-assisted internal simulation as a potential mechanism for compensating for sampling errors arising from limited experience. Our model also provides a testable framework for investigating how altered neural variability may impair internal-model fidelity in disorders such as Parkinson's disease.
计算语言学 (cs.CL)
58
cs.CL / 1 / 2609.17708
Confidence Comes from Experience: Experiential Confidence Estimation from Reasoning to Agents
Abstract
Reliable confidence estimation is increasingly central to the trustworthy deployment of language models: a calibrated estimate of the probability that an output is correct decides what to ship, what to escalate, and what to retry. Existing confidence estimators, however, share one design premise: they only read the current inference process, either by introspecting on it, scoring its token probabilities, or resampling it. We argue that the current inference is not a sufficient basis for confidence. We propose XConf (eXperiential Confidence): estimating confidence together with the model's accumulated experience. The experience is stored as a record of the model's own graded past episodes, each holding the task, the model's reflection, its stated confidence, the outcome, and a lesson written once the grade arrived. Given a new task, XConf's Recall stage retrieves past episodes on similar tasks met with a similar stated confidence, and reads off their historical success rate; its Reflect stage shows the model this record, has it name its recurring failure mode, and restate a confidence now informed by its own track records. Our estimator is format-general, requiring no logit access or weight updates, and costs only one answer generation. Across nine benchmarks spanning reasoning, coding, multimodal QA, and interactive agents, and four models from three families, XConf beats or matches ten-sample self-consistency in discrimination (AUROC) on 23 of 24 comparisons, with much lower calibration error (ECE), at a tenth of the generation cost. Used for selective prediction, abstaining on the 10% least-confident episodes raises the delivered success rate by up to 8.7 points on agent tasks. We therefore see experiential confidence estimation as a new paradigm for future general-purpose confidence estimation.
cs.CL / 2 / 2609.17747
Is Trump's Vocabulary Poor? Vocabulary Richness Across Texts of Different Lenghts
Abstract
This study explores the vocabulary richness of oral political communication. A model explaining the lexicon growth is proposed by subdividing the whole vocabulary into terms generated by general and specialized glossaries.
cs.CL / 3 / 2609.17755
Evolution of US Oral Political Language
Abstract
The analysis of US political language is usually based on the written form (e.g. presidential addresses) or posts broadcasted on various social networks. Oral production, however, which is even more frequent, can better reveal the style and mode of thinking of the speaker. This study covers this mode of linguistic communication by considering 19 candidates from the presidential elections between 1960 to 2024. Our main research objectives are to disclose the main trends hidden in those presidential debates. Do we observe a clear simplification of the US political language over time? Does Trump have poor language compared to the other candidates? Do unusual stylistic features occur only with a single, specific president? Moreover, can we detect a pattern explaining the success or failure of some nominees? Over time, this study demonstrates a significant reduction in political language complexity, a decrease of the mean sentence length, and a noteworthy decline of complex terms. Moreover, the emotional tone increases over the decades, while logical and rational thinking tends to lessen.
cs.CL / 4 / 2609.17762
Is Luke the Author of a Gospel and the Acts of the Apostles?
Abstract
According to Christian tradition, Luke is credited with authoring a Gospel and the Acts of the Apostles, even if his name does not appear in either book, both originally written in Koine Greek. Several biblical scholars assume that both texts were written by a common author, while others deduce the presence of two authors. Different studies have been found to support either finding, some based on qualitative evaluation, while a few others consider the occurrence frequency differences between the two books. To propose an enhanced quantitative analysis, this study is grounded on two recent authorship attribution models. The Burrows' Delta, applied with eleven different feature sizes, demonstrates common authorship. An author verification model confirms this finding. The following experiments consider several stylistic representations, feature sizes, and distance functions to confirm that Luke is the true author of both books.
cs.CL / 5 / 2609.17764
How Calibration Content Shapes Attention-Based Reranking
Abstract
Attention-based rerankers score documents by aggregating query-to-document attention and subtracting a null-query calibration pass to remove positional and structural bias. Although widely used, this calibration assumes that the null pass removes irrelevant signal from each document. We show that modern prompt content, e.g. constraints, instructions, personas, and demonstrations can violate this assumption when it enters the scoring readout, making the null pass relevance-aware rather than null. We find that calibration is especially harmful when applied to prompts containing longer, more detailed instructions as the null-pass step removes relevant signal. Based on these findings, we propose interpolated null calibration, a training-free modification that controls how much of the instruction content enters the null baseline. It recovers attention-based reranking performance on instruction-heavy tasks where standard calibration fails, while preserving calibration's benefits when the null pass remains relevance-agnostic. On instruction heavy tasks, the recovered rankings surpass generative rerankers. We also show that in-context demonstrations improve attention-based reranking with little calibration interference, since demonstrations act only through the query pass and leave the null pass unchanged.
cs.CL / 6 / 2609.17846
PrimeScientist: Strategic Allocation of Research Effort in Autonomous Research
Abstract
Autonomous research agents aim to automate scientific workflows, from proposing ideas to conducting experiments and analyzing results. Yet current AI and research agents can propose more directions than available resources allow them to pursue. Moreover, each attempt could consume substantial resources, requiring agents to reconsider how to invest in subsequent research. Thus, deciding how to invest research effort strategically should be a defining capability of autonomous research agents. Accordingly, we introduce PrimeScientist, which jointly determines research direction and resource investment across successive research attempts. Specifically, we formulate this challenge of strategic research effort allocation as a sequential decision problem where remaining resources should explicitly guide the research policy. We first introduce an executable plan tree that preserves competing plans and their outcomes across attempts. Building on this representation, we propose an adaptive MCTS-based allocation policy that balances exploration and exploitation using experimental feedback and remaining resources. Comprehensive evaluations across AI research, systems and code optimization, and machine learning engineering show that strategic allocation improves research quality and sample efficiency together. Across 12 AI research tasks, PrimeScientist improves average reward by 10.3% with 50.6% fewer research attempts than AutoResearch under the same resource budget. We believe making research effort allocation an explicit optimization target establishes effective resource use as a core research capability for autonomous agents to drive scientific breakthroughs at scale.
cs.CL / 7 / 2609.17848
SFT or RL for Tool-Calling Agents? A Controlled Study Across Data, Method, and Scale
Abstract
Limited controlled evidence exists on how training data, adaptation method, and model scale jointly affect tool-calling performance in language-model agents. We evaluate supervised fine-tuning (SFT) with LoRA, reinforcement learning (RL) via Group Relative Policy Optimization (GRPO), and SFT followed by GRPO across six Qwen3 models from 0.6B to 32B parameters, covering both in-distribution performance and cross-dataset transfer. SFT with LoRA is the strongest in-distribution method throughout the 0.6B-32B range and best in 15 out of 18 experimental settings. On cross-dataset transfer, the methods are closer: GRPO wins 29 out of 54 settings where training and test datasets differ, but its margin over SFT averages under one point, and SFT->GRPO is rarely strongest in either comparison. Dataset mixing gives consistently strong transfer while staying close to specialized in-distribution training, regardless of method. Additional analysis further confirms that LoRA outperforms full-parameter fine-tuning, demonstrating that LoRA better preserves pretrained agentic behavior.
cs.CL / 8 / 2609.17853
AfriSyCo: Measuring Assertive Framing, Verification, and Wording Sensitivity Around African-Language Content
Abstract
AfriSyCo studies answer switching around African-language factual content with two complementary layers: native-language follow-ups and a controlled cross-language factorial whose question, options, and target remain in the African language while the follow-up framing is English. We analyze 1,415 turn-1-correct model-language-item observations derived from 100 source questions across seven open-weight checkpoints and six languages; turn-1-correct denotes observed first-response accuracy, not demonstrated knowledge. Under native prompts, assertive endorsement produces 29.3 percentage points more any-turn false-target selection than mention-plus-verification (M+V), with a 19.0-point immediate T2 contrast. In the precommitted 2 x 2 factorial, averaged over three tested prompt families, assertive framing increases target selection by 30.4 points (95% CI [28.4, 32.3]); verification decreases it by 17.4 points, while the assertive effect rises from 20.5 points without verification to 40.2 with it (interaction +19.7). The effect remains 34.8 points among 611 observations correct after option reordering. Magnitude varies sharply by wording and checkpoint: prompt-family effects span 20.1-42.5 points, a Twi/Qwen3 paraphrase shifts target selection from 70.8% to 4.2%, and checkpoint effects span 9.2-47.0 points. Prompt realization is therefore part of the measurement problem.
cs.CL / 9 / 2609.17857
Who Judges Matters: Measuring Family-Conditioned Preference in LLM-as-Judge Panels
Abstract
Who the judge is can affect an LLM-as-judge result, but measuring that effect without confusing it with candidate quality is difficult. We study four open-weight families (Llama 3.1, Qwen 2.5, Gemma 2, and Yi 1.5) in a fully crossed pairwise design with 9,312 judgments. A common per-family statistic is strongly confounded with candidate quality and correlates with Bradley-Terry ability at r = 0.95. We derive a corrected estimator that holds the candidate family fixed and compares judges. All four families then show a positive same-family lift (3.4-8.4 percentage points), with global FPS 0.067 (95% CI [0.053, 0.084], permutation p = 0.0002). The effect remains under panel-based quality controls, an independent human-consensus anchor, and a float16 judging replication. Judge-side likelihood is closely related to the effect: adding likelihood advantage reduces the controlled coefficient by 61%, which we treat as descriptive attenuation rather than causal mediation. Position is a separate failure mode. Across the panel, 55.4% of AB/BA pairs reverse, and reversal above 50% is incompatible with a simple independent content-noise model. Relative to a family-balanced reference, panel composition changes 18.5% of pairwise outcomes. A complete reproducibility archive has been prepared for public release.
cs.CL / 10 / 2609.17956
TACTICS: Taxonomy-Aware Intelligent Corpus Sampling for Machine Translation
Abstract
Large-scale machine-translation (MT) systems are typically evaluated on random samples from a corpus whose distributional composition is an artifact of how it was assembled. Such a sample inherits the phenomena the collection happens to contain rather than the full space a system must handle, spanning rule-governed conventions (terminology, punctuation, currency formatting) and context-dependent phenomena (tone, honorifics, document-level coherence), and thus provides no coverage guarantee for assessing robustness. We propose TACTICS (Taxonomy-Aware Coverage-opTimized Intelligent Corpus Sampling), which recasts coverage as an explicit objective. TACTICS induces a hierarchical taxonomy from a locale style guide, classifies segments against it, and selects a fixed-budget subset jointly optimizing coverage of rare categories, document-level coherence, and distributional fidelity to the full corpus. Applied to MT evaluation across four translation directions, TACTICS improves coverage of rare categories over lexical and embedding-based selection. By targeting the phenomena that separate systems, TACTICS makes a fixed evaluation budget go further, recovering the true system ranking from far fewer segments than random sampling wherever a real quality gap exists and never signaling a difference where none exists.
cs.CL / 11 / 2609.17996
Modeling the Developmental Shift in Telicity Acquisition
Abstract
Acquiring telicity, which is the distinction between bounded (e.g., ate an apple) and unbounded (e.g., ate apples) events, requires first language (L1) learners to map surface-level and semantic cues to abstract event structures, but the computational trajectory of this mapping is not well understood. We introduce a Difference in Surprisal method that uses GPT2 token surprisal over paired temporal adverbial diagnostics (in an hour versus for an hour) to automatically label telicity across English CHILDES corpora, validated against expert linguist judgments. Using these labels, we train diagnostic logistic regression classifiers on 12 syntactic and lexical semantic features to compare how child speech and child-directed speech encode telicity. The two models diverge: the child model reaches near perfect accuracy through a single deterministic cue, the presence of a post-verbal determiner, while the adult model relies more heavily on verb class and other lexical semantic features, with the determiner cue neutralized. This trajectory supports Syntactic Bootstrapping: learners first exploit high-frequency structural cues as a scaffold to bootstrap, before developing fully compositional, verb-based event structures.
cs.CL / 12 / 2609.18011
Gaze as Evidence for Common Grounding: A Cross-Corpus Analysis of MapTask and MUNDEX
Abstract
In collaborative tasks with asymmetric information, participants coordinate their understanding through interaction. We ask whether gaze provides evidence about grounding across two such tasks. Working from discrete behavioral annotations, we map HCRC MapTask (Anderson et al., 1991) and MUNDEX (Türk et al., 2023) into a shared partner/task/away vocabulary and compute gaze features around task-relevant dialogue units. In both corpora, aligned reference interpretations (MapTask) and UND (understood) judgments (MUNDEX) are associated with more task-directed gaze and with less partner-directed gaze, lower gaze entropy, and fewer gaze transitions. The associations are clearest for the participant leading the task: in giver-produced references, and in explainer judgments, which also co-vary with the explainee's gaze. In same-speaker MapTask reference chains, the speaker's gaze entropy is lower at the mention where a previously non-aligned referent becomes aligned. The best gaze feature groups improve modestly over controls under grouped cross-validation: temporal features in MapTask and raw proportions in MUNDEX. Because effects are small and several weaken when recurring participants rather than dialogues are the unit of inference, we treat gaze as one contributing cue to grounding, to be interpreted alongside task and dialogue context.
cs.CL / 13 / 2609.18047
Exact semantic readout from compressed vector representations
Abstract
We characterize when compressed vector representations admit exact linear or affine readouts of a finite lexicon's truth conditions: one fixed map per predicate, sending each entity vector to the corresponding truth vector. A necessary and sufficient row-space condition determines existence; the augmented truth matrix has rank r, giving minimum dimension r in the linear case, and r-1 in the affine. Exact readouts return values in a shared truth basis on which Boolean connectives act unchanged; separability alone requires an intervening threshold. For binary relations, exact bilinear readout of identity or strict total order requires linearly independent entity vectors. Experiments with GloVe and word2vec distinguish exact affine recovery, linear separability, and held-out prediction: most predicates are strictly separable, but none admits an exact affine readout from the pretrained embeddings. Supervised transductive training attains exact affine recovery to numerical precision at every tested dimension meeting the bound. At the embeddings' original dimension, geometries constrained to exact linear recovery retain 98-99 percent of the pretrained variance on the feature norms, and 80-83 percent on the WordNet lexicon.
cs.CL / 14 / 2609.18135
DualSQL: Text-to-SQL with Multi-Agent Reinforcement Learning
Abstract
State-of-the-art Text-to-SQL systems are typically multi-agent pipelines centered around two fundamental tasks: schema linking and SQL generation. However, existing work trains separate models for each task, failing to leverage the synergy between these interrelated tasks. In this work, we propose DualSQL, a new Text-to-SQL system consisting of two agents powered by a single model backbone. The agents share the same model weights and agentic scaffold, enabling joint optimization through a robust multi-agent reinforcement learning (RL) framework. We design three database access tools to facilitate effective multi-step reasoning grounded to interactions with the databases. To improve training and avoid model collapse, we introduce a set of rollout guardrail mechanisms that stabilizes multi-agent RL training, supporting DualSQL to keep improving during training. We also introduce a new SQL correctness metric, robust execution match (REX), to more accurately judge SQL correctness and assign reward signals. Being trained on only 3755 examples, DualSQL-4B achieves an impressive 68.0% execution accuracy on the BIRD development set, matching previous 7B models. DualSQL-8B further improves to 71.1%, outperforming previous state-of-the-art single-model solutions with 32B parameters. These results demonstrate the strength of joint multi-agent reinforcement learning for building high performance Text-to-SQL pipelines.
cs.CL / 15 / 2609.18194
T-SANDHI: Tone Sandhi-aware Adaptive Network with Decoupled Hybrid Injection for Low-resource Taiwanese Hokkien Speech Recognition
Abstract
In Taiwanese Hokkien automatic speech recognition (ASR), prior studies often treat tone sandhi as a major challenge under the assumption that models fail to process implicit phonological variations. However, our experiments on Taiwanese Hokkien reveal that speech foundation models actually handle tone sandhi variations effectively, and the real performance bottleneck stems from a localized confusion between these variations and retained citation tones. To address this, we propose T-SANDHI to explicitly decouple surface acoustics from underlying lexical intent on top of a frozen Whisper backbone. Using a lexicon-guided multi-task learning structure driven by text-derived pseudo labels, our lightweight hybrid injection module integrates independent citation and sandhi phonetic streams via dynamic gating. Extensive evaluation on the TAT-MOE corpus and two blind test sets demonstrates that this explicit disentanglement effectively resolves tonal mapping confusion, outperforming baselines with strict parameter efficiency.
cs.CL / 16 / 2609.18203
Behavior2Value: Benchmarking and Empowering LLMs for Consumer Value Measurement from E-commerce Behaviors
Abstract
Human values are deep motivational orientations that shape human behaviors. In e-commerce, they reveal the stable drivers behind users' purchase decisions. Compared with short-term interests, consumer values better explain how users evaluate products before purchase. However, consumer values are often implicit in complex and fragmented behavioral trajectories, leaving value measurement from e-commerce behaviors largely underexplored. To this end, we propose the Behavior-to-Value (B2V) task, which aims to identify consumer values from e-commerce behavioral trajectories. Centered on this task, we first construct the E-commerce Consumption Value Taxonomy (ECVT) and introduce B2V-Bench, the first B2V dataset and benchmark, based on anonymized Taobao behavioral logs. B2V-Bench consists of real-world purchase decision episodes, covering 25 types of purchase behaviors, along with corresponding consumer value orientations manifested in each episode. To improve consumer value measurement accuracy, we further present B2V-Verifier, a behavior-to-value measurement model based on Value Verification Tuning, which learns to assess whether behaviors provide sufficient evidence for each value inference. Experiments show that B2V-Verifier outperforms strong LLM baselines, improving multi-label classification by 34\%. The dataset and code will be publicly released upon acceptance.
