Daily Research Digest
arXiv Papers
2026-09-09
349
Papers
8
Categories
69
Translated
收藏清单 0
精选 · Favorites
69
cs.AI / 1 / 2609.07879
Do Large Language Models Know What They Don't Know II? A Fully Behavioral, Non-Cognitive Measure of Epistemic Honesty
大型语言模型是否知道它们不知道的事？第二部分：一种完全行为性的、非认知性的认识论诚实度量
large language model
大语言模型相关
Abstract
Large Language Models (LLMs) are frequently confident, eloquent, and well versed. A natural question arises: do they know what they don't know? To answer this question, we borrow the concept of epistemic honesty and develop a novel metric to systematically evaluate whether an LLM appropriately acknowledges the boundaries of its knowledge. In this work, we introduce the Epistemic Honesty Quotient (EHQ), which reports three observable sub-scores across two operational axes (epistemic restraint and substantive-answer calibration), and construct EHQ-3000, a 3,000-question benchmark spanning Fabricated Entity, Post-Cutoff Event, Hyper-Niche True, and Context-Conditioned Questions. From a frozen registry of 21 model API routes, 15 completed the protocol after endpoint and eligibility checks; 14 entered the confirmatory analysis because severe provider-side truncation made one route's score indeterminate. The study reveals substantial variation across models, including a difference that can not be explained by their capability to extract explicitly available information. Composite EHQ ranges from 0.31 to 0.81 across the analysed panel, despite near-ceiling performance on the document-grounded capability probe. The two restraint criteria overlap strongly under the present category composition, whereas substantive-answer calibration varies across models and does not reliably co-vary with restraint; however, the small panel leaves substantial uncertainty. Thus, EHQ reveals behavioral differences that are not visible to conventional correctness-based assessment, while also showing why dataset composition, provider behavior, and confidence elicitation must remain part of the interpretation.
Chinese Translation
大型语言模型（LLMs）经常表现得自信、雄辩且见多识广。一个自然的问题出现了：它们知道自己不知道什么吗？为了回答这个问题，我们借鉴了认识论诚实的概念，并开发了一种新的度量指标，以系统评估一个LLM是否恰当承认其知识的边界。在本工作中，我们引入了认识论诚实商数（EHQ），它跨两个操作性维度（认识论约束和实质性答案校准）报告三个可观察的子分数，并构建了 EHQ-3000，一个包含3,000个问题、涵盖虚构实体、截止后事件、极度冷门真实和上下文条件问题的基准。在包含21条模型API路由的冻结登记清单中，15条在端点和资格检查后完成了协议；14条进入确认性分析，因为供应商侧的严重截断使一条路由的得分不确定。该研究揭示了模型间的显著差异，包括一种无法用她们提取显式可用信息的能力来解释的差异。尽管在基于文档的能力探测中表现接近上限，所分析面板中的综合EHQ从0.31到0.81不等。在当前类别组成下，两项约束标准高度重叠，而实质性答案校准因模型而异，且与约束没有可靠的共变关系；然而，较小的面板留下了很大的不确定性。因此，EHQ揭示了基于正确性的传统评估所看不到的行为差异，同时也说明了为什么数据集组成、供应商行为与置信度引出必须始终纳入解读范围。
cs.AI / 2 / 2609.07943
Beliefs and Behavior in Language Models
语言模型中的信念与行为
large language model
大语言模型相关
Abstract
There is significant uncertainty about whether abstractions like beliefs or desires usefully describe the behavior of large language models (LLMs). In addition to the inherent scientific interest of this question, these latent quantities are often invoked to explain the behavior of LLMs to users or to define and evaluate harmful behaviors which are relative to intent. Nevertheless, we currently lack a means to systematically test whether concepts like "belief" are well-applied to LLMs, and hence whether they are likely to be fruitful ingredients of attempts to align models with human interests. We propose an approach for empirically studying such questions, asking whether a single latent variable inferred from the LLMs' outputs -- interpreted as a degree of belief -- allows an observer to make interpretable predictions of how the LLMs' will respond to new prompts. We find that highly capable models are usefully described as holding beliefs and that, generally, the predictability of model outputs based on an inferred latent belief tracks overall trends in model capability. Building on these findings, we provide empirical strategies to study how beliefs in LLMs can be measured, the extent to which LLMs comply with instructed decision rules or payoffs, and how beliefs evolve within individual instances of an LLM over the course of reasoning.
Chinese Translation
关于像信念或欲望这样的抽象概念能否有效地描述大型语言模型（LLMs）的行为，存在显著的不确定性。除了这一问题本身固有的科学意义外，这些潜在量常常被用来向用户解释LLMs的行为，或定义和评估与意图相关的有害行为。然而，我们目前缺乏一种系统测试的方法，来检验像“信念”这样的概念是否很好地适用于LLMs，以及它们是否可能成为将模型与人类利益对齐的努力中富有成效的组成部分。我们提出了一种实证研究方法，询问从LLM输出中推断出的单一潜在变量——被解释为信念程度——是否能让观察者对新提示下LLM将如何回应做出可解释的预测。我们发现，高能力的模型可以被有效地描述为持有信念，而且一般来说，基于推断出的潜在信念对模型输出的可预测性，会跟随模型能力的整体趋势。在这些发现的基础上，我们提供了实证策略，以研究如何测量LLM中的信念、LLM在多大程度上遵循指令性的决策规则或收益，以及在一个LLM的单个实例中，信念在推理过程中如何演变。
cs.AI / 3 / 2609.08071
Automated Design of Inventory Policy with Large Language Models: An Exploratory Study
基于大语言模型的库存策略自动设计：一项探索性研究
large language model
大语言模型相关
Abstract
Firms making inventory decisions have access to operational data, optimization tools, and large language models (LLMs). Typically, data characterize the operating environment, optimization selects parameters within a prespecified inventory policy class, and LLMs support coding and decision analysis. We develop an integrated framework that combines these resources to automate inventory policy design. Given demand data, the framework iteratively uses an LLM to generate parameterized policy classes and an external solver to optimize its parameters within each class. Across 30 lost-sales inventory instances, the mean cost reduction relative to optimized base-stock benchmarks increases from 17.5% after one generation to 30.0% after ten generations. Parameter optimization is central to this performance: an LLM-only variant performs substantially worse, whereas optimization-guided feedback improves policy quality, accelerates search, and directs the LLM toward better policy classes rather than merely better parameter values within a fixed class. The strongest discovered policies are also interpretable: they combine recognizable inventory-control motifs, including capped orders, discounted or weighted pipeline inventory, and threshold-based replenishment logic. The search thereby produces new policy-class functional forms that, to our knowledge, have not previously been studied in the lost-sales inventory literature. These functional forms are not specified ex ante but emerge from the search process. Moreover, after their parameters are re-optimized, three discovered policy classes achieve average cost reductions of 21.75% to 22.60% across 10,064 new inventory instances. Overall, the results show that data-driven parameter optimization can guide LLM-based search over a broad space of inventory policy classes and identify high-performing, interpretable, and transferable decision rules.