cs.CL / 17 / 2609.18274
I code or AI code: A comparative evaluation of AI-rated scores in classroom observations
Abstract
Classroom observations are widely recognized as a key tool for establishing benchmarks of education quality and guiding pedagogical improvement, yet they remain resource-intensive and dependent on trained observers. This study evaluated the feasibility of using a LLM (GPT-5 model) to score teacher-child interactions in early childhood classrooms, benchmarked against human raters. The study analyzed 87 video-recorded observations from 38 classrooms across 30 kindergartens in Hong Kong. Using observation transcripts, the AI model was configured to apply the full Classroom Assessment Scoring System (CLASS) framework. AI-rated scores were then compared with human ratings by examining correlations and differences in mean scores of the CLASS domains and dimensions. The results showed greater convergence between AI and raters for the Emotional Support domain and, in particular, the Quality of Feedback dimension, which captures how teachers use feedback to extend children's learning. Greater divergence emerged for interactions that were more procedural or context-dependent, particularly within the Classroom Organization and Instructional Support domains. These findings suggest that transcript-based AI scoring may capture some of the relative variation in teacher-child interactions but cannot yet reproduce calibrated human judgements consistently across the full CLASS framework. AI-assisted observation may therefore be more appropriate as a preliminary screening tool rather than as a replacement for trained observers, providing teachers with evidence for reflection rather than high-stakes evaluation. Future research should examine whether domain-specific training and incorporation of contextual and visual information can improve alignment between AI and human rated scores.
cs.CL / 18 / 2609.18284
Made in Hungary: Comments on the performance of generative language models
Abstract
In recent years, three initiatives have emerged to develop generative language models in Hungary. The motivation behind them is the same. For Hungarian, no model with the given capability existed, or existing English-centric models offered limited proficiency. A detailed examination of the corresponding studies, however, reveals several methodological limitations. First, the reliability of the evaluation protocols is questionable. Contrary to the findings of Csibi et al. [2026], evaluation under the recommended inference settings shows that Qwen3-4B achieves higher scores than Racka-4B, its Hungarian-adapted version. Data contamination is evident in the work of Yang et al. [2025d] and Szentmihályi et al. [2025], potentially biasing the reported results. Second, the training pipelines fall short of current best practices in corpus curation and data mixture, which risks wasting substantial compute on low-quality data. The lack of controlled ablations prevents reliable assessment of these choices. Third, none of the three papers assessed forgetting or capability loss. Testing the adapted models on a subset of the original benchmarks indicates performance decline in all three cases, especially Racka-4B. These observations emphasize the importance of rigorous experimental design in language model development, given the significant computational and financial costs involved.
cs.CL / 19 / 2609.18310
SEA-LION-v4.8: A Technical Report
Abstract
We introduce Nemotron-SEA-LION-v4.8, a family of Southeast Asian Languages in One Network (SEA-LION) built upon NVIDIA Nemotron 3. The family includes 30B-A3B and 120B-A12B models, with both continued-pretrained base checkpoints and post-trained variants. We adapt the models using Southeast Asian, reasoning, code, and multilingual parallel datasets, followed by post-training with supervised fine-tuning and online on-policy distillation. On SEA-HELM, the 30B-A3B model improves the overall SEA score from 46.06 to 51.57, while the 120B-A12B model improves from 49.30 to 63.44. The strongest gains are observed in instruction following, natural language reasoning, and natural language understanding across seven Southeast Asian languages.
cs.CL / 20 / 2609.18385
Emotion Experience, Expression, and Perception: Emotion Analysis on Multimodal Social Media Posts
Abstract
Emotions are an essential aspect of human communication, particularly on social media, where authors frequently combine text and images to convey their emotions. Yet prior work on emotion analysis of social media posts has overlooked two important aspects in regard to measuring how well readers can reconstruct the authors' intent: (1)~the image modality, with most work focusing solely on text, and (2)~the real-world events that trigger the expressed emotions, and their relationship to the post content. We therefore study the relation between (a) the author's experience of the event that caused them to write a social media post and (b) the content of the post, with a focus on readers' capability to reconstruct that emotion expression. To do that, we introduce the Multimodal Multi-Emotion-Model dataset Mult2EMo, created by collecting annotations from both authors and readers on the posts and their triggering events. We find that reconstruction is possible but challenging for both human readers and computational models. We show that understanding the triggering event is crucial for accurate reconstruction, and that reconstruction is particularly challenging when posts rely heavily on the image to express emotion.
cs.CL / 21 / 2609.18417
Dependency-Aware Trajectory Refinement for Efficient Multi-Turn Agent Fine-Tuning
Abstract
Multi-turn agent trajectories often contain redundant rounds (failed tool calls, parallel sub-queries, verification-only steps) that inflate both training and inference cost. We propose viewing each trajectory as a \emph{round-level dependency DAG} that exposes which rounds are globally load-bearing for the final answer, and fine-tune agents on trajectories refined through this DAG. Given an LLM-annotated DAG, these edits are deterministic and interpretable, with optional rephrasing. Models trained on these refined trajectories consistently outperform those trained on the original trajectories at lower inference cost. Specifically, across four multi-modal QA benchmarks, our refinements improve downstream accuracy by up to $1.7$\,pp over vanilla SFT (and $5.7$\,pp over an LLM-deletion baseline) while reducing per-sample inference messages by up to approximately $40\%$ and inference tokens by up to approximately $48\%$, translating to substantial savings in compute and serving cost. Code is available.
cs.CL / 22 / 2609.18440
Planning or Improvisation? Stress-Testing the Poetry Planning Site on Open Models and Open Cross-Layer Transcoders
Abstract
Lindsey et al. (2025) report that Claude 3.5 Haiku plans rhymes: features for candidate rhyme words are active on the newline before a line is written, and a suppress-and-inject intervention redirects the line only when applied there (their Figure 13). We test how far this generalizes on seven cells crossing four open models (0.6B to 2.6B parameters) with six open cross-layer transcoders (CLTs), on one consumer GPU, decomposing the claim into position specificity (C1), newline site identity (C2), and a newline-resident plan (C3). This is a stress test rather than a faithful reproduction: attribution graphs are unavailable for these CLTs, so features are found bottom-up from decoder vectors. C1 generalizes, in every cell and in all 247 of 444 prompt-by-inject pairs with a detectable effect, but the effective position is the final prompt token, adjacent to emission, and only two cells reach behaviorally meaningful probabilities. C2 and C3 are not recovered by any probe: a census of every active feature finds no rhyme-anticipating enrichment at the newline, and steering the newline while the model composes the whole line, over 36 runs and 8,640 sampled lines, shows why. That intervention is strong but one token long, making the injected word the first word of the composed line in 703 of 720 samples and leaving the rhyme six words later untouched. A final test drops the transcoder entirely: patching the newline's whole residual, at every layer, from a minimal-pair poem whose third line ends on a different rhyme moves the rhyme in 11 of 1,260 composed lines against 4 at baseline, with a design resolving 1.4%. We read this as a boundary condition rather than a refutation: at this scale and with these transcoders, the causal site is emission-adjacent. We reproduce Figure 13's shape, not its mechanism. Code and data are public (code: github.com/PCfVW/poetry-planning-site).
cs.CL / 23 / 2609.18445
M-SQE: Multilingual Skill Quality Estimation for Enhancing Language Equality in Agentic Skill Use
Abstract
Agent skills, reusable procedural documents that extend LLM agents beyond their parametric memory, have become an important interface for deploying agents on real-world tasks. Community-maintained skill libraries built around this interface are growing rapidly. However, this ecosystem remains deeply English-centric: our audit finds that low-resource languages such as Swahili and Hindi have no in-language skill content, so retrieval often returns a skill written in a different language than the query, degrading accuracy and recall. A practical solution is to synthesize in-language skills for retrieval but the quality can be unreliable, so relevance in this setting alone often surfaces a related but unusable candidate. To address this, we propose M-SQE, a post-retrieval Multilingual Skill Quality Estimation framework that scores candidates via a Theory view for intrinsic quality and an Action view for task-grounded utility, unified into a domain-conditioned final score. We evaluate M-SQE across three skill-use domains: general, tool-use, and cultural tasks. Empirically, we build three-layer candidate skill pools mirroring today's ecosystem, where M-SQE's task success exceeds existing baseline's average by at least +3.5 points across three different retrievers. Particularly, M-SQE lifts the lowest-resource languages most (+12.9pp on Hindi and +5.6pp on Swahili) and achieves strong performance across all six culture regions, thereby moving agentic skill use toward linguistic and cultural equality.
cs.CL / 24 / 2609.18529
Machine Translation between English and Syriac (East Syriac Dialect) using Statistical Machine Learning
Abstract
UNESCO considers the Assyrian (Syriac) language an endangered language. Although Assyrians speak the language worldwide, the speaking population is uncertain (ranging from 500,000 to 1,500,000). Syriac is also one of the least studied languages in Natural Language Processing (NLP). Despite advances in Machine Translation (MT) over the past decade, the lack of publicly available corpora and the orthographic complexity of the Syriac script, specifically the Madnkhaya script, have left this language entirely ignored in the computational linguistics literature. This study develops the first phrase-based Statistical MT (SMT) model for English-to-Assyrian MT using the Moses framework. We created a dataset of 38,847 sentence pairs from the complete English and Syriac Bible, merging a pre-existing New Testament dataset with an Old Testament built from scratch through PDF extraction, using custom segmentation scripts and manual alignment review by three bilingual annotators. The Syriac side of the corpus undergoes diacritic removal and Byte-Pair Encoding tokenization to reduce orthographic sparsity before training. We trained and evaluated six models using different configurations and splitting-scheme ratios, language model order, distortion limits, and the inclusion of an Operation Sequence Model. The best-performing configuration achieves a word-level BLEU score of 23.54. Human evaluation by 11 native Assyrian speakers resulted in mean adequacy and fluency scores of 3.42 and 3.34 out of 5, respectively. These results are consistent with comparable low-resource SMT models trained on Biblical corpora for morphologically rich Semitic languages. The corpora, scripts, and trained model are publicly available, providing the research community with the first systematically curated English-Syriac dataset and a reproducible baseline for future MT and broader NLP work on this endangered language.
cs.CL / 25 / 2609.18533
A Probe Shift Is Not a Fairness Fix: The Limits of Representation Steering in Speech Models
Abstract
Automatic speech recognition (ASR) systems exhibit unequal error rates across speaker groups, motivating interventions on their internal representations. We ask whether speaker-linked attributes that are linearly readable from pretrained ASR encoders yield useful directions for reducing group word-error-rate (WER) gaps. Across Whisper-medium, HuBERT-large, and Wav2Vec2-large on Common Voice and the Speech Accent Archive, we probe every encoder layer for metadata-derived sex/gender, age, and native/accent labels; construct centroid and probe-derived directions; inject them at selected layers; and compare downstream probe trajectories with matched WER changes. Sex labels are highly decodable (best macro-F1 0.924--0.941), native/accent labels are also above chance (0.544--0.696), and age is weaker (0.354--0.397). Of 22 post-selected reruns, nine have 95% paired-bootstrap intervals entirely below zero, yet every absolute source-group WER reduction is below 0.7 percentage points. Conversely, a local target-class probe rate can rise from 8.09% to 99.87% while WER worsens. Linear readability is therefore neither evidence of causal use nor a reliable mitigation method. Our results motivate evaluating speech-bias interventions jointly at representation, propagation, and task levels.
cs.CL / 26 / 2609.18605
PACT: Can Enterprise AI Assistants Be Trusted Under Pressure?
Abstract
As corporate AI adoption continues to grow, enterprise-grade LLM agents are being deployed into sensitive contexts such as hiring, healthcare, and finance. In these contexts, compliance with rules specified in an agent's system context is a first-order legal concern. Currently, no evaluation framework systematically measures which LLM models tend to violate compliance rules, especially under pressure from a persistent user, a hurried manager, or circumstances where violation is convenient or attractive. We introduce PACT (Pressure-Applied Compliance Testing), a benchmark for rule-following under pressure in AI agents assisting employees in daily tasks across twelve regulated enterprise domains and forty-eight scenarios, each set in a realistic multi-turn conversation. Each benchmark item pairs a standing rule against a rule-violating shortcut, and applies a battery of pressures across different wordings and system-prompt modes. We construct PACT component by component under strict LLM-as-judge auditing to ensure samples are unambiguous, ungameable, and realistic enough to avoid eliciting evaluation-aware behavior. We use PACT to profile LLM compliance across six complementary metrics that create a holistic picture of an AI assistant's robustness under pressure and throughout multi-turn conversations, its transparency, and ability to correctly discern where a rule applies. We aggregate this profile into PACTScore, a reliability-weighted compliance rate over all items and modes. Our results across 22 common LLM models spanning multiple providers and sizes show substantial variability in compliance across models and metric dimensions. Even the strongest assistants mis-apply a rule on 6 to 10% of items, and ordinary user pressure raises the violation rate by 65% on average. PACT highlights compliance risks in LLM assistants, motivating guardrails and careful model selection.
cs.CL / 27 / 2609.18644
Fallacy Benchmarks Measure Scheme Recognition, Not Fallacy Detection
Abstract
Fallacy-detection benchmarks pair fallacy classes with a single "valid" or "none" class that takes everything data collection did not label as a fallacy. This construction is misleading: a classifier can learn cues that do well on this class without learning to tell a fallacy from a correct argument. We show that the low false-positive rates benchmarks report are an artifact of how the class is built, not evidence of detection ability. The most informative negative for a fallacy is a correct argument using the same argumentation scheme, and such arguments are at most a few percent of the valid class across the four benchmarks we examined. Evaluated on constructed scheme-matched negatives, false-positive rates rise from 16.6% to 58.9% on CoCoLoFa and from 5.7% to 62.0% on Reddit. That rate depends on how the negatives are written, so we also compare two conditions from the same pipeline that differ only in scheme identity. Classifiers label scheme-matched negatives as the source fallacy type 40.9 points more often than wrong-scheme negatives, which are instead identified as the scheme they actually use 85.9% of the time against 0.4% for the source type. The classifier has learned which scheme an argument uses, not whether it uses it correctly, and on the benchmarks' own test sets the two are indistinguishable. The same dissociation appears in three zero-shot LLM detectors that never saw these benchmarks, and the measurement is far lower on a negative class that was built deliberately. We release the items as Scheme Foils. A reported false-positive rate should not be trusted as a measure of detection until the valid class has been audited for scheme-matched coverage.
cs.CL / 28 / 2609.18649
DyMT-ESB: Dynamic Multi-Turn Evaluation of Social Bias in User-LLM Interactions
Abstract
Warning: This paper contains examples of stereotypes and social bias. LLMs are increasingly used in interactive settings by the general public, making the evaluation of model behavior in multi-turn conversational scenarios important for safety, including stereotyping-related harms. However, existing multi-turn social bias evaluations often rely on pre-specified or template-based user inputs that do not adapt to model responses and typically assume a fixed dialogue length in advance. In this paper, we study social bias dynamics in response-conditioned multi-turn interactions using a controlled evaluation protocol that generates follow-up user queries from the evolving dialogue history and allows evaluation over variable numbers of turns. Experimental results show that LLMs exhibit social bias even in coherent, response-conditioned multi-turn interactions, revealing late-emerging bias, non-monotonic bias patterns, and bias re-emergence. These results motivate evaluations that extend beyond fixed-turn, pre-scripted protocols. Our findings highlight the importance of analyzing social bias as a turn-level dynamic phenomenon.
cs.CL / 29 / 2609.18672
Selection Is Retrieval, Abstention Is Not: On-Device Tool Routing over 70 Korean-English Actions
Abstract
An AI assistant that calls tools makes two decisions on every request: which tool to invoke, and whether any available tool applies. In the usual design a single language model makes both, by emitting a call or by declining to emit one. On a device that has to answer without a server, the language model is what makes that design expensive, dominating both the latency and the memory of the router. The common alternative is to remove the model completely and rank the catalog of local actions with a retriever instead. That substitution is not symmetric across the two decisions. A retriever returns its highest-scoring candidate for every input and cannot signal that the catalog holds no valid action. Our earlier study found that constraining a decoder to a tool grammar repairs malformed output without improving the choice. What the substitution costs in each decision has not been measured. We evaluate the two decisions separately over 600 Korean and English requests and a catalog of 70 local actions. The router may also ask for a missing slot, reply, or delegate. Half the in-catalog requests reuse catalog vocabulary and half paraphrase it, separating lexical overlap from the action requested. Character 3-gram BM25 selects 162 of 164 lexically matched requests and 85 of 166 paraphrases. Restricting the candidate set to seven raises the paraphrase figure to a mean of 0.825 over five trials. No classifier over its score features separates in-catalog from out-of-catalog above 0.697 area under the curve, where the frozen encoder multilingual-e5-base reaches 0.806. Using that encoder for abstention alone keeps 376 of the requests local and misroutes 9 of the 150 needing delegation. Abstention, not selection, is where a neural component is required. A neural ranker improves every quality metric and is rejected on latency and memory rather than accuracy.
cs.CL / 30 / 2609.18677
Voice of Reason: Reinforcement Learning for Spoken Math
Abstract
Speech language models enable richer spoken interactions between humans and machines than cascaded systems, allowing access to paralinguistic information and lower latency. However, their accuracy on mathematical reasoning benchmarks has lagged behind those of text models. Reinforcement learning (RL) with verifiable rewards has been instrumental in extending text models' capabilities for solving complex problems and limiting hallucinations. In this work, we explore applying RL to the GLM-4-Voice speech model (Zeng et al., 2024) to bridge the gap between textual and spoken mathematical problem solving. We first adapt the model to the domain using supervised fine-tuning on synthesized spoken question-answering data. We then show that, even without extra reasoning tokens, RL improves the accuracy on GSM8K beyond levels previously achieved for speech models only with supplementary reasoning traces. When combined with existing streaming reasoning techniques, we show further gains to 74.8% free-form accuracy. This establishes a new state-of-the-art for mathematical spoken abilities with speech-native models.
cs.CL / 31 / 2609.18680
HearInContext: A Benchmark for Implicit Context in Speech Recognition
Abstract
Contextual ASR can benefit from semantic cues or from target words explicitly provided in the context. We introduce HearInContext, a Mandarin--English benchmark that pairs shared synthetic speech with assistant replies supporting different interpretations. The benchmark comprises 3,764 semantic test cases built around homophones. Implicit contexts exclude candidate words; explicit contexts name the target. No-context and unrelated-context controls measure the benefit of relevant history and sensitivity to irrelevant history. Context-capable models benefit from implicit cues but achieve higher target recall with explicit hints. Fine-tuning Qwen3-ASR-1.7B improves implicit-context target recall by 11.0 and 11.5 percentage points in Mandarin and English, respectively, while absolute CER/WER changes on AISHELL-1 and LibriSpeech remain below 0.1 percentage points. Gains extend to explicit conditions excluded from fine-tuning and to Mandarin hotword recognition on real recordings.
cs.CL / 32 / 2609.18697
Tracing individual knowledge trajectories in a changing field: the case of general relativity and gravitation
Abstract
Historians have reconstructed the twentieth-century transformation of general relativity and gravitation (GRG) at the field level and through individual careers, but connecting these scales requires a way to compare researchers with the changing field over time. We develop such a comparison, setting a researcher's publications and references against GRG field literature from the same, earlier, and later two-year periods. Building on Own Vocabulary and Embedding Density Estimation from our earlier two-case study (arXiv:2501.00391), we extend the analysis to the fifty most-published authors in a NASA/ADS corpus of about 180,000 GRG records (1911 to 2000) and add two citation-based measures, Referenced Vocabulary and Citation Identity. The four measures compare an author's written language, cited literature, semantic neighbourhood, and cited-authority configuration with the surrounding field. The earlier cases suggested that closer field-vocabulary alignment accompanies a denser semantic neighbourhood. Across the fifty authors this holds only partially. Written and cited vocabularies tend to move together, usually resembling later GRG literature as the field turned towards astrophysical and cosmological research. Semantic neighbourhoods more often lie where the field's publications were concentrated in earlier periods, while co-citation patterns follow no single temporal direction, and the two citation measures frequently place the same researcher differently despite drawing on identical reference lists. Individual trajectories can thus combine vocabulary tied to later field states with older semantic or citation structures, and these divergent cases mark patterns for closer historical investigation. The approach transfers to other fields with defensible corpus boundaries and adequate coverage of texts, references, and disambiguated author identities.
cs.CL / 33 / 2609.18720
LocQE: Principled Domain Adaptation for Localisation Quality Estimation by Leveraging Post-Edits
Abstract
Learned quality estimation (QE) models such as COMETKiwi are widespread and work well for general machine translation evaluation. However, they are known to struggle on unseen domains, limiting their performance in a real-world localisation context. We show that they are insensitive to some important factors in localisation, such as whether numbers are translated accurately, or even whether the correct number of spaces and punctuation are preserved in a translation. Further, a key capability for optimisation of machine translation is the ability of QE models to accurately rank different translations of a single segment, which suffers significantly from the domain transfer. In the absence of large-scale direct assessment data, we propose principled fine-tuning approaches to reduce the domain gap with even small amounts of post-editing data. Using a multi-task fine-tuning approach and a simple tokeniser intervention, we create a QE model which proves markedly better at distinguishing preferred post-edits from rejected initial translations in a localisation context. We show that preferences and artificial continuous scores stabilise each other, and argue that to calibrate metrics both in terms of their absolute scores and comparisons between translation of the same source, both types of signal are needed.
cs.CL / 34 / 2609.18772
Zero-Shot Cross-Lingual Recognition of Sign Language Handshapes
Abstract
Sign language processing advances rapidly for high-resource languages such as American Sign Language (ASL), yet most of the world's sign languages lack the phonological annotations new methods require. We present the first zero-shot cross-lingual framework for handshape recognition, transferring from ASL to Catalan Sign Language (LSC). Our approach leverages the decomposition of handshapes into five phonological features -- selected fingers, flexion, spread, thumb position, and thumb contact -- shared across both languages, to decode LSC handshapes from predicted features via a composite phonological distance metric. We evaluate three architectures (MLP, SL-GCN, SHuBERT) trained on two ASL corpora (PopSign, Sem-Lex) against a 37-handshape, single-signer LSC benchmark. Zero-shot transfer proves viable once recording-format disparities are harmonized, reaching 80.0% phonological feature accuracy and 54.5% expected handshape accuracy. Phonological decomposition thus offers a bridge for extending sign language technologies to low-resource languages without any target-language video training labels.
cs.CL / 35 / 2609.18804
Beyond frequency measures: Can contextual embeddings capture meaning change in scientific texts?
Abstract
Identifying technological trends is a core scientometric task, yet traditional frequency-based approaches struggle to capture substantial meaning shifts of domain-specific terms. We hypothesise that contextual embeddings can complement frequency dynamics to effectively track diachronic semantic change. We compare frequency and embedding-based approaches across Astrophysics and NLP corpora spanning from 2010 to 2024. Candidate terms are extracted using KeyBERT (utilizing SciBERT as its underlying language model) and filtered for significant frequency increases using Fisher's exact test. These terms are then evaluated for genuine semantic shift by domain experts to establish ground-truth labels. To quantify semantic drift, each term's contextual embedding ''clouds'' from the two discrete periods are compared using multiple metrics: cosine distance, average pairwise distance, Hotelling-type T 2 , and maximum mean discrepancy. Results indicate that frequency-based methods align slightly better with human judgments of ''trend-related terms'' than semantic metrics (Precision@50 of 0.62 vs 0.60 in Astrophysics). The two signals show a correlation of around 0.6. Several terms identified exclusively by embedding metrics (e.g., ''primordial black holes'') represent critical conceptual developments invisible to pure frequency analysis. These findings indicate that semantic metrics may capture complementary information, highlighting the value of integrating contextual embeddings into scientometric trend analysis.
cs.CL / 36 / 2609.18844
ReFigBench: Benchmarking Scientific Figure Reconstruction as Editable PowerPoint Artifacts
Abstract
Multimodal coding agents are expected to turn visual inputs into usable artifacts, and they act through a harness, the layer of tools, context management, and execution environment around the model. Existing evaluations often isolate short tool calls, API traces, or screenshot resemblance, and a low score under these proxies cannot say whether the model saw poorly, planned poorly, or was failed by its harness. We study scientific overview figure reconstruction, an agent task in which a source image must become an editable PowerPoint slide that preserves text, topology, layout, and native document structure. We introduce ReFigBench, a benchmark and evaluation framework built on 1,000 real overview figures retrieved from arXiv papers with full provenance. Coding agents from four model families reconstruct every figure under two workflows, direct code generation and a specialized PPTX workflow, and the strongest model runs inside two commercial harnesses, yielding ten configurations. Evaluation combines deterministic artifact checks, repeated automated scoring by judges from two model families, and blinded human comparisons. Perception remains a bottleneck that iterative rendering only partly repays. Whether workflow effort converts into quality depends on the model together with its harness, since the same model gains from the specialized workflow inside one harness and loses inside the other, and the harness shifts scores even under an identical direct prompt. The specialized workflow erases native connectors in every configuration, human judges still prefer its renderings in most matchups, and even the strongest agent falls short of the rubric ceiling. These results expose the tension between fidelity and editability as the central challenge for practical multimodal document agents.
cs.CL / 37 / 2609.18852
EviGen: Predictive Evidence Scaffolding for Verifiable Clinical Rationale Generation
Abstract
Longitudinal electronic health records (EHRs) capture years of patient history across notes, codes, labs, and procedures, and contain evidence needed to reason about likely clinical outcomes. However, comprehensive clinician review of these records is impractical, and LLM-based processing is costly and often unreliable, missing some relevant observations while hallucinating others. We therefore propose EviGen, a three-layer framework for verifiable clinical rationale generation that addresses these challenges. The first layer is a patient-conditioned retriever that uses learnable queries to find evidence predictive of, not just textually relevant to, a clinical outcome and ranks it by prediction attribution scores. The second layer is an LLM generator that consumes this ranked evidence as a scaffold to produce a clinical rationale grounded in the retrieved spans. The third layer is a process-supervised verifier that checks the generated rationale at the reasoning-step level, flagging unreliable claims. Across three medical prediction datasets, EviGen improves prediction performance and rationale faithfulness over full-context LLM and RAG baselines, and is preferred by clinical reviewers in a usability evaluation.
cs.CL / 38 / 2609.18861
PersonaPath: Towards Knowledge-Centric Personalized Learning Path Planning
Abstract
Adaptive learning systems commonly formulate learning path planning as Exercise-Centric (EC) recommendation, where the next step is inferred from item-level interaction logs. Evaluating goal-oriented guidance additionally requires explicit learner goals and curriculum-scale prerequisites: learners with similar exercise records may need different paths toward their targets. We therefore study Knowledge-Centric (KC) personalized learning path planning, where a planner must reason over learner profiles, mastery states, and prerequisite knowledge structures to decide which textbook, unit, and concept should be studied next. To support this setting, we introduce PersonaPath, a benchmark that pairs 2,000 fine-grained learner personas with a hierarchical knowledge graph of 347 textbooks, 1,751 units, and 4,092 concepts across 77 subjects. We evaluate representative LLMs on PersonaPath. Results show that even the strongest LLM reaches only a 29.5% final pass rate in Basic Education, and that the main bottleneck lies in adaptivity, where no model exceeds 44.7% in tailoring paths to individual learners.
cs.CL / 39 / 2609.18905
Structured Claim-Level Discourse Representations for Dense Health Narratives
Abstract
Health discourse in social media videos often contains densely entangled claims spanning multiple thematic aspects, stances, evidential frames, and rhetorical functions within short conversational spans. Existing approaches largely rely on coarse topic-level, sentiment-based, or stance-oriented representations that do not adequately capture this structure. Our analysis identifies an average of 13.22 atomic claims per minute, motivating richer claim-level discourse representations. We introduce a structured framework for claim-level discourse analysis in dense health narratives. Our framework models discourse through tuples linking atomic claims with thematic aspects, stance, and multidimensional pragmatic discourse attributes. To support this setting, we construct a benchmark spanning four health domains with 1,191 manually annotated claims from 60 videos. Using this framework, we evaluate automated structured discourse analysis under different discourse context settings. Results show that current LLMs achieve strong performance on thematic categorization and stance prediction, but struggle with high-dimensional pragmatic profiling. We also find that different discourse tasks benefit from different forms of contextual reasoning, suggesting that future systems may require task decomposition and specialized inference strategies.
cs.CL / 40 / 2609.18908
How Much is a Human Right Worth? ECtHR-NPD: A Benchmark for Predicting Non-Pecuniary Damage Awards
Abstract
Existing legal benchmarks cover diverse tasks, while continuous monetary remedies remain comparatively underexplored. We introduce ECtHR-NPD, to the best of our knowledge, the first benchmark for predicting non-pecuniary damage (NPD) awards at the European Court of Human Rights (ECtHR) from case information when no statutory formula or explicit calculation rule determines the amount. ECtHR-NPD contains 14,575 cases with case-level awards in nominal euros, chronological splits, and a protocol separating target construction from model input. We evaluate a battery of methods, including constant predictors, gradient-boosted trees, retrieval methods, fine-tuned encoder language models (LMs), prompted decoder LMs, and knowledge-augmented agents. Our results show that more sophisticated LM and agentic approaches do not consistently outperform the strongest feature-based baseline. All model families struggle to identify zero awards and to calibrate high-award predictions, with further degradation on the Challenging test view, making ECtHR-NPD a challenging testbed for current state-of-the-art open-weight and proprietary LMs.
cs.CL / 41 / 2609.18909
Beyond Outcomes: Dual-View Relational Learning for Efficient Agent Benchmarking
Abstract
Agent benchmarks are substantially more costly to evaluate than conventional LLM benchmarks. Benchmark compression is therefore a natural solution, yet existing methods primarily model redundancy in task--model final-score distributions, which is important in agentic evaluation. To address this limitation, we analyze large-scale trajectories and identify six complementary process signals that are systematically associated with final agent performance. To disentangle agent performance redundancy from a complete perspective, we propose DualViewEval, an agent benchmark compression method that jointly exploits outcome and process relations to learn an exact-size miniset and predict the full-benchmark scores. Across five agent benchmarks and five representative baselines, DualViewEval achieves the best results in all datasets. With only 20 tasks, it achieves $24\times$--$40\times$ compression on APEX-Agents and BFCL, reducing mean absolute error (MAE) by $14.5\%$--$28.2\%$ over the strongest competitors while improving Kendall's $τ$ by up to $7.2\%$ relative to EssenceBench on SWE-bench Verified. The selected minisets further reveal capability differences among different agents, providing compact and diagnostic feedback for efficient agentic model development.
cs.CL / 42 / 2609.18935
Long-Lived Characters, Local Inference: Incremental Memory Maintenance for Game NPCs
Abstract
A game character should not have to reread its entire life before every conversation. For locally deployed language-model characters, however, revising a few memories can invalidate a long reusable prefix. The resulting preparation cost competes with both foreground dialogue and the maintenance of other characters. This matters especially when dialogue feeds game-defined actions and value judgments: a fluent but incorrect account of who owns an item, or whether a transfer has already happened, can corrupt the input to otherwise deterministic rules. We study incremental memory maintenance for long-lived game NPCs in a quantized Qwen hybrid recurrent-attention model. Our runtime removes superseded attention KV entries, computes replacement records at the true sequence tail, and preserves the continuing recurrent state and unchanged KV. Existing local experiments combine multi-update dialogue replays, fixed-input placement ablations, and attention diagnostics. Independent block composition weakens query-conditioned memory selection without a uniform chunk-initial attention collapse. True-tail updates preserve important current-state and historical bindings across eight scripted maintenance rounds; a placement case recovers the full-refill quantity in three reconstructions, while slot-preserving alternatives repeat a double-subtraction error. Attention-distribution proximity alone does not explain these semantic differences. The results motivate treating a character's inference state as a maintained, history-dependent resource, rather than only a disposable encoding of its latest memory text.
cs.CL / 43 / 2609.19093
Reporting Practice Matters: The Impact of Reference Choice on Chest X-ray Report Evaluation
Abstract
Radiologists follow heterogeneous reporting practices. Two radiologists examining the same image and identifying the same clinical findings might nevertheless compose superficially distinct reports, varying in terminology, shorthand, formatting, and level of detail. These variations in reporting norms represent an under-appreciated obstacle in efforts to evaluate AI-based radiology report generation (RRG) models, where machine-generated reports are typically assessed based on their concordance with human-generated references. In this paper, we quantify the sensitivity of established evaluation metrics to variations in reporting practices, revealing impacts large enough to alter the rankings of models. We introduce a radiologist-informed taxonomy of variations in radiology reporting practice and a method (ReRef) that rewrites reference reports along the axes of our taxonomy while preserving clinical interpretation. For instance, when comparing the performance of nine RRG models on MIMIC-CXR using RadCliQ-v1, condensing the discussion of normal findings in the reference reports causes Libra to drop from first to second place while CheXOne rises from third to first. Our results suggest that many current metrics fail to decouple clinical interpretation from conformity to reporting practices and that choosing the ``right'' references that accurately reflect the desired reporting practices can be important in practice. To support future research, we release MIMIC-CXR-Ext-ReRef, a radiologist-validated dataset of 120 (original, alternative) reference report pairs derived from MIMIC-CXR.