Chinese Translation
做出库存决策的企业可以访问运营数据、优化工具和大语言模型（LLMs）。通常，数据描述运营环境，优化在预先指定的库存策略类别内选择参数，而LLMs支持编码和决策分析。我们开发了一个集成框架，结合这些资源来自动化库存策略设计。给定需求数据，该框架迭代地使用LLM生成参数化策略类别，并使用外部求解器在每个类别内优化其参数。在30个缺货库存实例中，相对于优化基础库存基准的平均成本降低从一代后的17.5%增加到十代后的30.0%。参数优化对这一性能至关重要：仅使用LLM的变体表现明显更差，而优化引导的反馈提升了策略质量、加速了搜索，并引导LLM朝向更好的策略类别，而不仅仅是固定类别内更好的参数值。发现的最强策略也具有可解释性：它们结合了可识别的库存控制模式，包括有上限的订单、折扣或加权的在途库存，以及基于阈值的补货逻辑。该搜索因此产生了新的策略类别函数形式，据我们所知，这些形式此前在缺货库存文献中从未被研究过。这些函数形式并非事先指定，而是从搜索过程中涌现出来的。此外，在对其参数重新优化后，三个发现的策略类别在10,064个新库存实例上实现了21.75%至22.60%的平均成本降低。总体而言，结果表明数据驱动的参数优化可以引导基于LLM的搜索在广阔的库存策略类别空间中进行，并识别出高性能、可解释且可迁移的决策规则。
cs.AI / 4 / 2609.08094
CIVI: A Framework for Diagnosing Search Agent Failures in Civic Information
CIVI：一个用于诊断公民信息中搜索智能体失败的框架
large language model
大语言模型相关
Abstract
Large Language Models are increasingly deployed in public-sector settings, where incorrect guidance can cause irreversible harm. We introduce CIVI, the first framework for diagnosing search agent failures in civic information. Its benchmark instantiation jointly spans cross-national, interjurisdictional government contexts (federal, state, and local) and functional categories from an internationally adopted United Nations standard. We evaluate ten frontier search agents and find that none matches an attentive human baseline. Alongside accuracy, CIVI measures search invocation rate, selective no-search accuracy, and how often agents cite authoritative government sources. To perform this diagnosis, we introduce ARISE, which decomposes agentic search failures into four mutually exclusive modes, isolated via source-injection ablation. ARISE attributes 72.1% of all observed failures to retrieval-bound causes rather than to gaps in the models' parametric knowledge.
Chinese Translation
大型语言模型越来越多地部署在公共部门环境中，在这些环境中，错误的指导可能会造成不可逆转的损害。我们介绍了CIVI，这是第一个用于诊断公民信息中搜索智能体失败的框架。其基准实例化共同涵盖了跨国家、跨辖区的政府背景（联邦、州和地方）以及来自一项国际采用的联合国标准的功能类别。我们评估了十个前沿搜索智能体，发现没有一个能与专注的人类基线相匹配。除了准确性之外，CIVI还衡量搜索调用率、选择性不搜索的准确率，以及智能体引用权威政府来源的频率。为了执行这种诊断，我们引入了ARISE，它将智能体搜索失败分解为四种互斥模式，通过源注入消融来隔离。ARISE将所有观察到的失败中的72.1%归因于检索相关的原因，而非模型参数知识中的空白。
cs.AI / 5 / 2609.08180
Less Is Personal: Learning Minimal Sufficient User Profiles for Personalized Language Models
少即是个人化：为个性化语言模型学习最小充分用户画像
large language model
大语言模型相关
Abstract
Retrieval-augmented personalization enables large language models to produce more accurate and preference-aligned outputs using relevant records retrieved from user histories. Personalized language models typically prepend a fixed number of retrieved user records, even when additional history is redundant, harmful, or unrelated to a user's distinctive behavior. We study minimal sufficient personalization: constructing the least costly ordered profile for each input while preserving the utility achievable from a retrieved candidate pool. We introduce ENOUGH, a method that iteratively appends behavioral records or emits STOP to construct profiles with adaptive lengths. Offline, bounded counterfactual search evaluates profile prefixes by jointly considering downstream gains, user specificity, and token costs. The resulting long-horizon targets are distilled into a multi-head value controller with explicit ranking and stopping supervision. At inference, the controller selects and orders records through lightweight decisions, and the frozen generator is invoked once after stopping. Extensive experiments on six personalized tasks demonstrate that ENOUGH consistently outperforms strong heuristic and retrieval-augmented baselines in both effectiveness and efficiency, achieving minimal sufficient profiles that preserve personalization utility while reducing unnecessary context costs.
Chinese Translation
检索增强的个性化使得大语言模型能够利用从用户历史中检索到的相关记录，生成更准确且更符合偏好的输出。个性化语言模型通常会前置固定数量的检索到的用户记录，即使额外的历史信息是冗余的、有害的，或与用户的独特行为无关。我们研究最小充分个性化：为每个输入构建成本最低的有序画像，同时保留从检索候选中可获得的效果。我们提出了ENOUGH，一种通过迭代追加行为记录或发出STOP来构建自适应长度画像的方法。在离线阶段，有界反事实搜索通过联合考虑下游收益、用户特异性和token成本来评估画像前缀。由此产生的长时程目标被蒸馏到一个多头价值控制器中，并带有显式的排序和停止监督。在推理时，控制器通过轻量级决策选择并排序记录，冻结的生成器在停止后被调用一次。在六个个性化任务上的大量实验表明，ENOUGH在有效性和效率方面始终优于强启发式基线和检索增强基线，实现了在保留个性化效用的同时减少不必要上下文成本的最小充分画像。
cs.AI / 6 / 2609.08189
Do Dynamic Routers Need Memory? HeRo: History-Aware Routing for Efficient LLM Inference
动态路由器需要记忆吗？HeRo：面向高效LLM推理的历史感知路由
large language model
大语言模型相关
Abstract
Dynamic layer routing reduces the inference cost of Large Language Models (LLMs) by learning to skip layers for individual tokens. Existing methods, however, treat each routing decision as a local operation conditioned solely on the current hidden state which is a formulation that overlooks the sequential, path-dependent nature of routing across depth: earlier decisions shape the representations seen by downstream routers, and the layer-usage objective couples all decisions jointly. We propose History-Aware Routing (HeRo), a dynamic routing framework that resolves this mismatch by introducing a router memory mechanism to maintain an explicit routing state across model depth. The memory is constructed via linear attention, incrementally aggregating preceding routing scores and their induced residual updates into a compact history representation. At each routed layer, the router conditions jointly on this accumulated state and the current hidden representation to select the executed branch. Instantiated for token-wise FFN routing, HeRo trains only lightweight routers and adapters on a frozen backbone, requiring no modification to pretrained parameters. Across Llama 3.1-8B, Llama 2-7B, and Llama 2-13B, HeRo consistently achieves the highest aggregate performance retention among ten baselines. On Llama 3.1-8B, it bypasses 26.87% of model parameters while achieving 100.24% of dense model performance across seven benchmarks, and retains 97.01% while bypassing 38.82% of model parameters under a tighter computation budget. Ablation studies confirm that removing routing history consistently degrades performance, most notably on multistep reasoning and code generation, validating that explicit routing memory enables more accurate and adaptive dynamic routing than solely conditioning on hidden state.
Chinese Translation
动态层路由通过为单个标记学习跳过层来降低大型语言模型（LLM）的推理成本。然而，现有方法将每个路由决策视为仅以当前隐藏状态为条件的局部操作，这种表述忽略了跨深度路由的序列性和路径依赖性：早期决策塑造了下游路由器所看到的表示，而层使用目标则联合耦合了所有决策。我们提出历史感知路由（HeRo），这是一种动态路由框架，通过引入路由器记忆机制来维护跨模型深度的显式路由状态，从而解决这一不匹配问题。该记忆通过线性注意力构建，将先前的路由分数及其引起的残差更新增量聚合为紧凑的历史表示。在每个路由层，路由器同时基于该累积状态和当前隐藏表示进行条件判断，以选择所执行的分支。在针对逐标记FFN路由的实例化中，HeRo仅在冻结骨干上训练轻量级路由器和适配器，无需修改预训练参数。在Llama 3.1-8B、Llama 2-7B和Llama 2-13B上，HeRo在十个基线中始终取得最高的总体性能保持率。在Llama 3.1-8B上，它绕过26.87%的模型参数，同时在七个基准测试中达到稠密模型性能的100.24%，并在更紧的计算预算下绕过38.82%的模型参数时保持97.01%的性能。消融研究证实，移除路由历史总是会降低性能，在多步推理和代码生成上尤为显著，这验证了显式路由记忆比仅以隐藏状态为条件能够实现更准确和更自适应的动态路由。
cs.AI / 7 / 2609.08216
Vision: Data-Centric Anchoring for Robust and Interpretable Agentic AI
愿景：以数据为中心的锚定，实现稳健且可解释的智能体AI
large language model
大语言模型相关
Abstract
Agentic AI systems built on large language models fail in two persistent ways that scaling does not fix: they break under distribution shift, and they cannot explain the decisions they make. We argue these are co-symptoms of one structural deficiency in the data lifecycle that governs how agents are trained, evaluated, and deployed. Observational interaction logs record what an agent did, not what it would have done otherwise. They encode spurious correlations without controlled variation, so they lack the counterfactual structure needed to separate causal signal from coincidence or to validate an explanation. No model-centric method can recover invariances the data never contained. We present Data-Centric Anchoring: robustness and interpretability should be engineered into the data environment, not extracted from models after training. Our central contribution is the Data-Centric Agentic Loop, a four-stage framework of Curate, Augment, Constrain, and Attribute. The ordering is structural, not stylistic. Curation precedes augmentation because generative models amplify whatever bias they are trained on. Augmentation precedes constraint because invariance objectives are vacuous without variation across environments to be invariant to. Attribution closes the loop, converting observed failures into targeted data interventions for the next iteration. Each stage manufactures the preconditions of the next, which makes the loop self-correcting rather than merely sequential. We ground the framework in a failure-driven taxonomy that links four core failure modes to the data lifecycle: spurious feature reliance, distribution-shift fragility, uncertainty miscalibration, and explanation unfaithfulness. We close with the limits of this approach and the open problems that stand between it and practical deployment at scale.
Chinese Translation
基于大型语言模型构建的智能体AI系统存在两种持续性的失败方式，扩展规模无法解决这些问题：它们在分布偏移下崩溃，并且无法解释自己做出的决策。我们认为，这些是数据生命周期中一个结构性缺陷的共同症状，而该数据生命周期支配着智能体的训练、评估和部署方式。观测性交互日志记录的是智能体实际做了什么，而不是它在其他情况下本会做什么。它们编码了虚假相关性，却缺乏受控变异，因此缺少将因果信号与巧合区分开来或验证解释所需的反事实结构。没有任何以模型为中心的方法能够恢复数据从未包含的不变性。我们提出“以数据为中心的锚定”（Data-Centric Anchoring）：稳健性和可解释性应当被设计进数据环境中，而非在训练后从模型中提取。我们的核心贡献是“以数据为中心的智能体循环”（Data-Centric Agentic Loop），一个由“策展”（Curate）、“增强”（Augment）、“约束”（Constrain）和“归因”（Attribute）四个阶段组成的框架。这种顺序是结构性的，而非风格性的。策展先于增强，因为生成模型会放大它们所训练的任何偏见。增强先于约束，因为如果没有跨环境的变异可供保持不变，不变性目标便是空洞的。归因闭合循环，将观察到的失败转化为下一轮迭代中针对性的数据干预。每个阶段都为下一阶段制造前提条件，这使得该循环是自我纠正的，而不仅仅是顺序执行的。我们将该框架建立在一种由失败驱动的分类法上，该分类法将四种核心失败模式与数据生命周期联系起来：虚假特征依赖、分布偏移脆弱性、不确定性校准错误和解释不忠实。最后，我们讨论了这种方法的局限性，以及横亘在它与大规模实际部署之间的开放问题。
cs.AI / 8 / 2609.08226
TTGBench: Benchmarking Topological Evolution and Semantic Drift in Text-attributed Temporal Graphs
TTGBench：文本属性时序图中的拓扑演化与语义漂移基准测试
large language model
大语言模型相关
Abstract
Temporal graph learning models the evolution of dynamic systems, where both structural interactions and semantic states change over time. However, existing benchmarks primarily emphasize structural evolution via temporal link prediction (TLP), while support for semantic evolution remains limited. Although temporal node classification (TNC) is sometimes included, it is typically restricted to simplistic binary settings that fail to capture realistic semantic drift. Moreover, commonly used datasets exhibit high link repetition, leading to inflated performance estimates and obscuring true model capability. To address these limitations, we introduce \textbf{TTGBench}, a new benchmark that jointly evaluates structural and semantic evolution. TTGBench comprises six real-world, text-rich datasets characterized by \emph{Dual Volatility}, enabling rigorous and fair evaluation of existing models. Notably, it is the first benchmark to support both multi-class and multi-label TNC, filling a critical gap in evaluating temporal semantic drift. We conduct a comprehensive evaluation of 17 state-of-the-art methods across Temporal Graph Neural Networks (TGNNs) and Large Language Model (LLM)-based paradigms. The results reveal a clear \emph{capability divide} between the two paradigms: TGNN-based methods excel at structural prediction but fail at semantic tracking, whereas LLM-based predictors show the opposite trend. Through in-depth analysis, we uncover their fundamental limitations and provide insights for developing more comprehensive temporal graph models.
Chinese Translation
时序图学习对动态系统的演化进行建模，其中结构交互和语义状态均随时间变化。然而，现有基准主要通过时序链接预测（TLP）强调结构演化，而对语义演化的支持仍然有限。尽管有时会包含时序节点分类（TNC），但它通常局限于过于简单的二分类设置，无法捕捉真实的语义漂移。此外，常用的数据集呈现出较高的链接重复性，导致性能估计虚高，并掩盖了模型的真实能力。为解决这些局限，我们提出了TTGBench，一个联合评估结构演化与语义演化的新型基准。TTGBench由六个真实世界、文本丰富的数据集组成，具有“双重波动性”（Dual Volatility）的特征，从而能够对现有模型进行严谨而公平的评估。值得注意的是，它是首个同时支持多分类和多标签时序节点分类（TNC）的基准，填补了评估时序语义漂移的关键空白。我们全面评估了时序图神经网络（TGNNs）和基于大型语言模型（LLM）这两种范式下的17种最先进方法。结果揭示了两种范式之间存在明显的“能力鸿沟”：基于TGNN的方法在结构预测上表现出色，但在语义跟踪上表现不佳，而基于LLM的预测器则表现出相反的趋势。通过深入分析，我们揭示了它们的根本局限，并为开发更全面的时序图模型提供了见解。
cs.AI / 9 / 2609.08267
Evidence-Aligned Entity Verification for Hallucination Detection in Retrieval-Augmented Generation
证据对齐的实体验证用于检索增强生成中的幻觉检测
large language model
大语言模型相关
Abstract
Hallucination detection is crucial for large language models (LLMs), as hallucinated content creates significant barriers in applications requiring factual accuracy. Current detection methods mainly depend on internal signals like uncertainty and self-consistency checks, using the model's pre-trained knowledge to identify unreliable outputs. However, pre-trained knowledge may become outdated and has coverage limitations, especially for specialized or recent information. To address these limitations, retrieval-augmented generation (RAG) has emerged as a promising solution by retrieving relevant evidence at inference time, grounding outputs beyond the model's parametric knowledge. In this paper, we target a critical and practical learning problem RAG-based hallucination detection (RHD), where RAG is employed to enhance hallucination detection by addressing information updating challenges. To address RHD, we propose a novel method Evidence-Aligned Entity Verification (EAEV), which detects entity-level hallucinations by leveraging RAG to align generated entities with retrieved evidence contexts. Specifically, EAEV evaluates entity-evidence alignment through three complementary dimensions and introduces counterfactual stability analysis to ensure robust alignments under evidence perturbations. Experiments across multiple RAG benchmarks demonstrate that EAEV achieves consistent improvements over existing methods with strong generalization capabilities.