cs.CL / 44 / 2609.19101
Monitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluations
Abstract
As models scale, reward hacking becomes more frequent, more sophisticated, and more consequential. Does it leave a telltale signature in model representations? This work analyzes how reward hacking is represented internally in frontier open source LLMs, and how those representations can be used to understand and discover the range of hacking behaviors a model displays. In particular, we find that simple difference of means vectors coherently represent reward hacking in Kimi K3, GLM 5.2, and Qwen 3.8 Max across a variety of behaviors in common evaluations. Despite their simplicity, these vectors are both generalizable and interpretable, and we can use them to reliably detect reward hacking. We first evaluate reward hacking in commonly reported benchmarks like DeepSWE and SWE-bench, finding that models reward hack excessively in these environments; GLM 5.2 hacks in 57.2% of rollouts on DeepSWE and in 73% of rollouts on SWE-bench. Catching these requires monitors; LLM monitors are effective, but expensive detectors. We show that DoM vectors are similarly effective but virtually free, catching 3.1% more hacks in Kimi K3 and 7.9% fewer hacks in GLM 5.2 on DeepSWE at a monitor matched false positive rate. DoM vectors run on the chain-of-thought also predict reward hacks in the model's subsequent actions, meaning we can run them online and catch potential hacks before they occur. Finally, we analyze probe-hits that LLM monitors do not catch and discover other undesirable behaviors, as well as show transfer to finding hacks in non-SWE evaluations. Together, these results provide evidence that simple, white-box methods can be used to scalably study and monitor reward hacking behaviors in frontier open source models
cs.CL / 45 / 2609.19113
Playing log(N)-Questions over Wikipedia Abstracts: Communication Efficiency Between Paired Frontier Models
Abstract
We evaluate six frontier language models on the two-agent $\log(N)$-Questions game. A questioner sees $N$ Wikipedia lead paragraphs and must identify a secretly chosen target using exactly $\log_2 N$ yes/no questions. An answerer sees only the target and the question, and replies with one word. Both roles run on the same provider, so the game measures how well a model communicates with itself across an information asymmetry. We run 408 games over document sets of 4 to 1024 paragraphs at a total API cost of \$363. One model finishes well behind the others: Claude Opus 5 wins 28 of 68 games, against 45 to 56 for GLM-5.3, GPT-5.6 Sol, Grok 4.6, Gemini 3.8 Flash and Kimi K3. The leading five are only marginally separable. Pooling those five, win rate declines with set size at $r=-0.973$ and is fit by a single per-round reliability parameter. The form is $\text{win}=p^{\log_2 N}$ with $p=0.928$. Losses divide into answer errors and discrimination failures in roughly equal measure, and models almost never name a document their own evidence excludes. Every unanimous answer error from the weakest model was inspected: 32 of 34 are ``No'' answers, on properties stated in the document's first sentence, under an instruction that explicitly warns against defaulting to ``No''. Information per question, estimated from answer balance, correlates with win rate at $r=+0.88$. The only two models to extract a full bit per question are the only two that partition on document titles, a strategy absent below $N{=}32$ and used in a quarter of questions above it. Reasoning-token expenditure varies $4.5\times$ across models with little relation to success, and the trace grows as the candidate set shrinks without a matching gain in reliability.
cs.CL / 46 / 2609.19134
ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments
Abstract
Scientific code repositories encode decades of human knowledge in executable models, methods, and tools. Yet fragmented toolchains, implicit domain conventions, and specialized correctness criteria make this knowledge difficult to convert into reliable learning experience-a challenge we call the scientific experience bottleneck. We introduce ScienceIDE, infrastructure for turning the world's scientific code into programmable environments for scientific agents. Guided by expert-defined scientific cases and acceptance criteria, agents transform repositories into executable environments that support task generation, execution, and scientific verification. These environments provide a shared foundation for supervised fine-tuning, reinforcement learning, and evaluation. Using verified interaction trajectories, we train PhAI-IDE-72B, PhAI-IDE-9B, and PhAI-IDE-4B. The model family shows gains in held-out scientific-code repair and across selected general-purpose benchmarks in code, reasoning, and knowledge, providing evidence of positive transfer from scientific experience to broader capabilities. ScienceIDE lays the foundation for an integrated workspace for agent learning and scientific practice, making humanity's scientific software a shared substrate for developing scientific intelligence. Code: https://github.com/aitofound/ScienceIDE
cs.CL / 47 / 2609.19145
Objective vs. Search: Decomposing What Makes a Good Tokeniser
Abstract
Two dominant tokenisation algorithms are used by modern language models: byte-pair encoding (BPE) and UnigramLM. These differ along two orthogonal axes: their optimisation objective (compression vs. log-likelihood) and their search procedure (bottom-up merging vs. top-down pruning). Existing comparisons confound these axes, making it unclear whether their observed differences stem from what is being optimised vs. how it is being optimised. We disentangle the two by introducing two new tokenisation algorithms that complete this 2x2 design space: BottomUpLL, a bottom-up likelihood-based tokeniser, and TopDownComp, a top-down compression-based tokeniser. We train language models with tokenisers produced by each algorithm, varying: model size, vocabulary sizes, and domain (English-only vs. multilingual). Evaluating models on bits-per-byte, we find that the search procedure -- not the objective -- is the dominant factor: bottom-up tokenisers consistently achieve lower bits-per-byte in most settings. Evaluating models on the BLiMP task, however, shows no consistent relationship between design choice and performance. Overall, our results disentangle the effect of tokeniser design choices on language modelling performance, offering concrete guidance for their more principled construction.
cs.CL / 48 / 2609.18690
RankGround: Efficient High-Resolution GUI Grounding via Lightweight Reranker-Guided Crop Selection
Abstract
Graphical User Interface (GUI) grounding is a fundamental perception task for multimodal agents, enabling them to interpret natural language instructions and interact with digital interfaces. Existing methods face a fundamental trade-off between accuracy and efficiency: direct full-image inference often fails to capture small or visually similar UI elements, while multi-crop strategies improve localization at the cost of multiple expensive Vision-Language Model (VLM) calls per query. To address this challenge, we propose RankGround, a two-stage framework that achieves accurate GUI grounding with a single VLM call per query. Central to our approach is GroundRanker, a lightweight multimodal reranker that identifies the most promising crop from a dense candidate set. Because no off-the-shelf ranking dataset is available, we construct ranking supervision data from existing grounding datasets. A strict containment criterion and boundary-aware positive augmentation improve alignment and spatial coverage in cluttered layouts. GroundRanker is then trained with a two-stage curriculum: a pointwise objective first learns coarse containment, and a listwise objective refines subtle semantic and spatial distinctions among visually similar crops. Experimental results show that RankGround consistently outperforms strong baselines while reducing computational cost. It achieves 1.4 times faster inference and improves localization accuracy by 5.5% on average over the second-best method across all backbones and screen scales, establishing a new state of the art in both efficiency and precision for GUI grounding.
cs.CL / 49 / 2609.19143
PANORAMA: Panoptic Grounded Captioning via Mask Proposal Selection
Abstract
Intelligent systems that act in the world require image understanding that is both comprehensive and spatially grounded. Current vision-language models (VLMs) can generate fluent and detailed image captions, but reliably associating them with image pixels remains challenging. Existing methods that combine dense captioning with pixel-level grounding often produce either incomplete descriptions or inaccurate segmentation masks. We study this problem through panoptic grounded captioning, a task that requires a VLM to describe both foreground objects and background regions while grounding each referring phrase with pixel-level masks. We make three contributions. First, we introduce PanoCaps, a human-annotated benchmark constructed from panoptic segmentation datasets. It provides dense captions with near-complete pixel coverage and image-text alignments at the entity level, supporting both training and evaluation. We further propose a phrase-mask matching protocol and a generalized Panoptic Quality (gPQ) metric that jointly evaluates textual and mask agreement. Second, we formulate phrase grounding as selection from a phrase-conditioned pool of mask proposals and introduce PANORAMA, a VLM that conditions a pretrained segmenter on contextualized phrase representations to obtain candidate masks and learns to select those corresponding to each phrase. Training this interface jointly with caption generation enables PANORAMA to produce high-quality masks while allowing each phrase to refer to a single region or multiple instances. Third, PANORAMA achieves the best overall grounding on PanoCaps and matches or exceeds specialized models across several pixel-level grounding tasks. Experiments show that our method produces precise entity-level segmentations while maintaining detailed, mask-consistent captions. Code, data and models are available at https://www.di.ens.fr/willow/research/panorama/.
cs.CL / 50 / 2609.18341
Understanding AI Provider Recommendations in Local Service Markets
Abstract
When someone asks an AI assistant which doctor to see or which firm to trust with their savings, the answer is a referral. We audit AI provider recommendations in four registry-backed service domains across the 100 largest U.S. metropolitan areas, matching every recommendation against the official registry for its domain (Medicare clinician and facility records, and SEC adviser disclosures), under three conditions: an open-weight model, a proprietary model without web search, and the same proprietary model with search. Without search, both models largely fabricate recommendations in the domains the web covers thinly. Only 4% of the open-weight model's recommended doctors and 11% of the proprietary model's match a clinician in the queried city, and the open-weight matches are name coincidences: its matched clinicians are no likelier to be primary-care doctors than names drawn at random from the registry. With search, 64-71% of recommendations in the same domains match a real provider. Search also changes who is recommended. Without it, recommended advisory firms carry SEC misconduct disclosures at 3.6 times the registry base rate, even after adjusting for firm size; with search, significantly below it. Restaurants, where quality and visibility are separately measurable, show a 3-5x review-count premium but a rating premium of at most a tenth of a star. Finally, search largely removes the metro-size penalty: without it, real recommendations concentrate in the largest metros; with it, match rates are similar across metro-size terciles. Whether an AI referral is trustworthy depends strongly on its retrieval configuration rather than on the underlying model alone, yet an answer produced without retrieval often carries no sign that its recommendations were never verified.
cs.CL / 51 / 2609.18729
"If I Had to Buy Just ONE: Galaxy S26 Ultra": Auditing AI-Generated Product Recommendations
Abstract
Consumers increasingly use AI chatbots for advice on what to buy. With companies like OpenAI and Google monetising their AI through advertising, this raises difficult questions about the bias and impartiality of such advice. In response, we conduct an AI audit of popular chatbots using real commercial-advice queries. First, we curate a dataset of 2,528 real commercial-advice queries (ConsumerQ). Then, we evaluate 1,536 responses to product queries from popular AI chatbots: ChatGPT (chatbot and API), Google Gemini (chatbot and API), and Google Search (AI Overviews). We find that ChatGPT expresses a first-person product preference in 79% of product-recommending responses, compared with 7% for Gemini and 2% for AI Overviews, while the products recommended often change across repeated requests. Displayed sources vary strongly: for the same query, the ChatGPT and Gemini interfaces share only 5.4% of domains on average, with no domain in common in 76.7% of comparisons. APIs provide a different view from their corresponding interfaces, with mean domain overlaps of 12.0% for ChatGPT and 14.8% for Gemini, and also differ in the types and layers of source information they expose. Our findings show that neither isolated responses nor API observations can be assumed to represent the commercial advice consumers encounter. Independent audits of AI-mediated commercial advice should therefore account for repeated responses, consumer-facing conditions, and the source layer being observed.
cs.CL / 52 / 2609.18154
PageRecall: Measuring Page Selection in Literature-Grounded Question Answering
Abstract
We describe our system for LitTraceQA (GroundLM @ EMNLP 2026): given a research question, retrieve the relevant papers from a pool of 27,487, cite the page and the table or figure where the answer lives, and answer in a requested format. Our main finding is that evidence grounding is limited by retrieval, not by reading. The page selector put the annotator's page, which we call the gold page, in front of the model that locates evidence only about half the time (52.6% gold-page recall), while that model, given the page, cited the right one in 45 of the 48 locators it emitted (94%). When the page was missing it rarely said so: of 45 such cases it returned nothing 14 times, a wrong page 24 times, and a correct page 7 times, so the pipeline failed quietly almost twice as often as it failed visibly. Since the failure was that the right page was never shown, the fix is to stop choosing: each retrieved paper fits in the model's context, so we show it whole. Page ranking survives only as a fallback inside papers too long to fit, which no test-split paper was, and gold-page recall reaches 100% on the papers we can parse. Separately, questions that identify their target by position rather than content, such as "the first author of the 24th reference", are served by parsing rather than retrieval: we resolve the bibliography into an addressable list, which also supplies identifiers the evidence metric scores. The final system scores 0.762 paper $F_1$, 0.441 evidence $F_1$ and 0.920 multiple-choice accuracy on the held-out test split. Because the pipeline depends on a closed model without seed control, we release a harness that verifies the paper's central claims against committed artifacts.
cs.CL / 53 / 2609.18470
Divide and Conquer: Mixture-of-Bottleneck Experts in Informative Ordinal Space for Video-based Multimodal Sentiment Analysis
Abstract
Video-based Multimodal sentiment analysis (MSA) must handle information from text, audio, and image sequence in human speaking videos, yet current methods often fail to integrate modalities with task awareness. Most models treat video sentiment prediction as a single task, overlooking its ordinal nature, and their fusion strategies struggle to capture diverse unique and synergic cues across modalities. To address these limitations, we adopt a divide-and-conquer perspective by reformulating MSA as an ordinal regression problem and decoupling it into polarity recognition and intensity prediction. Driven by information theory, we introduce a Mixture-of-Bottleneck (MoB) framework that assigns different latents to polarity- and intensity-specific experts for different modalities. With the learning of information bottleneck, each expert learns compact and task-relevant representations while filtering out redundancy and noise. A multimodal bottleneck routing fusion module then fuses these expert latents with hard mining strategy, guiding the prediction in the ordinal sentiment space. Extensive experiments on 4 MSA datasets and 4 language models show that MoB effectively leverages informative latents from diverse modalities and captures general sentiment structure. Beyond stronger performance, MoB comprehensively captures fine-grained intra- and inter-modal dynamics, enabling more trustworthy localization of nuanced video sentiment signals.
cs.CL / 54 / 2609.18748
TeleAntiFraud 2.0: A Refreshable, Profile-Grounded, and Audio-Based Benchmark for Telecom Fraud Detection
Abstract
Telecom fraud scripts evolve rapidly and are often designed to resemble routine service conversations, creating two key requirements for audio-based telecom-fraud evaluation. First, benchmarks must incorporate newly observed scam patterns without overwriting previously established test sets. Second, they must distinguish fraud from lawful, near-domain calls rather than relying on topic-separated negative examples. We present TeleAntiFraud 2.0, constructed with our Mixed-Tree Anti-Fraud Generation Pipeline and evaluated under a monthly frozen evaluation protocol. The pipeline transforms online fraud-case abstracts into profile-grounded scenarios, expands them through mixed-tree generation, realizes fraud and non-fraud dialogue paths under shared contexts, renders validated dialogues as role-matched speech, and freezes the resulting audio, labels, prompts, manifests, and provenance records for each monthly evaluation set. Each frozen set contains 900 Chinese calls, comprising 600 fraud and 300 near-domain non-fraud cases. Controlled text experiments show that three classifiers achieve perfect macro-averaged F1 (Macro-F1) when evaluated against unrelated or ordinary negatives, but drop to 0.65-0.68 with near-domain sibling negatives. Full-set audio and automatic-speech-recognition plus large-language-model (ASR+LLM) evaluations further reveal class-prior shortcuts, prediction collapse, and snapshot sensitivity. Together, these findings establish near-domain construction and collapse-aware reporting as core requirements for evaluating audio-based telecom-fraud models under realistic confusable conditions. The accompanying research artifact includes the construction code, evaluation scripts, manifests, and documentation. Our dataset and code are available at https://anonymous.4open.science/r/TeleAntiFraud-2_0-EEB2/.
cs.CL / 55 / 2609.18766
FRAUDSkill: Structured Frozen-Weight Skill Optimization for Audio Anti-Fraud Detection
Abstract
Large audio-language models have shown promise for anti-fraud detection by directly processing speech and reasoning over fraud-related evidence. Their deployment, however, requires predictions to follow a predefined label space and a structured decision protocol consisting of service-scenario identification, fraud detection, and conditional fraud-type classification. Existing fine-tuning and prompt-based approaches typically encode task knowledge, constraints, and decision rules into model parameters or manually maintained prompts, making them difficult to adapt as fraud patterns and labeling policies evolve. To this end, we propose FRAUDSkill, a structured frozen-weight adaptation framework that leaves the underlying audio-language model unchanged while optimizing an external layer of skill programs, route-specific policies, and decision rules. We further combine structured output control with validation-guided multi-path inference to ensure protocol-compliant predictions. On the TeleAntiFraud benchmark, FRAUDSkill achieves 73.50% Macro-F1, outperforming the shared frozen-model baseline by 31.96% while reducing invalid outputs to 1.94%. Extensive experiments demonstrate that external skill optimization provides an effective and adaptable solution for structured audio anti-fraud detection without modifying the underlying model. The source code is available at https://anonymous.4open.science/r/FRAUDSKILL-114514.
cs.CL / 56 / 2609.18009
G-Mamba: Sparse Graph-Guided Mamba for Audio-Visual Speech Enhancement
Abstract
Lightweight audio-visual speech enhancement (AVSE) models face a critical trade-off between computational efficiency and cross-modal alignment accuracy. While simple concatenation lacks relational expressiveness, dense cross-attention incurs computational overhead and is prone to unreliable cross-modal correspondence under strong acoustic interference. We propose Sparse Graph-Guided Mamba (SG-Mamba), a lightweight AVSE framework that integrates a sparse heterogeneous graph with a linear-complexity Mamba backbone. The graph explicitly models modality-specific relations through content-adaptive attention and cross-frame audio-visual connections, while Mamba captures long-range temporal context. We further introduce an audio skip connection to preserve spectral detail without sacrificing noise suppression. Evaluated on LRS3, SG-Mamba achieves competitive or superior performance against strong lightweight baselines and reaches 13.091 dB SI-SDR under noise-only condition. It also remains robust in cluttered multi-speaker conditions with a competitive cost of 3.45 G MACs (or 6.90 G FLOPs). Results on VoxCeleb2 further suggest that explicit structural priors improve robustness, generalizability, and computational efficiency in lightweight AVSE.
cs.CL / 57 / 2609.18041
Correlation-Guided Encoder Selection for Multi-Encoder Large Audio-Language Models
Abstract
Multi-encoder fusion extends Large Audio-Language Models (LALMs) beyond speech-centric recognition, but selecting encoders via intuition or exhaustive search often introduces redundant representations and inflates an already constrained compute budget. We propose CUES (Correlation-gUided Encoder Selection), a lightweight heuristic that estimates complementarity through task- and category-level Pearson correlations between encoders' performance profiles, scoring a candidate set from single-encoder evaluations alone--without fusion training during selection. Evaluated on the XARES-LLM benchmark with a frozen SmolLM2-135M backbone (LoRA-adapted) via five-fold cross-validation, CUES consistently identifies the same configuration per track from held-out development splits alone, without using test data for selection. For the broad Track~A suite, CUES selects a cross-family trio (Whisper-medium, mHuBERT-147, and Dasheng-base), achieving a 4.3% relative gain over Whisper-medium (0.771 vs. 0.739). For Track~B text generation, it re-anchors on a focused, speech-only pair (mHuBERT-147 and WavLM-base-plus) and actively abstains from adding a divergent encoder, outperforming mHuBERT-147 by 6.3% (0.589 vs. 0.554). Rather than a failure to scale, this divergence is consistent with a diversity--interference trade-off that CUES navigates per track from correlation signals alone: across the evaluated pool, added cross-family diversity tends toward an inverted-U on broad audio tasks but toward steady degradation on text generation, which favors a focused, speech-anchored set.
cs.CL / 58 / 2609.18565
Variational Quantum Transformer Architecture for Synthetic Language Generation
Abstract
We propose a compact NISQ-compatible quantum transformer architecture for synthetic QNLP sequence modelling. The model preserves the autoregressive next-token interface of a classical transformer, but replaces attention and feed-forward sublayers with variational quantum encoder blocks, connector circuits, decoder blocks and a direct two-qubit measurement readout. Token contexts are angle-encoded into small quantum registers, processed by parallel variational heads and encoder integration circuits and conditioned through decoder ancillae to produce a distribution over a four-token vocabulary. We evaluate several architecture variants on deterministic and lexicographic grammar-generation tasks against a compact classical transformer baseline. The quantum models are trainable end-to-end and learn nontrivial grammar structure, including perfect deterministic generation in individual runs and high lexicographic validity in the strongest variant. The classical baseline remains more accurate and stable and the quantum models are sensitive to initialization. The contribution is therefore not a claim of quantum advantage, but a concrete architecture and evaluation of transformer-inspired QNLP sequence modelling under near-term quantum constraints.
多智能体系统 (cs.MA)
3
cs.MA / 1 / 2609.18223
ABM-SIRTEM: A Hybrid Agent-Based and Epidemiological Model for Pandemic Response
Abstract
The COVID-19 pandemic has had profound impacts on global health, social structures, and economies. It disproportionately affected lower socioeconomic groups and those reliant on interaction-based jobs. Regulatory bodies faced the challenge of designing policies that preserve public health while limiting disruption to economic stability and productivity. Epidemiological models such as SIR and agent-based models (ABMs) have been used to study disease dynamics and the socioeconomic impacts of disease and interventions. Population-level models often simplify individual heterogeneity, while detailed ABMs can become computationally expensive as the numbers of agents and interactions increase. We propose ABM-SIRTEM, a hybrid model that incorporates occupation categories, economic productivity, and welfare at the individual level while dynamically modeling compliance with government interventions. We calibrate the model against historical positive and negative test counts from four U.S. states and examine the resulting compliance dynamics. This framework provides a basis for studying the interaction between disease spread and socioeconomic behavior in pandemic-response planning.
cs.MA / 2 / 2609.18929
Social Laws for Multi-agent Coordination in Stochastic Environments
Abstract
In multi-agent environments, coordinating agents to prevent interference and ensure robust individual performance is a critical challenge. Previous research on social laws for multi-agent systems has primarily focused on deterministic, goal-based settings. This paper extends the concept of social laws to stochastic, reward-based environments, proposing a formalism for defining and verifying their robustness under various conditions. We introduce the notion of $α$-robustness, a measure of the guaranteed utility each agent retains while pursuing its optimal single agent policy, assuming all agents obey the social law. We then present an approach for robustness verification of social laws in stochastic settings, based on a reduction to solving a series of Markov decision processes. Empirical evaluations on toy environments illustrate the potential of our framework.
cs.MA / 3 / 2609.17797
Set-membership localization of intermittent RF sources using a fleet of collaborating UAVs
Abstract
This paper proposes a set-membership approach (SMA) to localize radio frequency (RF) sources observed by a collaborating fleet of Unmanned Aerial Vehicles (UAVs). Considering frequency-separable RF transmitters with intermittent and periodic emission patterns, %and unknown but bounded periods the SMA evaluates set estimates of the source locations and a set free of sources. Simulation results show that SMA outperforms a Bayesian baseline approach in terms of localization accuracy and convergence speed.
软件工程 (cs.SE)
8
cs.SE / 1 / 2609.17959
A11yLTLNav: Automatic Detection of Accessibility Navigation Failures
Abstract
For blind and low-vision (BLV) screen-reader users, a website that appears accessible in a static snapshot can become difficult or impossible to navigate once interaction begins. Yet, most automated accessibility checkers miss failures involving focus, interface state, and accessible feedback across interactions. We present A11yLTLNav, a property-based approach for automatically detecting accessibility navigation failures. Through a structured review of prior research, we organize accessibility navigation failures into a failure taxonomy and formalize a browser-observable subset as executable Linear Temporal Logic properties over action-state traces. A11yLTLNav combines random keyboard exploration with runtime property monitoring to detect these failures during interactions. We evaluate A11yLTLNav on 31 generated websites based on real-world websites and tasks. It reported 309 accessibility failures, of which 274 were confirmed, achieving 88.7% precision and identifying more confirmed failures than the comparison checkers. Our results show that A11yLTLNav transforms accessibility knowledge into reusable checks of interface behavior over time.
cs.SE / 2 / 2609.17844
Detecting Argument-Swap Bugs Using Context-Enhanced Code Representations
Abstract
Names of source code elements convey rich semantic information and have been widely used in software engineering tasks such as bug detection, code completion, type prediction, and code classification. Prior studies exploit lexical similarity between method arguments and formal parameter names to detect bugs caused by incorrectly ordered arguments, typically relying on establishing mappings between method calls and their corresponding definitions. However, such mappings are often difficult to obtain in dynamically typed languages like Python. In this paper, we present BugProbe, a learning-based approach for detecting incorrectly ordered arguments in Python method calls that does not require call-to-definition mappings. Our approach leverages multiple sources of contextual information, including local context and argument usage context, and combines name-based similarity with machine learning to construct expressive representations of method arguments. We collect a new dataset of 132,739 Python source files from the top-1,000 starred GitHub repositories, yielding 3,371,244 synthetic training examples, and contribute a curated benchmark of 55 real-world argument-swap bugs manually verified from commit histories. We evaluate our approach on this dataset and show that it achieves high accuracy and consistently outperforms a state-of-the-art baseline across standard evaluation metrics. These results demonstrate that effective detection of argument-ordering bugs is possible without relying on explicit call-to-definition resolution, making the approach well suited for dynamically typed language settings.
cs.SE / 3 / 2609.17993
Experimental Settings in LLM-Based Program Repair: A Study of Inputs, Tool Access, Feedback, and Validation
Abstract
Evaluations of automated program repair (APR) systems commonly report the benchmark, the number of repaired defects, and the tests used for final patch validation, but these items no longer fully specify the repair task presented to a system. Recent LLM-based systems differ in the information supplied before repair, the repository and testing operations permitted during repair, and the feedback returned after unsuccessful attempts, allowing the same benchmark to instantiate substantially different repair tasks ranging from localized patch generation to repository-level diagnosis and iterative repair. We present a framework for explicitly specifying the experimental settings associated with reported APR results. We analyze reported experimental settings from systems evaluated on Defects4J and SWE-bench and characterize each result by its task unit, fault-localization assumptions, initial input, tool access, repair-time feedback, final validation, and resource budget. Our analysis shows that benchmark identity alone is insufficient to reconstruct the evaluated task or determine the appropriate scope of comparison across reported repair rates. We therefore introduce a machine-readable schema for specifying each experimental setting to improve reproducibility and make the scope of cross-system comparisons explicit.
cs.SE / 4 / 2609.18291
Relationally Guided Use Case Modeling with LLMs
Abstract
Use case flows are important elements of use case modeling because they support downstream software engineering activities, including requirements analysis, architectural and detailed design, and test case generation. However, constructing them manually is costly and expertise-intensive, while existing automated approaches still struggle to preserve semantic consistency, control-flow logic, data-flow logic, and the intended system boundary, especially when identifying branch points and generating alternative flows. To address this problem, we propose FlowGen for complete use case flow construction. FlowGen uses LLM-based Semantic Information Processing (SIP) to extract semantic elements, constructs a Semantic Relational Graph (SRG) encoded by an enhanced R-GAT for basic flow generation (BFGen), and further supports branch point prediction through BPP and branch-conditioned alternative flow generation through AFGen. Evaluations on 13 public and 7 industrial datasets show that FlowGen consistently outperforms competitive baselines in all three core components. In particular, BFGen improves over the best baseline by 14% in Precision, 7-25% in Recall, 11-30% in F1, and 10-19% in AUC; BPP improves Precision by 30-110%, Recall by 33-91%, and F1 by 32-117%; AFGen improves Precision by 8-23%, F1 by 5-18%, and AUC by 0.6-2.5%. Moreover, we validate the effectiveness of the LLM-based SIP module and the attention preservation factor in BFGen, analyze the impact of requirement completeness on BFGen, and examine how different scopes of branch-related context affect AFGen.
cs.SE / 5 / 2609.18298
A Study of the Reliability of Agentic AI-Generated Programs
Abstract
Agentic-AI based software development offers the promise of faster completion of the software, greater programmer efficiency, and more reliable code. The question is how can we verify these claims in an objective way? In this project, we attempted to answer this question based on three practices. First, we applied a typical best-practices agentic AI workflow for software development. Second, our target programs were ten well-known, release-quality human-written Linux utility programs so that we could compare the AI-generated code against a concrete ground truth. Third, we based our measure of reliability on a widely used testing technique, fuzz random testing. For this testing, we used both classic black box, generational testing and more modern coverage guided (gray box, mutational) testing using AFL++. We found that the AI-generated versions of the utility programs were typically as reliable - often more reliable - than the latest human-generated versions of these programs. While the AI-generated versions did have some failures, they were less common than the code from the standard repositories. Interestingly, the AI-generated code was less likely to have failures such as memory errors (such as buffer overflows) but more likely to have hangs such as infinite loops. In addition, we verified that generating robust and reliable software using agentic AI requires careful practice and human supervision. The quality of the code is highly dependent on the prompts and skills used, and how the human directing the process responds. We also demonstrated that using agentic AI workflow for software development (with its prompts and skills) can become a specification of the code that leads to cost-effective sustainability of the software.
cs.SE / 6 / 2609.18391
From Component Snapshots to Lifecycle Traces: Agent-Based Software Composition Analysis
Abstract
Software supply-chain security requires accurate identification of third-party components and an understanding of how they evolve from development to execution. Existing software composition analysis (SCA) approaches examine manifests, build environments, release artifacts, containers, or runtime states, but typically produce only stage-specific views of software composition. As dependencies are resolved, removed, repackaged, and transformed across lifecycle stages, a single snapshot cannot capture both where a component originates and where it ultimately ends up. Combining snapshots from multiple stages still leaves their cross-stage relationships unresolved. We present SCA-Agent, an agent-based approach to lifecycle-aware SCA that reconstructs evidence-backed component lifecycle traces across Code, Build, Release, Deploy, and Runtime. SCA-Agent adaptively explores project-specific analysis paths, gathers stage-specific evidence, and correlates observations across stages to recover component identities, versions, introduction paths, propagation relationships, and final lifecycle states. We evaluate SCA-Agent on 105 real-world projects from the Java, JavaScript, and Python ecosystems. SCA-Agent achieves the highest component detection F1 across all lifecycle stages and ecosystems. For vulnerability exposure assessment, it reaches an F1 score of 96.69%, exceeding the best traditional SCA tool by 18.76 percentage points. These results show that lifecycle-aware SCA supports traceable component provenance and more accurate software supply-chain risk assessment.
cs.SE / 7 / 2609.18801
Evaluating the Health of Open-Source Smart City Platforms
Abstract
To manage the complexity of smart cities, a variety of smart city platforms (SCPs), both proprietary and open-source, have been proposed. Typically acting as middleware between IoT devices at a lower layer and smart services at a higher layer, these platforms simplify the management of smart cities by reifying a variety of requirements common to different municipalities. Open-source software (OSS) is typically free of charge, making open-source SCPs an attractive option for municipalities with smaller budgets that still wish to improve efficiency and quality of life for residents and visitors. Furthermore, using an OSS SCP promotes digital sovereignty, since municipalities can control where the data will reside and will be less prone to vendor lock-in. A lack of support for OSS hinders its adoption by municipalities' IT teams, often with scarce technical resources. An active (aka "healthy") ecosystem, including well-orchestrated developers and users, the availability of installation and user manuals, non-vulnerable and documented code, test batteries, and other artifacts, is an important decision factor before adopting an OSS SCP. In this paper, we evaluate different open-source SCPs using the concept of "OSS health", covering several of the aforementioned characteristics.
cs.SE / 8 / 2609.18805
ProgramDistill: From Interactive Web Apps to Verifiable Reference-Guided SWE Tasks
Abstract
Coding agents are typically evaluated with desired behavior specified through issues or instructions. In practical web development, however, agents may need to infer behavior from working software and implement it in an incomplete application. We introduce ProgramDistill, a benchmark evaluating coding agents on features discovered through interaction with fully functional reference applications. We build ProgramDistill by factorizing applications into features of different granularities, each associated with replayable behaviors executable via its gold patch. Our pipeline, mine-craft-patch, discovers 1,975 replay-verified behaviors across 26 applications and constructs 4,063 tasks without human intervention. Across nine frontier coding agents, GPT-6 Astra and Claude Opus 5 achieve 49.2% and 28.8% success on cumulative workflows in full-application reconstruction. In partial-application reconstruction, success falls from 100% to 64.0% and from 96% to 32% as restoration depth increases from 1 to 8. ProgramDistill thus provides a scalable benchmark with controlled difficulty for evaluating and diagnosing coding agents, and a natural basis for future curriculum-based training.
操作系统 (cs.OS)
1
cs.OS / 1 / 2609.18633
Netkit: Specializing Linux Packet Delivery for Container Networks
Abstract
Cloud-native microservices architectures rely on network namespaces for isolation, with the overhead of container communications remaining a critical performance bottleneck. While colocating containers on the same host mitigates some of this overhead, it cannot match the performance of communication within a single network namespace. Existing solutions either require application rewrites or fail to support the full Linux network stack expected by containerized applications. In this paper, we present netkit, an eBPF-based datapath that specializes the Linux networking stack to eliminate redundant backlog queue traversals during network namespace transitions. netkit leverages eBPF to transparently redirect packets between namespaces, bypassing unnecessary buffering while preserving compatibility with existing container applications. Our implementation in the Linux kernel, integrated with minimal changes to the Cilium network plugin for Kubernetes, improves throughput by up to 37\% and achieves parity between container-to-container and process-to-process communications, effectively closing the performance gap introduced by namespace isolation.