Chinese Translation
幻觉检测对于大语言模型（LLMs）至关重要，因为幻觉内容在需要事实准确性的应用中会造成重大障碍。当前的检测方法主要依赖内部信号，如不确定性和自一致性检查，利用模型的预训练知识来识别不可靠的输出。然而，预训练知识可能过时且存在覆盖范围限制，尤其是对于专业或最新信息。为解决这些限制，检索增强生成（RAG）通过在推理时检索相关证据，将输出锚定在模型参数化知识之外，已成为一种有前景的解决方案。在本文中，我们针对一个关键且实用的学习问题——基于RAG的幻觉检测（RHD），其中使用RAG通过解决信息更新挑战来增强幻觉检测。为解决RHD，我们提出了一种新方法——证据对齐的实体验证（EAEV），该方法通过利用RAG将生成的实体与检索到的证据上下文进行对齐，来检测实体级别的幻觉。具体而言，EAEV通过三个互补维度评估实体-证据对齐，并引入反事实稳定性分析，以确保在证据扰动下的稳健对齐。在多个RAG基准上的实验表明，EAEV相比现有方法取得了持续改进，并具有较强的泛化能力。
cs.AI / 10 / 2609.08288
LEBGen: An LLM-Enhanced Bayesian Network Framework for Few-Shot Travel Survey Data Generation
LEBGen：一种用于少样本出行调查数据生成的LLM增强贝叶斯网络框架
large language model
大语言模型相关
Abstract
Travel survey data are essential for transportation planning and travel behavior analysis, yet collecting large-scale representative samples is costly and time-consuming. A practical alternative is to generate synthetic survey records from a few-shot sample. However, such samples provide incomplete coverage of heterogeneous traveler groups and insufficient evidence for recovering the complex dependencies between demographic characteristics and travel behavior. Existing approaches have complementary limitations. Probabilistic generative models such as Bayesian networks (BNs) offer explicit distributional control, but structures learned from few-shot samples may omit meaningful dependencies or retain spurious ones. Large language models (LLMs) can help address these difficulties in BN structure learning by providing behavioral knowledge that complements the limited statistical evidence. We therefore propose LEBGen, an LLM-enhanced BN framework that uses this knowledge to refine network structure for few-shot travel survey data generation. Specifically, the LLM first identifies traveler personas from demographic attribute and travel behavior statistics, then recovers dependencies missed by the persona-augmented BN structure and prune spurious ones. The refined BN is parameterized exclusively from the observed data to generate synthetic records. Under a 2% few-shot setting on the 2022 Hong Kong Travel Characteristics Survey, LEBGen reduces the mean marginal Jensen-Shannon divergence from 0.0671 to 0.0091 and the mean absolute Cramer's V error by 14.3% over the best-performing baseline, substantially improving both distributional and dependency fidelity.
Chinese Translation
旅行调查数据对于交通规划和出行行为分析至关重要，然而收集大规模代表性样本既昂贵又耗时。一种实用的替代方案是从少样本（few-shot）样本中生成合成调查记录。然而，此类样本对异质出行者群体的覆盖不完整，并且为恢复人口统计特征与出行行为之间的复杂依赖关系提供的证据不足。现有方法各有其互补的局限性。诸如贝叶斯网络（BN）之类的概率生成模型提供了显式的分布控制，但从少样本中学习到的结构可能会遗漏有意义的依赖关系或保留虚假的依赖关系。大语言模型（LLM）可以通过提供补充有限统计证据的行为知识，帮助解决BN结构学习中的这些困难。因此，我们提出了LEBGen，一种LLM增强的BN框架，利用这一知识来优化网络结构，以用于少样本旅行调查数据生成。具体而言，LLM首先从人口统计属性和出行行为统计中识别出行者角色（personas），然后恢复被角色增强的BN结构遗漏的依赖关系，并修剪虚假的依赖关系。优化后的BN仅从观测数据进行参数化，以生成合成记录。在2022年香港出行特征调查的2%少样本设置下，与表现最佳的基线相比，LEBGen将平均边际Jensen-Shannon散度从0.0671降至0.0091，并将平均绝对Cramer's V误差降低了14.3%，显著提升了分布保真度和依赖保真度。
cs.AI / 11 / 2609.08452
SRPO: Setwise Relative Policy Optimization for Multi-Agent LLMs
SRPO：用于多智能体大语言模型的集合级相对策略优化
large language model
大语言模型相关
Abstract
Multi-agent large language models solve complex tasks by coordinating several policies in a shared environment. However, existing reinforcement learning methods usually optimize each response or trajectory separately, even when several outputs jointly cause one state transition. Consequently, the update unit differs from the action executed by the system. To address this problem, we propose SRPO (Setwise Relative Policy Optimization), which treats the active set the minimal set of outputs consumed by one transition, as one multi-agent action. Specifically, SRPO combines member log-ratios into one cardinality-normalized set ratio, assigns one relative advantage, and clips the set once. This formulation unifies division of labor and joint co-evolution as actions with different set sizes. Experiments on mathematical reasoning and multi-turn search demonstrate one training interface for fixed, mixed, and dynamically routed workflows across four model scales, with the strongest macro-average results among the reported comparisons. Optimization diagnostics further characterize its stability under different event reductions and set sizes.
Chinese Translation
多智能体大语言模型通过协调多个策略在共享环境中执行复杂任务。然而，现有的强化学习方法通常单独优化每个响应或轨迹，即使多个输出共同导致一次状态转移时也是如此。因此，更新单元与系统所执行的动作不一致。为解决这一问题，我们提出了SRPO（集合级相对策略优化），它将活动集合——即一次转移所消耗的最小输出集合——视为一个多智能体动作。具体而言，SRPO将成员的对数比值合并为一个基数归一化的集合比值，分配一个相对优势值，并对集合进行一次裁剪。这种表述将分工与联合共同演化统一为具有不同集合规模的动作。在数学推理和多轮搜索上的实验表明，在四种模型规模下，该训练接口适用于固定、混合及动态路由的工作流，并在所报告的对比中取得了最强的宏平均结果。优化诊断进一步表征了其在不同事件归约和集合规模下的稳定性。
cs.AI / 12 / 2609.08572
AgentGrad: Intervention-guided Prompt Optimization for Multi Agent Systems
AgentGrad: 干预引导的多智能体系统提示优化
large language model
大语言模型相关
Abstract
Large language model (LLM)-based multi-agent systems (MAS) achieve strong performance by employing specialized multiple agents, yet their performance depends on the prompt design of each agent. For MAS prompt optimization, textual gradient methods that guide prompt updates using natural-language feedback have emerged as a leading paradigm. In this paper, we identify limitations in two stages of existing textual gradient approaches: gradient extraction and gradient aggregation. In gradient extraction, previous works select a target prompt without verifying whether modifying it resolves the failure, and derive gradients without agent-level supervision over the corresponding agent's intermediate output. In gradient aggregation, individual gradients are randomly grouped and concatenated, often mixing unrelated failure modes and producing prompts that fail to generalize. To address these limitations, we propose \textbf{AgentGrad}, a prompt optimization framework for multi-agent systems based on sequential intervention and semantic textual gradient abstraction. For each failure, sequential intervention modifies the behavior of one agent at a time to identify the target agent whose modification resolves the failure. The modified output of the target agent then serves as agent-level supervision for extracting a fine-grained gradient. Semantic textual gradient abstraction clusters semantically similar gradients to prevent mixing unrelated failure modes, and abstracts each cluster into a generalized gradient that captures the shared corrective pattern. Experimental results show that AgentGrad achieves state-of-the-art performance across five MAS benchmarks and reduces wall-clock optimization time by $2.5\times$ on average compared to the next-fastest baseline.