硬件架构 (cs.AR)
7
cs.AR / 1 / 2609.17730
FAME: An FPGA-Based Platform for Approximate Multipliers Evaluation with Pattern-Guided DNN Retraining
Abstract
Approximate multipliers can reduce hardware area and energy consumption in Deep Neural Network (DNN) inference; however, they introduce computational errors. Assessing the accuracy of numerous approximate multiplier designs across diverse DNN models and large-scale datasets remains challenging due to prohibitive evaluation times. This overhead primarily stems from the slow emulation of approximate multiplier behavior using look-up tables (LUTs) on CPU and GPU platforms. Moreover, the resulting accuracy degradation must be carefully quantified and, if necessary, mitigated (e.g., through retraining), further increasing the overall evaluation cost. To address these challenges, we propose FAME, an FPGA-based platform for evaluating approximate multipliers. The platform exploits the reconfigurable logic of Field-Programmable Gate Arrays (FPGAs) to implement approximate multipliers directly in hardware, eliminating the need for LUT-based emulation on CPU/GPU platforms and thereby enabling efficient DNN inference while significantly reducing evaluation time on large datasets. Furthermore, we introduce a pattern-guided DNN retraining technique to mitigate accuracy degradation induced by approximate multipliers. Specifically, retraining is guided by multiplier-specific patterns to effectively recover potential accuracy losses. We evaluate FAME using two DNN models, ResNet-18 and MobileNetV2, on the ImageNet dataset across 27 approximate multipliers. During inference, our approach achieves up to a 3.47x speedup in approximate multiplier evaluation compared to prior LUT-based emulation methods. Furthermore, the proposed retraining technique improves accuracy by up to 65.5% over existing retraining approaches for the evaluated multipliers. The code is publicly available at: https://github.com/gicLAB/FAME
cs.AR / 2 / 2609.17922
Demystifying Gate-Level Localization of RTL Trojans
Abstract
Hardware Trojans are malicious modifications that compromise functionality or leak sensitive data. They pose a severe threat, particularly when inserted at the Register Transfer Level (RTL). After synthesis, these Trojans are often concealed by optimizations in gate-level netlists. Recent efforts, including the ICCAD 2025 contest, emphasize golden-chip-free detection using machine learning (ML) on labeled netlists. In this work, we show that RTL Trojans exhibit stable structural and signal-flow patterns post-synthesis, enabling effective detection through targeted heuristics rather than generic ML feature learning. We propose LoRD, a lightweight heuristic-based approach that exploits these distinctive subgraph signatures, achieving near-perfect detection and localization on the contest testcases. Com- pared to a transformer-based ML baseline and top five teams, LoRD achieves on-average a score of 2.957 (out of 3) for Trojan- implanted designs without the data and tuning overhead.
cs.AR / 3 / 2609.18022
VeriBugBench: An Empirically Grounded Framework for Constructing Verilog RTL Debugging Benchmarks
Abstract
RTL source-level debugging research requires benchmark artifacts that provide faulty designs together with precise change locations, executable test stimuli, and reproducible configurations. Available Verilog resources usually provide only a subset of these elements. We present VeriBugBench, a framework for constructing Verilog RTL debugging benchmarks through empirically grounded fault construction, LLM-based testbench enhancement, and execution-based retention. The mutation library maps recurring, multi-granularity repair patterns observed in RTL bug-fix histories to 19 executable inverse operators. For each project, an LLM generates a design-specific stimulus phase from the clean DUT and original testbench; the phase is composed with the original testbench for candidate execution. Applying the framework to 45 open-source projects yields VeriBugBench-v1.0, with 2,608 executable single-fault instances whose effects are observable at design outputs. Across the 45 projects, the assembled testbenches increase mean project-level fault observability from 36.01% to 39.54% and improve line coverage and execution-trace diversity on average. VeriBugBench provides versioned RTL variants, source-level ground truth, testbenches, and execution artifacts for evaluating RTL debugging methods.
cs.AR / 4 / 2609.18662
Automated Instruction Encoding Synthesis for Modern GPU ISA Compression
Abstract
Modern GPU kernels increasingly stress the instruction supply path, while fixed instruction containers can leave substantial footprint slack. This paper presents an automated encoding-synthesis framework that treats instruction layout as a constrained slot-assignment problem over a validated instruction-form field specification. The formulation separates semantic field identity from physical bit positions and supports tied, pinned, and free placement constraints, making it applicable when recurring decoded fields are not frozen by a public format contract. We instantiate the framework for NVIDIA SASS: raw public encoding text is normalized into a machine-readable specification, a SASS disassembler is validated against nvdisasm on 3.78M instructions, released as an open benchmark, and CP-SAT synthesis is used for fixed-length and variable-length encodings. On 142 Blackwell kernel inputs, variable-length synthesis reduces instruction footprint by 33%, with comparable reductions after re-synthesis on Ampere and Hopper; fixed-length synthesis on the same specification reduces decoder area by 16% against a decoder generated from the NVIDIA-observed 128-bit layout by the same generator and flow. Generated fetch/decode RTL meets 1.5 GHz in TSMC 22 nm with a replicated area delta of 0.12% of a GA100-class die; a same-node SRAM comparison shows the footprint reduction corresponds to about 9x this added logic in instruction-SRAM bit-cell area.
cs.AR / 5 / 2609.18792
Quantifying the Effect of HCLs on a Fixed-Microarchitecture MXFP4 Accelerator
Abstract
Hardware Construction Languages (HCLs) aim to improve hardware design productivity while generating register-transfer-level (RTL) circuits without changing the designer's microarchitecture. However, most comparisons between HCLs are either qualitative or evaluate quality of results (QoR) across different designs, making it difficult to separate language effects from design effects. This paper compares the most widely used HCLs using the same fixed design, the OCP MXFP4 block dot product, a quantization primitive at the heart of edge Physical-AI inference, implemented as a single 12-stage, II=1 pipeline. A SystemVerilog baseline is followed by implementations in Chisel, SpinalHDL, Amaranth, Clash, Bluespec, and C++ for high-level synthesis (HLS). Every variant goes through the same flow on the same Artix-7 device set at 100 MhZ, driven by a RISC-V soft core. With the micro-architecture held constant, the comparison is clean: every variant meets timing, and the HCLs match or even undercut hand-written RTL in area. The remaining differences stem not from the algorithm but from how each back end lowers arithmetic, and from a single width choice that silently toggles DSP inference. Unlike HLS, where design decisions are limited to pragmas, the HCLs achieve comparable area and timing. Therefore, the choice comes down to ecosystem fit and interface needs rather than QoR.
cs.AR / 6 / 2609.18846
Locus: A Framework for Exploring and Optimizing Point Addition Hardware for Zero-Knowledge Proofs
Abstract
Zero-Knowledge Proofs (ZKPs) are critical for privacy-preserving and verifiable computation, but their cryptographic primitives impose high computational overheads. One such primitive is point addition (PADD) on elliptic curves. Several prior works have implemented PADDs in hardware, but only for a few specific elliptic curves and design points, leaving a large design space unexplored, and lacking systematic guidance on hardware design trade-offs. To address this gap, we present Locus, a framework dedicated to optimizing and exploring point addition hardware. Given the parameters of any elliptic curve in a supported equation form, Locus automatically generates ASIC and FPGA implementations of PADD, enabling systematic exploration of the PADD design space. Using Locus, we conduct the first comprehensive hardware-focused study of PADD designs, exploring trade-offs over 1,000 design points. On a 12nm technology node, our framework produces PADD designs that yield a $2.71\times$ geomean speedup and $3.11\times$ geomean area reduction compared to prior ASICs, $34.67\times$ geomean speedup over CPU, and $3.15\times$ geomean speedup on end-to-end proof generation when integrated into a prior ZKP accelerator at iso-area. Locus is available at https://github.com/cryptolets/cryptolets/tree/locus.
cs.AR / 7 / 2609.17903
Composability rather than computation sets the cost of an analog EML hardware fabric
Abstract
The operator eml(x, y) = exp(x) - ln(y) with the constant 1 generates the elementary functions, a continuous counterpart to NAND. Whether it yields a useful fabric had not been asked of hardware. We ask in network models, circuit simulation and SkyWater 130 nm layout. Four bipolar junctions evaluate the operator for 13 fJ, beating a width-matched digital datapath by 4-134x. The fabric assembled from them is not cheap: it loses to resource-matched baselines, and over the reals its grammar excludes trigonometry. Amplifiers holding those junctions' operating points take 74.5% of a cell's current, so a cell costs 3000 times what they spend. Extracted non-idealities cost 2.6x when a cell must hold a value and nothing when it need only be repeatable. Sharing them across cells recovers two of the three orders. The premise was that a universal primitive licenses a uniform machine. It survives in the primitive and fails in the machine.
密码学与安全 (cs.CR)
35
cs.CR / 1 / 2609.18020
Improved lower bounds for decomposable randomized encoding
Abstract
A decomposable randomized encoding (DRE) for a function $f$ allows $n$ parties, using shared randomness, to encode their individual inputs locally so that the collection of encodings reveals $f(x_1,\ldots,x_n)$ and nothing else. DREs are widely used in efficient multiparty computation. Their main complexity measure is size, the total bit length of the local encodings. Yet the optimal DRE size remains poorly understood even for the $n$-bit OR function. We prove the first superlinear lower bound for OR and, more generally, for every non-periodic symmetric function. Under an additional symmetry assumption, we prove a sharp $Ω(n\log n)$ lower bound for OR, matching the classic construction of Feige, Kilian, and Naor (STOC 1994). We also prove the first $Ω(n^2)$ lower bound on DRE size for non-explicit Boolean functions.
cs.CR / 2 / 2609.17817
Reflections on Trusting Trust, Revisited: Contaminating Self-Modifying AI Coding Agents with Poisoned Benchmarks
Abstract
Thompson's "Reflections on Trusting Trust" showed that a compiler can be poisoned to reinsert its own backdoor, so that even recompiling clean source reproduces the Trojan. Today, substantial coding work is done by AI coding agents -- and increasingly, those agents generate new versions of themselves. We reconsider Thompson's attack when the "compiler" is a self-modifying coding agent. Can an adversary supply poisoned benchmarks to the agent's self-evaluation and self-improvement process to induce future versions of the agent to write vulnerable code on clean, held-out tasks? We instantiate this attack against three recently proposed self-modifying coding agents: the Darwin Gödel Machine (with our experimental modifications), the Self-Improving Coding Agent, and Hyperagents (both substantively unmodified). We demonstrate successful proofs-of-concept: for example, with Hyperagents powered by Sonnet 4.5, our poisoned benchmark leads the agent to self-evolve instructions that disable HTTPS certificate validation on neutral URL-fetching tasks. From our experiments, we distill properties of the vulnerability, benchmark, model, and agent scaffolding that are sufficient to enable a benchmark poisoning attack. Moreover, we show that contamination often persists even when a poisoned agent is subsequently evolved against clean benchmarks. We discuss defensive directions and argue that self-modifying coding agents must be designed to be more resilient to such attacks.
cs.CR / 3 / 2609.17897
When AI Agents Meet MEV: Cross-Chain Arbitrage in the Agentic Economy
Abstract
We study cross-chain arbitrage when autonomous AI agents, rather than humans or bots, are the searchers. We model agents as both arbitrage extractors and Maximal Extractable Value targets, derive the optimal trade size for a risk-averse agent under mean-variance utility with stochastic bridge delays, and formalize multi-chain path selection as a belief-weighted online learning problem whose belief estimates converge under a Robbins-Monro schedule. Using 23,000 Uniswap V3 swap events across Ethereum, Arbitrum, and Base, we find that Ethereum-Arbitrum price gaps average 0.044% at 10-second resolution and Arbitrum--Base gaps average 0.013%, so $10,000 trades clear in 63% of L2-L2 windows via CCTP while L1-L2 routes require $50,000 or more for comparable viability. Our adaptive path-selection algorithm outperforms standard baselines by 11% on average, and moderate randomization cuts MEV exposure by over 50% with only modest profit loss.
cs.CR / 4 / 2609.17902
Ghost-Filled Orders: Detecting and Testing Atomicity Violations in Non-Custodial Prediction Markets
Abstract
Blockchain based prediction markets combine offchain order management with onchain settlement. This architecture supports user controlled custody, since users keep funds in wallets or smart contracts while submitting signed orders to an offchain order book. However, it creates an atomicity gap. An order may be valid when accepted or matched offchain, but become invalid before the corresponding onchain settlement transaction is executed. This behavior, often called ghost filled orders by the community, can cause trades that appear filled offchain to fail onchain. This paper studies this atomicity gap through a case study of Polymarket. We show how the delay between offchain order acceptance and onchain settlement allows adversaries to invalidate unfavorable orders after observing market outcomes or price movements. We then quantify the scale and financial impact of this behavior over a nine-month period from August 12, 2025, to May 22, 2026, using 1.8 million reverted transactions involving Polymarket official smart contracts. Our analysis separates attacker profit from market and user loss, and develops conservative measurement rules to avoid overclaiming impact. We further develop a methodology for testing other blockchain based prediction markets and apply it to three additional markets. We find that all three are vulnerable to the same class of attack, and that one design enables direct attacker profit. We have reported the findings to all three projects; one project had acknowledged the issue at the time of the study.
cs.CR / 5 / 2609.18120
PentestChain: A Cost-Aware, MCP-Orchestrated Framework for Automated Penetration Testing with Free-Tier LLMs
Abstract
AI-driven penetration testing has been demonstrated with premium frontier models such as GPT-4, but the per-engagement token cost makes continuous, automated testing unaffordable for the smaller organisations that need it most. This paper presents PentestChain, a ten-phase automated penetration testing framework that couples a curated, deterministic exploit map with a cost-aware AI cascade-a local Ollama model (qwen2.5-7b) first, then free-tier OpenRouter and Cerebras, with a rule-based fallback that always produces output-and exposes the full pipeline through a Model Context Protocol (MCP) server with eleven tools. We make three contributions. First, we treat US-dollar cost per engagement as a measured, first-class evaluation metric and show that a 7B-parameter local model, kept off the critical path by a deterministic backbone, sustains end-to-end operation at zero measured paid-API cost. Second, we analyse the attack surface that an MCP-exposed offensive engine introduces, grounding a four-position threat model in the 2025 MCP incident record (the CVE-2025-6514 remote-code-execution flaw in mcp-remote, the postmark-mcp supply-chain backdoor, and the tool-poisoning-rug-pull-line-jumping class), and contribute four mitigations. Third, we specify a reproducible, containerised evalua-tion protocol aligned with the standardised testbeds now expected at top-tier venues-AutoPenBench, a Cybench subset, and the PentestGPT 182-sub-task benchmark-with multi-trial statistics (more than 10 trials per configuration, pass-at-k, non-parametric significance tests and effect sizes) and direct, same testbed reproduction of the PentestGPT and PentestAgent baselines rather than citation of their published numbers. On the legacy targets measured to date, the framework detected 26 services, enriched 34 CVEs, produced
cs.CR / 6 / 2609.18122
"Your Robot Was Trained on a Lie": Collision Mesh Poisoning Attacks on Robotic Manipulation
Abstract
Learning-enabled robotic manipulation increasingly relies on robot simulators for policy training and evaluation before real-world deployment. Inside a simulator, a 3D asset contains two separate geometries: a visual mesh used for rendering and a collision mesh used for physical interaction. For computational efficiency, the collision mesh is deliberately a coarse approximation that need not have the same geometry as the visual mesh, a legitimate and pervasive discrepancy we call the Visual--Collision Gap (V--C Gap). We show that the V--C Gap opens a new and practical attack surface, and propose Collision Mesh Poisoning (CMP), the first poisoning attack against robotic manipulation delivered through the 3D asset supply chain. An attacker modifies only the collision mesh of a 3D asset, leaving the visual mesh and all other components unchanged. A policy trained and evaluated with the poisoned asset behaves normally throughout simulation, yet degrades, fails, or creates physical safety risks once deployed in the real world. Since current asset review practices cover malware, copyright, and format compliance, but not visual--collision consistency, poisoned assets can be distributed through legitimate supply chain channels. We evaluate several defenses and our results show that they are insufficient to defend against CMP, highlighting the need for new defenses.
cs.CR / 7 / 2609.18158
Bridging the Opacity: Evidence-Backed Cross-Chain Transaction Correspondence Reconstruction Across Heterogeneous Blockchains
Abstract
Cross-chain bridges enable interoperability, but they also break the transaction trails needed to trace illicit funds. Third-party investigators typically cannot access the source-to-destination mappings maintained by bridge backends, and our survey of 131 bridges finds that only 16.79% provide complete public tracking. Existing approaches depend on official APIs, EVM-specific assumptions, or fragile temporal heuristics, limiting their ability to trace transfers across heterogeneous ledgers. We present XSplicer, an evidence-driven system for reconstructing cross-chain transaction correspondence (xTCR) without privileged access to bridge backends. XSplicer derives unified semantic specifications from public protocol documentation and transaction examples, translates them into lightweight parsers and verifiers, and links source and destination transactions by prioritizing hard evidence and using soft clues only when necessary. We evaluate XSplicer on seven bridge protocols spanning EVM, Bitcoin, and Solana. XSplicer achieves 92.5% global recovery rate and up to 98.61% on individual protocols. Under adversarial noise, its hard-evidence verifier retains the correct match in 100% of tested cases, while soft-clue matching degrades as ambiguity increases. In two real-world case studies, XSplicer recovers more than 1,900 historical transaction pairs after Multichain ceased operations and identifies 754 illicit cross-chain transfers worth 105.6 million USD in the Bybit laundering incident. These results show that public protocol invariants can support practical cross-chain forensics without privileged bridge mappings.
cs.CR / 8 / 2609.18196
ISIA-AF: Orchestrating Reproducible Attacks and Multi-Source Data Collection for OT Systems
Abstract
Operational Technology (OT) environments require realistic, reproducible security datasets, yet existing approaches often lack automation, multi-source data capture, and sufficient documentation for reuse. This paper presents ISIA-AF, a modular attack framework for orchestrating reproducible attack execution and automated dataset generation on industrial systems. The framework coordinates distributed attack clients, records network traffic and operational data, ultimately leading to a multi-source dataset. We derive functional and non-functional requirements from prior work and stakeholder discussions, and realise the framework following a design science research approach. A case study on the ISIA testbed, comprising a real industrial system and a simulated process, demonstrates how the framework supports centralised control, low communication overhead, and flexible deployment across network segments. The result is a practical basis for generating extensible, multi-source OT security datasets for intrusion detection research.
cs.CR / 9 / 2609.18217
Measuring and Exploiting Implicit Trust in LLM Tool-Calling Pipelines
Abstract
The Model Context Protocol (MCP) enables LLMs to invoke external tools, but every tool interaction exposes the model to attacker-controlled text through multiple input channels (tool descriptions, tool results, sampling messages) that share a single context window without privilege separation. In this paper, we present a framework to measure the trust profile of an arbitrary LLM based on a variety of payload framings sent through different channels. Following this assessment, we devise cross-channel fragmentation attacks that distribute seemingly benign payloads across two or three channels; no individual channel carries a complete injection, yet the LLM compiles the fragments into credential exfiltration. We evaluated our attacks across 12 frontier models, three production clients, and six payloads, totalling over 15,000 trials. Our evaluation reveals that cross-channel attacks are an unexplored attack surface: models that fully resist single-channel injection (0% compliance) exfiltrate sensitive data at up to 100% under two-channel fragmentation (e.g., GPT-4o, Llama 70B, Composer 2, Haiku 4.5). We further demonstrate value-aligned exploitation, where a tool's stated purpose requires the data the attacker targets, and a sampling system prompt override that injects persistent instructions via VS Code's MCP implementation. Finally, we evaluated our attacks against seven third-party MCP security tools and three prompt-based defenses. All tools failed to detect fragmented payloads, and prompt defenses proved model-specific rather than universal.
cs.CR / 10 / 2609.18275
Witness Encryption via Prime-Order Generic Groups
Abstract
We unconditionally construct witness encryption for NP in the classical generic-group model, using an ordinary cyclic group of prime order. For SAT instances of size $n$, the encryption algorithm runs in time poly$(n)$, and any satisfying assignment can be used to decrypt in poly$(n)$ time with correctness error $2^{-n^{Ω(1)}}$. If no satisfying assignment exists, then every generic adversary making at most $n^{Θ(\log n)}$ group queries has distinguishing advantage at most $n^{-Θ(\log n)}$. Along the way, we prove the first superconstant-factor NP-hardness of approximation result for homogeneous MinRank under randomized polynomial-time reductions, achieving a logarithmic gap even when the rank-one witness has a Boolean right factor.
cs.CR / 11 / 2609.18338
Autonomy in Check: Governor-Mediated Adaptive Security at the Edge
Abstract
Adaptive security at the network edge increasingly relies on automated planners, including rule-based controllers, learned policies, and LLM-assisted agents, that translate observations into enforcement actions. Once such a planner can influence live policy state, syntactic validity is not enough. A semantically wrong action, produced from incomplete or manipulated observations, can be faithfully executed by an enforcement substrate that cannot judge mission context. We address this problem by treating the boundary between planner output and kernel enforcement input as the primary security object. We propose a split-control architecture in which an untrusted planner emits typed security intents, a deterministic governor checks each intent against safety, resource, temporal-stability, and proportionality invariants, and only admitted actions are bound to signed receipts and compiled into pre-installed eBPF map updates. The paper formalizes this trust-boundary problem, defines three threat classes, develops the governor admission predicate, and reports an end-to-end prototype. Across rule-based and LLM-assisted planners on a Raspberry Pi 5 testbed connected to the university 5G Test Network, the governor admits, rejects, and bounds intents at microsecond cost without disrupting protected-flow regularity. The contribution is conceptual as much as empirical: adaptive security does not need to trust the author of an action. It needs a mediation boundary that decides whether the action is admissible.
cs.CR / 12 / 2609.18344
Detecting Logic Vulnerabilities Across the Contract and Device Layers of Blockchain-Enabled IoT With Multi-Agent Heterogeneous Graph Attention
Abstract
Blockchain-enabled Internet of Things (IoT) systems integrate smart contracts with embedded devices to support decentralized device management and access control. Their security therefore depends jointly on the logic of on-chain contracts and off-chain device firmware. Logic flaws in either layer can violate the same system invariants, such as unauthorized access, improper state changes, or unguarded privileged operations. Existing approaches rely on contract analysis, firmware analysis, and graph-based vulnerability detection. However, these methods typically focus on a single layer or artifact and often depend on predefined vulnerability patterns, emulation fidelity, or homogeneous representations that obscure security-relevant component roles. They also lack a unified architecture that supports different security tasks while remaining deployable on resource-constrained gateways. To address these limitations, we extend MA-HGAT into a cross-layer multi-agent heterogeneous graph attention framework that models contracts, firmware artifacts, device fleets, and transaction streams with a unified four-role, nine-relation schema. Role-aligned agents exchange heterogeneous evidence through cross-attention, while graph-, link-, and node-level heads support multiple detection tasks and a role-based gateway--cloud partition enables lightweight edge inference. MA-HGAT thus provides a unified and deployable framework for detecting logic vulnerabilities across the contract and device layers of blockchain-enabled IoT systems.
cs.CR / 13 / 2609.18411
The Verifiable Action Card: Trustworthy Human-in-the-Loop Control for Secure Autonomous Agents
Abstract
Agentic browsers can execute security-sensitive actions under a user's authenticated session, making indirect prompt injection and deceptive confirmation interfaces a direct threat to action integrity. Existing human-in-the-loop (HITL) safeguards are insufficient when the approval prompt itself can be influenced by untrusted page content or model-generated text. We present the \emph{Verifiable Action Card} (VAC), an architectural defence that reconstructs approval information from the ground-truth pending browser action and trusted intent provenance, renders it out-of-band in the trusted browser chrome, and binds approval to the exact action re-verified at dispatch. VAC combines provenance fencing, a ground-truth action descriptor, default-deny confirmation, provenance-aware risk gating, and execution binding. We implement VAC in a complete agentic browser and evaluate it on a 24-scenario benchmark covering confused-deputy attacks, Lies-in-the-Loop dialog forging, indirect prompt injection, adaptive action substitution, provenance evasion, and legitimate tasks. Across the evaluated LLMs, attack success without VAC ranges from $68\%$ to $100\%$, whereas VAC reduces attack success to $0\%$ on every model, with $78\%$ legitimate-task completion and a $0\%$ false-block rate. These results show that grounding approval in the action that will actually execute provides architectural protection against security failures that prompt-level defences and conventional HITL confirmation cannot reliably prevent.
cs.CR / 14 / 2609.18457
AIJon: Automated Generation of Annotations for Fuzzing
Abstract
Modern fuzzers use code coverage as feedback to guide their exploration which has proven to be an effective strategy for driving exploration. However, this strategy overlooks inputs that may be interesting to the target program even without uncovering new code paths. Fortunately, prior research has shown that annotations generated by human domain experts can provide additional feedback, guiding the fuzzer towards interesting parts of the program. In this paper, we replicate experiments presented in IJON and extend them to real-world vulnerability detection at scale. To mitigate the scalability challenge, imposed by the need for human domain expertise, we propose utilizing LLMs to automatically generate annotations. We demonstrate the applicability of LLMs for this purpose and observe that LLMs can generate annotations that perform comparably to human-generated annotations. Motivated by this finding, we design AIJON, a system that leverages LLMs to automatically generate IJON-style annotations. We evaluate AIJON on the Magma benchmark and surprisingly observe that annotation-based fuzzing does not perform strictly better than AFL++. We conduct several experiments to identify the cause of our results and identify key insights regarding the impact of annotations on fuzzing campaigns, including their effect on the energy distribution of the fuzzer. Notably, we observe that LLMs can generate annotations that achieve comparable results to human generated ones, thus opening the door for future research to perform further studies on the impact of annotations at scale.
cs.CR / 15 / 2609.18459
SEEK: Secure and Efficient Encrypted Keyword Search For Privacy-Preserving Messaging Protocols
Abstract
Encrypted communication protects sensitive user data but can facilitate harmful or unlawful exchanges, creating a trade-off between detecting dangerous messages and preserving end-user privacy. To address this, we propose SEEK, a practical and efficient encrypted keyword-search protocol for privacy-preserving messaging that combines homomorphic encryption with secure two-party computation (2PC). SEEK first partitions messages into ciphertext fragments with the minimum sufficient overlap, then homomorphically correlates them using encrypted keyword trapdoors. For long messages, this design can reduce sender-side encryption and upload overhead by up to two orders of magnitude over state-of-the-art baselines. It supports ASCII case-insensitive matching with one fixed-size encrypted trapdoor and one homomorphic multiplication per fragment, yielding up to 5.47x faster correlation computation than the strongest fragmentation-based baselines. SEEK then invokes 2PC-based selected decoding, blinded zero testing, and secure aggregation, revealing only the keyword presence-or-absence bit while hiding the keyword, its length, message contents, match counts, and locations. SEEK achieves 100% accuracy under case variations that result in exact-matching failures, without requiring additional trapdoors or online communication. We further realize SEEK as an end-to-end web and cross-platform mobile application. Prototype evaluation on a weekly messaging history yields an online computation time of 1.92 s per search, demonstrating the practical feasibility and efficiency of SEEK.
cs.CR / 16 / 2609.18477
A Global Readiness and Sovereignty Capability Model for Post-Quantum Cryptography Migration
Abstract
Cryptographic dependence predates the quantum era, but the migration to post-quantum cryptography (PQC) opens a rare window to reshape it, because the algorithms, implementations, hardware, and standards adopted now can lock in dependence or sovereignty for decades. This paper introduces the Readiness-Sovereignty Capability Model (RSCM), a national measurement model that operationalizes PQC readiness together with cryptographic sovereignty, which current maturity models score only as readiness and the sovereignty literature defines without measuring. RSCM decomposes sovereignty into three distinct constructs, indigenous cryptographic capacity, indigenous post-quantum control, and external dependency, and certifies a post-quantum maker only through a gate requiring demonstrated, institutionally sustained creation in at least one core layer, whether design, implementation, or validation. Applying it to fifty-seven documented cryptographic actors coded from cited public evidence, and testing that coding with an independent second coder, a plausible-state bootstrap, and convergent-validity checks, we find that twenty countries clear the gate, fifteen as full-stack makers and five as research makers, eleven hold strong general capacity without post-quantum control, one is a ready adopter, and twenty-five are dependent. The gate cells show substantial weighted agreement, a quadratic-weighted kappa of 0.71, and the maker classification is stable in its core though uncertain at the threshold. Readiness tracks independent cyber indices at rank correlations up to 0.70, while post-quantum creation shows no significant correlation with the commitment index, a rank correlation of only 0.22 that separates control from readiness. The paper contributes the framework, the evidence-graded assessment, and policy directions for building indigenous quantum-safe capacity.
cs.CR / 17 / 2609.18496
MiST: Mid-Training LLMs for Cybersecurity
Abstract
Cybersecurity combines high-stakes analysis with complex technical language, making it an impactful and challenging domain for LLMs. We present MiST (Mid-trained Security Transformer), a suite of 8B and 32B models that achieve strong performance on public cybersecurity benchmarks. We use mid-training as an intermediate adaptation stage between general pre-training and cybersecurity training. Rather than performing continual pre-training over large volumes of raw domain text, we curate a compact, expert-vetted seed corpus, and transform it into high-quality domain-specific synthetic training data. The final MiST checkpoints improve mean cybersecurity accuracy by +13.1 and +8.6 absolute percentage points over the corresponding Qwen baselines for 8B and 32B, respectively, corresponding to relative gains of +27.0% and +15.8%. Ablation results further show that these cybersecurity gains arise in the mid-training and supervised fine-tuning stages through a combination of the synthetic data generation flows. Furthermore, we show that MiST provides a stronger initialization for downstream task-specific fine-tuning adaptation and reinforcement learning.
cs.CR / 18 / 2609.18518
Robot Visions: Breaking reCAPTCHA at Zero Cost and Zero Shot
Abstract
Google reCAPTCHA is the most widely deployed visual CAPTCHA service, protecting hundreds of thousands of websites from automated bots. It serves as a critical line of defense against automated attacks, including credential stuffing, bulk account creation, and automated form abuse. It has proven largely effective since its introduction in 2007. However, the rise of accessible AI now threatens its efficacy. Prior work has demonstrated that commercial cloud-based vision-language models (VLMs) can solve visual CAPTCHA challenges, but at non-trivial monetary cost per attempt. In this paper, we show that free and locally-run models can break Google reCAPTCHA. We conduct a comprehensive study of reCAPTCHA and present a taxonomy of its challenge types: Type A (independent image tiles, with static and dynamic sub-variants) and Type B (a single image partitioned into a 4x4 grid), each demanding a distinct solving strategy. We design zero-shot, no-cost solvers built entirely on open-source local models, specifically CLIP (58% per-challenge accuracy on Type A) and OWLv2 (43.5% on Type B), requiring no model training and no API access. Our end-to-end automated solver achieves a 92.6% per-session success rate across 500 real-world reCAPTCHA sessions. We further demonstrate that reCAPTCHA can be defeated by a non-technical adversary, using only natural-language instructions to a commodity AI assistant. This collapses the practical attacker skill floor to near zero and fundamentally changes the threat model for challenge-based CAPTCHAs. Although reCAPTCHA increasingly favors reputation-based verification, visual challenge-based fallback persists as a safety net that, paradoxically, has become the weakest link in the defense chain, suggesting that challenge-based visual CAPTCHAs may have reached the end of their useful life.
cs.CR / 19 / 2609.18658
A Security Risk Assessment Framework for AI-Powered Development Tools
Abstract
AI-powered development tools are now widely used to generate code and assist developers with routine programming tasks. Although existing work has identified vulnerabilities in AI-generated code, security-oriented work is often focused on vulnerability detection rather than risk assessment. To address this gap, this paper presents a Security Risk Assessment Framework (SRF) to evaluate the security risks of AI-generated code. SRF combines threat modeling, security analysis, and a quantitative risk evaluation approach based on vulnerability criticality. The framework is applied to a set of security-relevant programming tasks, where code generated by multiple AI-powered development tools is analyzed using Bandit and Semgrep. The results show that AI-generated code can introduce security vulnerabilities across all evaluated tools. They also show that risk levels vary by task type, as input processing and file handling tasks showed higher risk, while simpler tasks remained low-risk. Differences between tools exist but are smaller than differences across task categories. Overall, SRF enables reproducible evaluation of AI-generated code and provides a practical framework for assessing its security implications.
cs.CR / 20 / 2609.18674
CaMeLoT: CaMeL orchestrated with Temporal logic for static verification and liveness
Abstract
LLM-based agents generate and execute multi-step plans that invoke external tools which can access private data or execute commands. In this setting, security is a property of the entire execution that a plan creates, not just any single step. The plan itself is a critical artefact that captures the tool calls, control flow, and data dependencies. We present CaMeLoT, a complement to CaMeL, an existing defence against prompt injection in tool-using LLM agents. CaMeLoT extends CaMeL by adding a static verification layer that checks an agent's plan before any tool is invoked. CaMeLoT translates a generated plan into a finite-state transition system, labels it with tool calls, provenance and taint information, and checks it against temporal policies expressed in CTL using the nuXmv model checker. Because verification happens before execution, unsafe plans are rejected without using LLM calls or tool calls, saving tokens that runtime could have cost, as well as the need to unwind changes or teardown temporary sandboxes. When a verification fails, the model checker returns a counterexample to give feedback to the agent to repair the plan. We evaluate CaMeLoT on policies derived from the AgentDojo benchmark, SOC workflows, and prompt-extraction experiments, showing that it verifies a broad class of temporal properties before execution while preserving CaMeL's runtime-checkable coverage.
cs.CR / 21 / 2609.18706
Echo: Learning-based Matching Decompilation using Trusted Back Translation
Abstract
Neural decompilers can recover readable and recompilable source code from binaries, but their predictions remain difficult to trust. Matching decompilation addresses this problem by searching for source code whose recompiled assembly exactly matches the target, providing stronger evidence of correctness. However, exact matching remains challenging for optimized binaries under unknown compilation configurations. We present Echo, a matching decompilation system based on trusted back-translation. Our key insight is to use compilation not only for verification, but also as trusted feedback to guide iterative search. Echo first uses a domain-specific model to generate candidate programs and compilation configurations. It recompiles these candidates, measures assembly-level similarity, and synthesizes promising code-configuration pairs. Remaining mismatches are then progressively repaired using rule-based rewriting, neural refinement, and reasoning-based refinement. We evaluate Echo on function-level benchmarks and the Mirai malware binary. Compared with the strongest baseline, Echo produces 2.43x more exact matches on average and achieves the highest structural similarity to ground-truth source code. On Mirai, Echo matches 2.75x and 7.4x as many functions as GPT-5.6 and Codex, respectively.
cs.CR / 22 / 2609.18751
Normal Alignment: Improved Cryptanalytic Sign Recovery on Hard-Label Networks
Abstract
At EUROCRYPT 2025, Carlini et al. proposed a breakthrough in the cryptanalytic extraction on hard-label (S1) deep neural networks (DNNs), demonstrating polynomial-time signature and sign recovery. However, Carlini et al.'s sign-recovery method (which we call Future Toggle) suffers only a marginal advantage over random guessing, producing high-confidence wrong sign predictions in deeper layers. Such errors trigger expensive exponential-time enumeration. This work presents Normal Alignment, a novel statistical sign-recovery approach for S1 DNNs. Drawing on the expected length difference between projected normals of adjacent decision facets at dual points, our method infers neuron signs via normal-signature alignment. It delivers higher voting accuracy and pushes erroneous predictions to low-confidence ranks, which further enables a more efficient combined method, eSOE + Alignment, by combining Normal Alignment with the hard-label SOE extension. This combined strategy removes heavy enumeration overhead and realizes exact polynomial-time full sign recovery. Experiments demonstrate the effectiveness of our method, especially for deep layers. For example, with our method, the signs for CIFAR-10 (architecture 192-64$\times$8-10) and MNIST (architecture 64-96$\times$3-32-10) models can be fully recovered in polynomial time; in contrast, Carlini et al.'s sign-recovery method would require exponential-time enumerations involving $2^{52}$ or $2^{82}$ guesses of the signs, respectively.