Chinese Translation
基于大型语言模型（LLM）的多智能体系统（MAS）通过采用多个专门化智能体实现了强大性能，但其性能取决于每个智能体的提示设计。对于MAS提示优化，利用自然语言反馈指导提示更新的文本梯度方法已成为领先范式。在本文中，我们指出现有文本梯度方法在梯度提取和梯度聚合两个阶段存在的局限性。在梯度提取阶段，先前的工作选择目标提示时不验证修改它是否能解决故障，并且在没有对应智能体中间输出的智能体级监督下推导梯度。在梯度聚合阶段，单个梯度被随机分组并拼接，常常混合不相关的故障模式，并产生无法泛化的提示。为解决这些局限性，我们提出AgentGrad，一种基于顺序干预和语义文本梯度抽象的多智能体系统提示优化框架。对于每个故障，顺序干预一次只修改一个智能体的行为，以识别其修改能解决故障的目标智能体。然后，目标智能体修改后的输出作为智能体级监督，用于提取细粒度梯度。语义文本梯度抽象将语义相似的梯度聚类，以防止混合不相关的故障模式，并将每个聚类抽象为捕捉共享修正模式的广义梯度。实验结果表明，AgentGrad在五个MAS基准上取得了最先进性能，并且与次快基线相比，平均将实际优化时间减少了$2.5\times$。
cs.AI / 13 / 2609.08599
Graph-Based Personalized Memory for LLM Agents: Representation, Evolution, Retrieval, and Evaluation
基于图的 LLM 智能体个性化记忆：表示、演化、检索与评估
large language model
大语言模型相关
Abstract
Large Language Model (LLM) agents are evolving from single-session tools toward long-term personal assistants that must adapt to individual users across tasks, contexts, and interactions. This shift makes memory a core requirement for personalization, since user preferences, goals, constraints, relationships, and past experiences are accumulated gradually and often change over time. Graph-based personalized memory provides a structured way to model such user information through explicit relations, temporal context, and evidence links. Such representations can model not only what an agent remembers about a user but also how memories are connected, revised, and retrieved to support personalized decisions. However, existing work remains fragmented across personalized agents and generic graph memory frameworks, making it difficult to understand the design space as a whole. This survey develops a lifecycle-oriented view of graph-based personalized memory for LLM agents. We organize existing studies around memory representation, memory evolution, memory retrieval, and memory evaluation. We further compare key design choices, discuss current evaluation practices, and open challenges in building reliable long-term personalized agents. This survey aims to clarify how graph-based memory can support adaptive, controllable, and user-centric LLM agents.
Chinese Translation
大语言模型（LLM）智能体正在从单会话工具演变为长期个性化助手，必须在不同任务、情境和交互中适应个体用户。这一转变使记忆成为个性化的核心需求，因为用户偏好、目标、约束、关系以及过往经验是逐渐积累的，并且往往随时间而变化。基于图的个性化记忆通过显式关系、时间上下文和证据链接，为建模此类用户信息提供了一种结构化方式。这类表示不仅能建模智能体关于用户记住了什么，还能建模记忆之间如何连接、修订和检索，以支持个性化决策。然而，现有工作仍零散地分布在个性化智能体与通用图记忆框架之间，使人难以整体理解其设计空间。本综述针对 LLM 智能体的基于图个性化记忆提出了一种生命周期导向的视角。我们围绕记忆表示、记忆演化、记忆检索和记忆评估来组织现有研究。我们进一步比较了关键设计选择，讨论了当前的评估实践，以及在构建可靠长期个性化智能体方面所面临的开放性挑战。本综述旨在阐明基于图的记忆如何支持自适应、可控且以用户为中心的 LLM 智能体。
cs.AI / 14 / 2609.08772
It's All in the Way You Say It: The Role of Information Representation in LLM-Based Glycemic-Event Prediction
全在于表达方式：信息表示在基于大语言模型的血糖事件预测中的作用
large language model
大语言模型相关
Abstract
Large Language Models (LLMs) are increasingly being investigated for physiological time-series prediction, yet their effectiveness may depend not only on the model itself, but also on how physiological information is represented and presented at inference time. This study investigates prompt-based general-purpose LLMs for postprandial hyperglycemia and hypoglycemia prediction in individuals with type 1 diabetes. Using the OhioT1DM dataset, we evaluate multiple open-weight LLMs under zero-shot and few-shot inference across prediction horizons of 30, 60, and 90 minutes. The analysis varies both the textual representation of the available physiological information and the amount of information exposed to the model, ranging from glucose observations alone to derived descriptors and additional contextual variables related to insulin, meals, carbohydrates, and physical activity. Performance is compared with conventional patient-specific supervised models and with Gluco-LLM, a language-model-based architecture explicitly adapted to glucose time-series forecasting. Results show a marked task-dependent behavior. Conventional supervised models achieve the strongest performance for hyperglycemia prediction, whereas the best observed prompt-based LLM configurations improve performance for hypoglycemia across all investigated horizons. The effectiveness of prompt-based inference is also strongly influenced by how physiological information is represented, while providing additional contextual information does not lead to a systematic improvement. Overall, these findings highlight physiological information representation as a central design factor in prompt-based LLM approaches to glycemic-event prediction.
Chinese Translation
大语言模型（LLM）正越来越多地被研究用于生理时间序列预测，但其有效性不仅可能取决于模型本身，还可能取决于生理信息在推理时如何被表示和呈现。本研究考察了基于提示的通用大语言模型在1型糖尿病患者餐后高血糖和低血糖预测中的应用。利用OhioT1DM数据集，我们在30、60和90分钟的预测时间范围上，评估了多种开放权重的大语言模型在零样本和少样本推理下的表现。分析同时改变了可用生理信息的文本表示方式以及暴露给模型的信息量，范围从仅血糖观测值，到衍生描述符，再到与胰岛素、进餐、碳水化合物和体力活动相关的额外情境变量。性能与传统的患者特定监督模型以及Gluco-LLM（一种专门适配于血糖时间序列预测的基于语言模型的架构）进行了比较。结果显示了一种显著的任务依赖性行为。传统的监督模型在高血糖预测中取得了最强性能，而所观察到的基于提示的大语言模型配置在所有研究的时间范围上均改善了低血糖预测性能。基于提示的推理的有效性也受到生理信息表示方式的强烈影响，而提供额外的情境信息并不会带来系统性的改进。总体而言，这些发现强调了生理信息表示是基于提示的大语言模型方法用于血糖事件预测时的一个核心设计因素。
cs.AI / 15 / 2609.08832
Closing the Consistency Gap: Self-Evolving Agents That Learn to Stay on Course
弥合一致性差距：学会保持航向的自我进化智能体
large language model
大语言模型相关
Abstract
Large language model (LLM)-powered agents can be accurate on average yet unreliable in production, a discrepancy that has been observed but remains largely unaddressed. When given the same task five times, a ReAct agent on the AppWorld benchmark using GPT-4.1 succeeds in all five runs only 53% of the time, even though its per-run pass rate averages 77%. We call this 24-point shortfall the consistency gap, and we argue that addressing it is a precondition for trustworthy AI agent deployment. We present a self-evolving agent framework that reduces this gap by identifying unstable, low-consistency steps in agent trajectories and converting them into episodic memory the agent can draw on in future runs. At its core is a Consistency Analyzer that pinpoints where and why a trajectory is likely to flip across executions, and a Guideline Generator that converts the diagnosis into targeted guidelines, committed to memory and injected into future agent executions on similar tasks. On AppWorld with ReAct/GPT-4.1, our framework raises the fraction of tasks that succeed in all five runs by +16 points on same-task evaluation and +13 points on similar-task generalization.
Chinese Translation
大语言模型（LLM）驱动的智能体在平均意义上是准确的，但在生产环境中却不可靠，这种不一致性已被观察到，但在很大程度上仍未得到解决。在 AppWorld 基准上，使用 GPT-4.1 的 ReAct 智能体在五次执行同一任务时，只在 53% 的情况下五次运行全部成功，尽管其单次运行通过率平均为 77%。我们将这 24 个百分点的缺口称为一致性差距，并认为解决这一差距是可信赖 AI 智能体部署的前提条件。我们提出一种自我进化的智能体框架，通过识别智能体轨迹中不稳定、低一致性的步骤，并将其转化为智能体可以在未来运行中利用的情景记忆，来减小这一差距。其核心是一个一致性分析器，用于精确定位一条轨迹在不同执行之间可能发生翻转的位置和原因；以及一个指南生成器，将诊断结果转化为有针对性的指南，这些指南被写入记忆，并在未来类似任务的智能体执行中被注入。在 AppWorld 上使用 ReAct/GPT-4.1 时，我们的框架使五次运行全部成功的任务比例在同任务评估中提高了 +16 个百分点，在相似任务泛化中提高了 +13 个百分点。
cs.AI / 16 / 2609.08965
PlannerForge: LLM Agents for Scenario-Based Testing of Motion Planners in Autonomous Driving
PlannerForge：用于自动驾驶运动规划器场景化测试的LLM智能体
large language model
大语言模型相关
Abstract
Ensuring the safety of autonomous driving is a critical challenge. Scenario-based testing is a systematic process used to validate Autonomous Driving Systems (ADSs), but it remains a fragmented modular pipeline in which scenario generation, retrieval, modification, ADS execution, and results analysis are performed by separate tools with little interaction. Large Language Model (LLM) agents have shown promise across ADS sub-systems such as perception, planning, and control. However, no prior work covers the whole scenario-based testing pipeline for ADSs with a unified LLM-agent framework. We present PlannerForge, an LLM-agent framework that extends all scenario-based testing stages (from Scenario Generation to ADS Assessment) and adds two further LLM-enhanced stages: ADS Enhancement and ADS Benchmarking. We evaluate PlannerForge with 10 off-the-shelf LLMs across all tasks (Generation, Selection, Modification, Module Routing, Planner Testing, and Enhancement) under 5 prompt conditions. Best-per-task scores range from 0.88 to 1.00, and open-source 20-35B backends match commercial APIs on most tasks. Open-source models such as Qwen3.6:35B match commercial APIs on three of the five tasks. Chaining the modules end-to-end retains 83% / 78% of seed queries (commercial / open). It outperforms Scenario Factory 2.0 (Finkeldei et al., 2025) on natural-language generation (193 vs. 144 executable of 200) and realises 92-96% of requested city, road and vehicle attributes. It outperforms BM25 (Robertson and Zaragoza, 2009) at rank 1 selection (92.0% vs. 67.5%) and From-Words-to-Collisions (Gao et al., 2025) on physically valid edits (>=94% vs. 31%). At N=400, cost-tuning lifts planner success from 50.4% to 70.2% and cuts collisions from 19.0% to 8.4%, without domain-specific fine-tuning.
Chinese Translation
确保自动驾驶的安全性是一个关键挑战。基于场景的测试是用于验证自动驾驶系统（ADSs）的系统化过程，但它仍然是一个碎片化的模块化流水线，其中场景生成、检索、修改、ADS执行和结果分析由各自独立的工具完成，彼此之间几乎不交互。大型语言模型（LLM）智能体已在感知、规划和控制等ADS子系统中展现出潜力。然而，此前尚无工作使用统一的LLM智能体框架覆盖ADS的完整基于场景的测试流水线。我们提出了PlannerForge，一个LLM智能体框架，它扩展了所有基于场景的测试阶段（从场景生成到ADS评估），并新增了两个由LLM增强的阶段：ADS增强和ADS基准测试。我们在5种提示条件下，使用10个现成的LLM对所有任务（生成、选择、修改、模块路由、规划器测试和增强）评估了PlannerForge。各任务最佳得分范围为0.88至1.00，并且在大多数任务上，开源的20-35B后端模型可与商业API相匹配。开源模型如Qwen3.6:35B在五项任务中的三项上与商业API持平。将各模块端到端串联后，保留了83%（商业）/78%（开源）的种子查询。它在自然语言生成方面优于Scenario Factory 2.0（Finkeldei等，2025）（每200个中可执行数量为193对比144），并实现了所要求城市、道路和车辆属性的92-96%。它在排名第一的选择上优于BM25（Robertson和Zaragoza，2009）（92.0%对比67.5%），并且在物理有效编辑上优于From-Words-to-Collisions（Gao等，2025）（≥94%对比31%）。在N=400时，成本调整将规划器成功率从50.4%提升至70.2%，并将碰撞率从19.0%降至8.4%，且无需领域特定的微调。
cs.AI / 17 / 2609.09115
MeClear: Cooperative Game-Theoretic Attribution and Risk-Aware Memory Clearance for Long-Horizon LLM Agents
MeClear：面向长时程LLM智能体的合作博弈论归因与风险感知记忆清除
large language model
大语言模型相关
Abstract
Long horizon Large Language Model (LLM) agents rely on external memory systems to preserve user preferences and task knowledge across extended interactions. Conventional retrieval mechanisms optimize semantic compatibility rather than downstream utility, frequently introducing outdated, misleading, or conflicting evidence into the active context. We present MeClear, a task conditioned memory clearance framework that identifies memories featuring negative downstream utility through cooperative attribution and selectively suppresses them from agent execution. MeClear combines Leave One Out screening with sampled cooperative Shapley attribution to distribute utility across interacting evidence, effectively resolving redundant conflict masking where single removal evaluations fail. Utilizing attribution rankings, MeClear executes a query scoped minimal clearance strategy over a nested filtration, verifying task recovery on the cleared context without permanently altering the persistent memory bank. Comprehensive experimental evaluations across ten long dialogue memory pools demonstrate that MeClear achieves a target recall of 85.9% and an overall task recovery rate of 82.3%, representing a 25.5 percentage point improvement over Leave One Out (LOO) baselines.
Chinese Translation
长时程大语言模型（LLM）智能体依赖外部记忆系统，在长时间交互中保存用户偏好与任务知识。传统检索机制优化的是语义兼容性而非下游效用，经常将过时、误导性或相互冲突的证据引入活动上下文。我们提出MeClear，一种任务条件化的记忆清除框架，通过合作归因识别具有负面下游效用的记忆，并选择性地将其从智能体执行中抑制。MeClear将留一法筛选与采样合作Shapley归因相结合，将效用分配给相互交互的证据，有效解决了单一移除评估无法处理的冗余冲突遮蔽问题。利用归因排序，MeClear在嵌套过滤结构上执行查询范围限定的最小清除策略，并在清除后的上下文上验证任务恢复情况，同时不永久改变持久记忆库。在十个长对话记忆池上的综合实验评估表明，MeClear达到了85.9%的目标召回率和82.3%的总体任务恢复率，相较于留一法（LOO）基线提升了25.5个百分点。
cs.AI / 18 / 2609.09153
Procedural Graphs: Self-Evolving Execution Structures for LLM Agents
程序化图：LLM智能体的自演化执行结构
large language model
大语言模型相关
Abstract
Large language models are increasingly deployed as agents that plan over long horizons and act through external tools. Most agents select actions through unconstrained generation over an accumulating history, leaving implicit the procedural knowledge of what to do, in what order, and under which conditions. As trajectories lengthen, agents can lose track of their objectives, invoke tools out of order, and repeat unproductive actions. We introduce the Procedural Graph: just as a knowledge graph organizes factual knowledge into (entity, relation, entity) triplets for what-is questions, a Procedural Graph organizes procedural knowledge into (procedure, relation, procedure) triplets for what-to-do questions. At each decision step, the framework localizes the agent's active node, and a guidance model translates the surrounding subgraph into step-level situational guidance that biases the solver's next action without dictating it. The graph is self-evolving: an LLM refiner contrasts failed trajectories with successful ones and edits the graph's topology and attributes, committing edits that preserve or improve held-out validation performance while retaining rejected ones to discourage repetition. Starting from a minimal skeleton, the loop builds graphs that match or surpass hand-designed ones. It can also repair a flawed expert prior. Across multiple datasets, task types, and LLMs, the Procedural Graph delivers consistent gains over memory-based baselines, and self-evolution further improves performance without manual engineering.