cs.CR / 23 / 2609.18783
s-MDM: Generative Virtualization of Multi-Device Hardware Variations for Portable DL-SCA
Abstract
Deep Learning-based Side-Channel Analysis (DL-SCA) frequently suffers from catastrophic performance degradation across unseen hardware due to printed circuit board routing differences, silicon process variations, and measurement noise shifts. This poster presents the Synthetic Multiple Device Model (s-MDM), a zero-target-trace generative framework designed to improve cross-device portability. s-MDM combines a structured cVAE generator, a Walsh-Hadamard leakage anchor, continuous style modulation, and decoupled leakage-style--domain critics to synthesize virtual source-device profiles offline. Benchmarked on 32-bit side-channel traces (AES_PTv2), s-MDM maps a precise operational boundary: while physical MDM remains superior on identical electrical clones (D4), s-MDM achieves consistently low key rank on the layout/acquisition-shifted Pinata target, where physical baselines are unstable or misaligned.
cs.CR / 24 / 2609.18811
Differential Trust: Dynamic Multi-Authority Anonymous Credentials with Epoch-Weighted Updates
Abstract
Anonymous credentials (ACs) are fundamental to privacy-preserving authentication, allowing users to prove possession of attributes without revealing their identities. State-of-the-art ACs distribute credential issuance across multiple authorities, typically employing techniques such as Shamir's secret sharing or aggregate signatures. While this approach enhances system robustness and eliminates a single point of failure, it treats all authorities equally in the credential issuance phase. This uniform treatment disregards the varying levels of trustworthiness or stake held by different authorities. Such a limitation has become particularly problematic in modern decentralized systems like Proof-of-Stake networks, where the inherent trust differentiation among nodes cannot be leveraged in the credential issuance process. To address this limitation, we propose the notion of Multi-Authority Anonymous Credentials with Epoch-Based Weights (MA-ACEW), the first Multi-Authority Anonymous Credential (MA-AC) model that considers authorities' weight distribution in credential issuance. Crucially, MA-ACEW enables efficient credential updates when authority weight distributions change across epochs. The core of MA-ACEW is our novel Epoch-Bound Pointcheval-Sanders Signature (EB-PS) primitive, which binds signatures to specific time epochs. This temporal binding enables both weight-based credential issuance within epochs and efficient non-interactive credential updates across epochs. We formalize the EUF-eCMA unforgeability requirement for EB-PS and prove our construction satisfies it under a novel STB-GPS assumption. We then prove that our MA-ACEW construction achieves unforgeability, anonymity, and blindness. Finally, we present benchmarks demonstrating the efficiency of EB-PS and MA-ACEW. Remarkably, presenting a credential aggregated from 128 partial ones takes only 10.68 ms on average.
cs.CR / 25 / 2609.18862
CASHEWS: Source Preprocessor for LLM-based Malicious Package Detection
Abstract
Malicious npm package detection tools now leverage LLMs' semantic understanding of source code to detect malicious intent at scale. This capability has proven invaluable in identifying packages involved in recent supply-chain attacks such as Shai-Hulud. However, threat actors exploit the limited context windows of LLMs through JavaScript techniques such as code obfuscation that yields high token density and bundling malicious code with benign packages, causing detectors to skip large files or miss malicious behavior. This creates an attack surface for evading detection. In this paper, we present CASHEWS, a JavaScript preprocessor that reduces file size by rewriting source code to remove code that is irrelevant to analysis or likely to mislead the model. Given a package source file, CASHEWS deobfuscates it through iterative decoding, extracts bundled modules and dynamically executed code, identifies malicious sinks and computes backward slices that reach them, and abbreviates long literals and identifiers to produce a compact representation for the detector. Across 512 large package files, two scanner types, and three LLMs, CASHEWS increases analysis coverage from 69.1--85.7% to 98.8--100% and reduces the false-negative rate by up to 18.6 percentage points. CASHEWS also has a median preprocessing time of 30 seconds while reducing net analysis cost by 34.6%, making registry-wide LLM-based analysis more practical. By preprocessing source code before analysis, CASHEWS enables researchers and industry practitioners to use more powerful models for malicious package detection at the same or lower analysis cost as less powerful models.
cs.CR / 26 / 2609.18864
ASLEval: Measuring Privacy Exposure Displacement in LLM Agent Sessions
Abstract
Privacy evaluations of tool-using LLM agents often inspect a designated action, final response, or attacker report. These local proxies can miss unauthorized exposure elsewhere in a multi-step session and lack common ground truth across outlets, reports, and tool paths. We introduce privacy exposure displacement, the mismatch between a local evaluation proxy and target-grounded session exposure, and ASLEval, an authorization-aware framework that pre-registers a hidden target set, measures all declared visible exits, and reserves internal traces for diagnosis. Across multiple enterprise-style environments and independently implemented runtimes, we observe three recurring patterns. An expected-outlet-only view misses 46.9% of exposure recovered by the visible-exit union; attacker self-reports combine omissions with high false discovery; and schema-aligned internal evidence usually precedes visible exposure at the request/probe level. Reducing model-visible returns changes this path but can eliminate normal-task success. Independent human review supports the adjudication pipeline while identifying harder console and candidate cases. These findings motivate benchmarks that declare the complete visible boundary, ground claims in pre-specified targets and authorization, and report privacy together with task utility.
cs.CR / 27 / 2609.18866
Hamming Ideals and Grobner Bases for ISD-like Syndrome Decoding
Abstract
We investigate an algebraic approach to the Syndrome Decoding Problem, based on a reformulation of the Hamming weight constraint and its integration with the Information Set Decoding paradigm. We begin with a systematic analysis of the Hamming variety, deriving its defining equations in terms of elementary symmetric functions. Since these equations may have high degree, we exploit convolution identities for elementary symmetric functions, together with factorizations based on Lucas' identity, to derive an equivalent formulation with auxiliary variables and equations of bounded degree. Building on this modeling, we generalize the ISD paradigm through an ISD-like decoding strategy, implemented by the GBDecode algorithm, in which only a subset of an information set is fixed. This approach reduces the size of the combinatorial search space at the cost of solving the associated multivariate nonlinear systems. To handle this algebraic component, we employ the MultiSolve algorithm, which replaces a single Grobner basis computation with a collection of computations on simpler systems, obtained by exhaustively assigning a varying number of indeterminates over the finite field. This provides a tunable balance between combinatorial search and algebraic solving. We evaluate the resulting approach experimentally on instances of the Syndrome Decoding Problem for random binary linear codes, using parameters corresponding to the NIST Security Category 1 parameter set of the Classic McEliece cryptosystem. The experiments assess the feasibility of this combinatorial-algebraic approach and provide insights into the practical behavior of Grobner basis techniques within an ISD-like decoding framework.
cs.CR / 28 / 2609.19091
When Agents Look Like Beacons: NIDS Evasion by Model Context Protocol Traffic
Abstract
The Model Context Protocol (MCP) standardizes communication between autonomous Artificial Intelligence (AI) agents and remote tools over Streamable HTTP. This shift introduces a class of machine-generated, authenticated, and high-frequency JSON-RPC traffic directly into enterprise networks. Enterprise network defenders have historically relied on machine-like cadence as an Indicator of Compromise (IoC). In this study, we show that without explicit network-layer indication, MCP traffic structurally and temporally resembles Command and Control (C2) beaconing behavior, specifically the polling architectures used by advanced persistent threats like Cobalt Strike. Counter to theoretical assumptions about machine-generated polling, our measurements reveal a visibility gap: standard enterprise Intrusion Detection Systems (IDS) and behavioral beacon-scoring frameworks do not classify MCP remote tool usage as anomalous within our testbed scope. Through a controlled Docker-based testbed simulating eleven mathematically defined traffic profiles across three TLS conditions (Opaque, TLS-Inspected, and Cleartext), we evaluate Suricata signature matching and RITA behavioral scoring against MCP JSON-RPC patterns. Our results show that MCP traffic, regardless of temporal smearing (jitter) or TLS inspection visibility, evades detection within this configuration, yielding a consistent 0.0 behavioral beacon score and near-zero IDS content alerts under the Emerging Threats (ET) Open ruleset. While opaque TLS obscures HTTP content, it exposes agent traffic to flow-level temporal analysis; however, NIDS heuristics tuned to identify traditional malware do not flag the lognormal inter-arrival distributions characteristic of generative AI reasoning loops. To address this gap, we propose an agent-native network indication standard including Agent-Native ALPN and standardized out-of-band headers.
cs.CR / 29 / 2609.19100
Characterizing Network Centralization and Observability in the Remote MCP Ecosystem
Abstract
The Model Context Protocol (MCP) has emerged as the dominant interface for connecting autonomous agents to external data sources and execution environments. The ecosystem's transition from local process execution to remote Streamable HTTP deployments introduces unmeasured architectural and security constraints at scale. This paper presents a three-tier observability framework comprising catalog metadata (O_0), passive compliance signals (O_1), and live vulnerability analysis (O_2), applied to empirically characterize the public MCP server ecosystem. Evaluation of a stratified sample of 179 remote endpoints across two primary public registries reveals significant infrastructural consolidation. The Herfindahl-Hirschman Index (HHI) computed over the Autonomous System Number (ASN) distribution yields a value of 0.736, well above the 0.25 threshold for a highly concentrated market. Analysis further indicates that server authentication is strongly correlated with hosting platform choice rather than individual operator configuration, with 95\% of commercial PaaS-hosted servers enforcing gateway-level OAuth 2.1 with PKCE. The empirical results identify a Security-Observability Tradeoff observed in the current ecosystem: the platform-level authentication mechanisms that secure the majority of servers simultaneously limit automated vulnerability scanning capabilities, constraining the ability of AI gateway operators to assess tool-poisoning vectors without prior credential provisioning.
cs.CR / 30 / 2609.19111
Analog Pin Directionality as an Exfiltration Attack Surface in Mixed-Signal ICs
Abstract
Mixed-signal SoCs rely on nominally input-only analog pins to acquire off-chip signals, but the directionality of these interfaces is generally treated as a functional property rather than explicitly verified as a security property. This work identifies and experimentally demonstrates a directionality-based class of analog and mixed-signal (AMS) exfiltration attacks in which data-dependent circuit-offset modulation converts a nominally input-only pin into an outbound information channel. We analytically model the attack mechanism and identify three enabling host conditions: a closed-loop amplifier, an exposed amplifier input, and sufficiently high impedance at that pin. This attack class is validated through a representative silicon case study using a photoplethysmography (PPG) analog front-end (AFE) fabricated in a commercial 55-nm CMOS process. The payload incurs $<$0.001\% area overhead relative to typical biosensing AFEs. Under the evaluated conditions, payload activation reduces the filtered PPG-output SNR by only 0.03~dB, while the maximum HT-induced perturbation of 5.9\% of the PPG amplitude remains within the 34.3\% benign variation at the exposed sensor-input pin across process and temperature. The raw exfiltration SINR remains below -20~dB, while targeted filtering increases it above 14~dB and enables signal recovery. Silicon measurements demonstrate data exfiltration through the input pin at bit rates up to 10~kbps and error-free recovery of a PRBS message. These results expose a conventional test-observability gap and establish analog pin directionality as an AMS security property requiring explicit verification, test coverage, and defense rather than being inferred from nominal signal flow.
cs.CR / 31 / 2609.19140
AgentLSD: Evaluating AI Security Agents Under Adversarial Task Contamination
Abstract
AI agents for security inspect web pages, source code, logs, configuration files, and command outputs. These environments may contain deceptive artifacts that influence the agent's behavior. We call this adversarial task contamination. Whereas prompt injection relies on attacker-supplied instructions, task contamination also includes non-instructional evidence, such as fake results and decoy endpoints. We present AgentLSD, a controlled framework for studying adversarial task contamination. AgentLSD uses Capture the Flag (CTF) challenges as its experimental environment. We inject trap artifacts, such as fake flags, misleading hints, decoy endpoints, and hidden cues, while preserving the intended CTF solution. The framework supports paired clean and trap-augmented experiments with deterministic trap generation, runtime injection, telemetry, and delivery verification. We evaluate six models on 11 web CTF challenges. In the clean condition, agents capture 41% of the flags, and no model solves every challenge. We then measure the impact of task contamination. Even when the agent still recovers the flag, traps increase the number of turns (+20) and reasoning tokens (+2k). Solve-rate effects are more heterogeneous, as some model-challenge pairs are largely unaffected while others follow decoys or submit wrong flags. These results show that clean CTF performance understates vulnerability to deceptive task evidence. AgentLSD isolates this effect and provides a reproducible benchmark for studying it. We release the framework, configurations, trap specifications, and raw traces.
cs.CR / 32 / 2609.17856
Investigating Adversarial Robustness of Heterogeneous Cooperative Perception
Abstract
Heterogeneous cooperative perception (CP) enables connected vehicles with diverse sensor setups to share spatial awareness via compact feature maps, where receivers reconcile these maps using learned translation modules for fusion and inference. Prior attacks against CP in a homogeneous setting reveal that the data exchange introduces a critical attack surface: a single malicious agent can transmit crafted features that erase real objects from a neighbor's fused scene. Yet, it is widely hypothesized that heterogeneity naturally defends against these attacks, as the attacker lacks knowledge of the victim's detector and the translation module scrambles adversarial gradients. We demonstrate that this protection is largely an illusion. Using a matched-objective harness to standardize the perturbation budget, objective, and forward path, we show that properly tuned iterative attacks close or reverse the apparent robustness gap. However, these optimization-based attacks require ground-truth labels and iterative backpropagation, meaning they do not represent a practical field threat running in real-time. To bridge this gap, we introduce HetPoison, a learned generator that crafts a removal perturbation in a single, label-free forward pass. HetPoison transfers across major heterogeneous designs without requiring access to the victim's detector, matching or exceeding the effectiveness of expensive optimizer-based attacks. Since heterogeneity itself is not a defense, we propose HetShield, a lightweight trust layer that validates the spatiotemporal consistency across features, recovering 83--95% of the accuracy degraded by attacks, outperforming prior art.
cs.CR / 33 / 2609.18886
Fluid Notarization: Verifiable Evolution of Concurrently Edited Structured Documents
Abstract
Traditional blockchain-based document notarization follows a snapshot-oriented model in which each document revision is represented as an independent state anchored on-chain through a cryptographic reference. While effective for immutable artifacts, this approach becomes inadequate when documents evolve through collaborative editing. Concurrent modifications create divergent document versions that must be reconciled outside the notarization layer, while even minor changes require generating and distributing new document snapshots. Conversely, collaborative replication frameworks such as CRDTs provide deterministic reconciliation of concurrent updates, but do not inherently provide independently verifiable evidence of when contributions were published. This paper introduces Fluid Notarization, a notarization paradigm in which document evolution itself becomes the object of notarization. Rather than certifying isolated states, Fluid Notarization certifies a graph of causally related evolution artifacts generated by a JSON-native delta-CRDT. The proposed model builds upon Melda, which represents document changes as compact, content-addressed deltas linked through causal dependencies. Blockchain notarization is reduced to recording identifiers of these evolution artifacts, while synchronization, reconstruction, and conflict resolution remain entirely off-chain. The resulting architecture combines two complementary guarantees: deterministic convergence provided by the CRDT and independently auditable proof-of-existence, provenance, and publication evidence provided by the blockchain. A prototype implementation and validation scenario based on collaboratively edited electronic health records demonstrate the feasibility of the approach and highlight the advantages of notarizing document evolution rather than successive document snapshots.
cs.CR / 34 / 2609.18829
Epsilon-Nash Equilibria in History-Dependent SA-MDPs
Abstract
We study state-adversarial Markov decision processes (SA-MDP) as a game of observation-space attacks: at each step, an agent selects an action from a received observation while an adversary$\unicode{x2014}$who knows the true state the agent is in$\unicode{x2014}$chooses a perturbed observation within a state-dependent proximity set. While existing work focuses on Markovian policies, we develop a solution concept and computational approach for SA-MDPs under history dependence. This is motivated by results showing that history dependence can materially change equilibrium outcomes and can force both the agent and the adversary to adapt their strategies. First, we prove the non-existence of universal (agnostic of the initial state distribution) history-dependent equilibrium policies. In response to this finding, our main result presents the first algorithmic route to computing $ε$-approximations of initial-state dependent equilibria. We do so by reducing SA-MDPs to a strategically equivalent constrained zero-sum one-sided partially observable stochastic game. We conclude by testing our algorithm on small analytically verifiable games and showing it scales to larger, more realistic benchmarks, including Atari Freeway rollouts with a 12-period ahead horizon.
cs.CR / 35 / 2609.18876
Low-Rank Masking for Single-Server Matrix Multiplication
Abstract
We study the statistical privacy of outsourcing matrix multiplication over a finite field ${\mathbb F_q}$ to a single server using additive masks of rank at most $r$. For independent uniform $n\times n$ inputs, we show that uniform \emph{rank-ball masks} and products of independent uniform factors give maximal-correlation secrecy of at most $q^{-r}$ against the complete server view, with $O(n^2r)$ field operations for encoding and decoding. This secrecy captures how effectively the server is prevented from estimating functions of the inputs. We prove an asymptotically matching lower bound of this secrecy measure for $r=o(n)$, showing that both sampling methods are asymptotically optimal among input-independent additive masks of rank at most $r$, even when secret invertible transformations are allowed. We also characterize the posterior distribution for uniform rank-ball masks under arbitrary joint input distributions and prove approximate individual security for rows and columns under independent uniform inputs. Finally, we show that every input-independent additive mask of rank at most $r=o(n)$ requires $δ\to1$ in entry-level $(\varepsilon,δ)$-differential privacy for fixed field size $q$ and bounded $\varepsilon$.