Chinese Translation
大型语言模型越来越多地被部署为智能体，它们会在长时间跨度上进行规划并通过外部工具采取行动。大多数智能体通过基于不断累积的历史的无约束生成来选择动作，这使关于做什么、按什么顺序做以及在什么条件下做的程序化知识隐含起来。随着轨迹变长，智能体可能会丢失对目标的追踪、无序调用工具，并重复无效的动作。我们引入了程序化图：正如知识图谱将事实性知识组织为(实体, 关系, 实体)三元组以解答“是什么”的问题，程序化图将程序化知识组织为(程序, 关系, 程序)三元组以解答“该做什么”的问题。在每个决策步骤中，框架定位智能体的活动节点，并且一个引导模型将周围的子图转化为步骤级别的情境引导，这种引导会偏向求解器的下一个动作，但不会强制规定它。该图是自演化的：一个LLM精炼器将失败的轨迹与成功的轨迹进行对比，并编辑图的拓扑和属性，提交那些能够保持或改进留出验证性能的编辑，同时保留被拒绝的编辑以防止重复。从一个最小骨架开始，该循环构建出的图能够匹配或超越手工设计的图。它还可以修复一个不完善的专家先验。在多个数据集、任务类型和LLM上，程序化图相对于基于记忆的基线带来了一致的提升，并且自演化在不进行人工工程的情况下进一步提高了性能。
cs.AR / 19 / 2609.08682
HDA-MoE: Hybrid Parallelism and Dynamic, Adaptive Scheduling for Mixture-of-Experts with 3D Near-Memory Processing
HDA-MoE：面向三维近内存处理的专家混合模型的混合并行与动态自适应调度
large language model
大语言模型相关
Abstract
Mixture-of-Experts (MoE) architectures have become a key technique for scaling Large Language Models (LLMs), enabling high model capacity with reduced computational cost. However, this efficiency comes at the expense of increased memory capacity and bandwidth demands. Recent 3D Near-Memory Processing (NMP) architectures, which vertically integrate memory and compute through hybrid bonding, provide high internal bandwidth and energy efficiency, making them attractive for accelerating MoE inference. Nevertheless, the distributed memory and compute organization of NMP systems introduces new challenges for mapping MoE workloads. Existing parallelization strategies, such as Tensor Parallelism (TP) and Expert Parallelism (EP), suffer from either high communication costs or unbalanced computation utilization, leading to inferior efficiency. In addition, the dynamic routing behavior of MoE models further complicates efficient deployment. To address these challenges, we present HDA-MoE, a framework that optimizes MoE execution on NMP architectures through hybrid parallel deployment and runtime scheduling. HDA-MoE integrates an offline hybrid parallel mapping algorithm with an online dynamic and adaptive scheduling mechanism to reduce communication overhead while improving computation utilization. Experimental results show that HDA-MoE achieves a speedup of 1.1x--3.4x over TP, 1.1x--1.5x over EP, 1.1x--3.7x over the Hybrid TP-EP compute-balanced baseline, and 1.1x--1.3x over HD-MoE. Source code is available at https://github.com/PKU-SEC-Lab/HDA-MoE-TCAD26.
Chinese Translation
专家混合（MoE）架构已成为扩展大型语言模型（LLM）的关键技术，能够在降低计算成本的同时实现高模型容量。然而，这种效率是以增加内存容量和带宽需求为代价的。近年来，通过混合键合技术垂直集成内存与计算的3D近内存处理（NMP）架构，提供了高内部带宽和能效，使其在加速MoE推理方面颇具吸引力。然而，NMP系统的分布式内存与计算组织给MoE工作负载的映射带来了新的挑战。现有的并行化策略，如张量并行（TP）和专家并行（EP），要么存在较高的通信开销，要么存在计算利用率不平衡的问题，导致效率较低。此外，MoE模型的动态路由行为进一步增加了高效部署的复杂性。针对这些挑战，我们提出了HDA-MoE，一个通过混合并行部署和运行时调度在NMP架构上优化MoE执行的框架。HDA-MoE将离线混合并行映射算法与在线动态自适应调度机制相结合，以降低通信开销并提高计算利用率。实验结果表明，HDA-MoE相较于TP实现了1.1倍至3.4倍的加速，相较于EP实现了1.1倍至1.5倍的加速，相较于混合TP-EP计算均衡基线实现了1.1倍至3.7倍的加速，相较于HD-MoE实现了1.1倍至1.3倍的加速。源代码可在https://github.com/PKU-SEC-Lab/HDA-MoE-TCAD26获取。
cs.CL / 20 / 2609.07883
Deadline-Aware Adaptive Prefill Chunking for Efficient Large Language Model Serving
面向高效大语言模型服务的截止期限感知自适应预填充分块
large language model
大语言模型相关
Abstract
Continuous batching improves large language model (LLM) serving throughput, but long prompt prefills can delay decode iterations and violate inter-token latency objectives. Chunked prefill mitigates this interference, yet its chunk size is normally fixed: small chunks protect decode latency but repeatedly pay launch overhead, while large chunks improve prefill efficiency but create latency spikes. We introduce SLOWeave, an online scheduling method that selects the largest prefill chunk predicted to finish before the earliest active decode deadline. The decision requires no workload-specific chunk-size tuning and is computed by a logarithmic-time search over a monotone iteration-cost model. We prove that, whenever a decode-only iteration is feasible and the cost predictor is accurate, SLOWeave maximizes immediate prefill progress among decisions that preserve every active request's next-token deadline. We evaluate the method in a reproducible event-driven simulator and an iteration-level GPU runtime across chat, mixed-context, long-context, and bursty workloads. Under a 25ms time-per-output-token objective, SLOWeave improves goodput over the strongest fixed-chunk baseline by 39% on mixed requests and 38% on long-context requests. Under a stricter 10ms objective, the gains rise to 3.3$\times$ and 2.4$\times$, respectively. These results isolate adaptive chunk sizing as a useful serving primitive and provide an implementation-ready controller for integration with iteration-level LLM runtimes.
Chinese Translation
连续批处理提高了大语言模型（LLM）的服务吞吐量，但长提示词的预填充可能延迟解码迭代并违反令牌间延迟目标。分块预填充缓解了这种干扰，但其块大小通常是固定的：小块保护解码延迟但反复付出启动开销，而大块提高预填充效率但造成延迟尖峰。我们提出 SLOWeave，一种在线调度方法，它选择预计能在最早的活跃解码截止期限之前完成的最大预填充块。该决策无需针对工作负载进行块大小调优，并且通过对单调迭代成本模型的对数时间搜索来计算。我们证明，只要仅解码迭代可行且成本预测器准确，SLOWeave 就在保持每个活跃请求的下一个令牌截止期限的所有决策中最大化即时预填充进度。我们在一个可复现的事件驱动模拟器和一个迭代级 GPU 运行时中，在聊天、混合上下文、长上下文和突发性工作负载上评估了该方法。在 25ms 每输出令牌时间的目标下，SLOWeave 相对于最强的固定分块基线，将混合请求的有效吞吐量提高 39%，将长上下文请求的有效吞吐量提高 38%。在更严格的 10ms 目标下，收益分别提升至 3.3$\times$ 和 2.4$\times$。这些结果将自适应分块大小确立为一种有用的服务原语，并为与迭代级 LLM 运行时的集成提供了一个可立即实现的控制器。
cs.CL / 21 / 2609.08131
Jacap: Robust KV Cache Eviction via Jacobian-Based Nonlinear Information Capacity Preservation
Jacap：基于雅可比非线性信息容量保持的鲁棒KV缓存驱逐
large language model
大语言模型相关
Abstract
Key-value (KV) cache eviction is essential for scaling long-context inference in Large Language Models. However, existing policies predominantly rely on empirical heuristics, lacking a rigorous characterization of token utility under the inherently nonlinear softmax attention mechanism. In this work, we rethink KV cache eviction through the lens of local information geometry, modeling the attention process as a nonlinear Gaussian communication channel. By performing a first-order Taylor expansion of the attention mapping, we derive the Jacobian Information Capacity, a novel objective that explicitly captures query relevance, softmax sensitivity, and structural diversity. Guided by this theory, we introduce Jacap, a capacity-aware eviction method that utilizes softmax-aware importance weighting and statistical leverage scores for subset selection. Extensive experiments across diverse architectures and benchmarks demonstrate that \textsc{Jacap} delivers superior performance in most scenarios, particularly in high-compression regimes.
Chinese Translation
键值（KV）缓存驱逐对于扩展大型语言模型中的长上下文推理至关重要。然而，现有策略主要依赖经验启发式，缺乏对内在非线性softmax注意力机制下令牌效用的严格刻画。在本工作中，我们从局部信息几何的视角重新思考KV缓存驱逐，将注意力过程建模为非线性高斯通信信道。通过对注意力映射进行一阶泰勒展开，我们推导出雅可比信息容量（Jacobian Information Capacity），这是一个新颖的目标，明确捕获了查询相关性、softmax敏感性和结构多样性。在此理论的指导下，我们提出了Jacap，一种容量感知的驱逐方法，利用softmax感知的重要性加权和统计杠杆分数进行子集选择。跨多种架构和基准的大量实验表明， extsc{Jacap} 在大多数场景中均展现出优越性能，尤其是在高压缩率场景下。
cs.CL / 22 / 2609.08156
When Metrics Reward the Worst Translations: Internalizing Cultural Reasoning for Social Media Translation Evaluation
当指标奖励最差的翻译：将文化推理内化用于社交媒体翻译评估
large language model
大语言模型相关
Abstract
Automatic translation quality metrics trained on general-domain corpora systematically fail on social media content, where communicative intent is encoded in culturally loaded expressions (internet slang, homophonic ciphers, and platform-specific idioms) rather than surface token patterns. We conduct a systematic empirical analysis demonstrating that standard metrics including COMET, XCOMET, and BERTScore exhibit near-zero or negative correlation with human cultural judgments, and even display a severity inversion in which scores increase as translation quality deteriorates. We further show that this failure extends to large language model judges: Qwen3-235B achieves Cohen's kappa of only 0.162, revealing that the bottleneck is not reasoning capacity but cultural grounding: models lack the domain-specific cultural knowledge needed to identify which aspects of a translation require scrutiny. To address this, we propose CuRIL, a reinforcement learning framework that internalizes cultural reasoning: cultural annotations are prepended inside the model's reasoning, excluded from policy gradients via a token-level loss mask, and injected with a probability that decays to zero over training, progressively forcing autonomous cultural judgment. On a 1,444-sample human-annotated social media translation benchmark, Qwen3-8B trained with CuRIL achieves Cohen's kappa 0.370 and Exact Match accuracy of 45.22%, approaching Gemini-3.1-Pro with 30x fewer parameters and surpassing models up to 235B in scale. We further demonstrate that our judge produces reliable reward signals for downstream translation optimization, reducing the low-quality translation rate by over 20 percentage points under independent human evaluation.
Chinese Translation
在通用领域语料上训练的自动翻译质量指标，系统性地在社交媒体内容上失效，因为其交际意图编码在具有文化负载的表达（网络俚语、谐音密码和平台特有习语）中，而非表面词元模式。我们进行了系统的实证分析，表明包括COMET、XCOMET和BERTScore在内的标准指标，与人类文化判断表现出接近零或负相关，甚至显示出严重性反转：随着翻译质量恶化，分数反而增高。我们进一步表明，这种失败也延伸到大型语言模型评判者：Qwen3-235B仅达到0.162的Cohen's kappa，揭示瓶颈并非推理能力，而是文化基础：模型缺乏识别翻译中哪些方面需要审查所需的领域特定文化知识。为了解决这个问题，我们提出CuRIL，一个将文化推理内化的强化学习框架：文化注释被前置到模型的推理内部，通过词元级损失掩码从策略梯度中排除，并以随训练衰减至零的概率注入，逐步迫使模型进行自主文化判断。在包含1,444个样本的人工标注社交媒体翻译基准上，使用CuRIL训练的Qwen3-8B达到Cohen's kappa 0.370和精确匹配准确率45.22%，以30倍更少的参数接近Gemini-3.1-Pro，并在规模上超越高达235B的模型。我们进一步证明，我们的评判者为下游翻译优化产生可靠的奖励信号，在独立人工评估下将低质量翻译率降低了超过20个百分点。
cs.CL / 23 / 2609.08171
EviSI: An Evaluation Agent for Simultaneous Interpreting
EviSI：一种用于同声传译的评估智能体
large language model
大语言模型相关
Abstract
Simultaneous speech-to-speech translation requires understanding, translation and spoken delivery while the source stream continues. To support timely delivery and limit accumulated delay, systems adopt reformulation and summarization, which can preserve meaning while departing from written references. BLEU and COMET may not reliably distinguish such variation from semantic loss. We introduce EviSI, a large language model evaluation agent adapting the error analysis and penalty principles of Multidimensional Quality Metrics (MQM). It constructs shared source evidence, assesses semantic fidelity and oral expression, reconciles overlapping errors and scores deterministically. EviSI recovers the aggregate human system ranking for English to Chinese. Mean Kendall agreement with human system rankings within corpora reaches 0.707 for English to Chinese and 0.467 for Chinese to English, exceeding evaluated baselines. An extension across five directions shows positive concordance with COMET without human ratings. Individual output agreement with humans remains mixed.
Chinese Translation
同声语音到语音翻译要求在源语流持续进行的同时完成理解、翻译和口语输出。为了支持及时输出并限制累积延迟，系统采用了改写和摘要策略，这些策略在偏离书面参考的同时能够保留语义。BLEU和COMET可能无法可靠地将这种变化与语义损失区分开来。我们提出EviSI，一种采用多维质量指标（MQM）的错误分析和扣分原则的大语言模型评估智能体。它构建共享源证据，评估语义保真度和口语表达，协调重叠错误并确定性评分。EviSI恢复了英语到中文的总体人类系统排名。与人类系统排名在语料库内的平均Kendall一致性在英译中方向上达到0.707，中译英方向达到0.467，超过了所评估的基线。扩展到五个方向的实验显示，在没有人类评分的情况下，与COMET呈正一致性。在单条输出层面与人类的一致性仍然参差不齐。
cs.CL / 24 / 2609.08410
Compositional Multilingual and Behavioral Attribute Steering
多语言与行为属性的组合式引导
large language model
大语言模型相关
Abstract
This study examines the compositionality of steering vectors for language and behavioral control in large language models. Focusing on language, jailbreak, and conciseness, we investigate whether additive, training-free composition of attribute steering vectors can preserve the intended steering effect of each attribute, across four instruction-tuned models from two model families and two size scales. We find that single-attribute steering is reliable for all three attributes, but only within an appropriate combination of intervention layer and steering strength, with abstract behaviors (jailbreak, conciseness) favoring middle layers and language favoring earlier layers. We show that additive composition of two attribute vectors succeeds in steering both attributes simultaneously when each is injected at its own best-performing layer, and that this partially extends to three simultaneously composed attributes, addressing an inconsistency left open by prior work on training-free composition. We further analyze the geometric properties of these steering vectors, finding that they are approximately orthogonal in the residual stream, consistent with their compositional behavior.
Chinese Translation
本研究考察了大型语言模型中用于语言和行为控制的引导向量的组合性。聚焦于语言、越狱和简洁性，在来自两个模型家族和两个规模尺度的四个指令微调模型上，我们研究了属性引导向量的加性、免训练组合能否保留每种属性的预期引导效果。我们发现，单属性引导对全部三种属性均是可靠的，但只有在干预层与引导强度的适当组合下才成立；其中抽象行为（越狱、简洁性）偏好中间层，语言偏好较早层。我们表明，将两个属性向量分别注入各自表现最佳的层时，其加性组合能够同时成功引导这两种属性，并且这一结果部分地扩展到三个属性的同时组合，从而解决了之前关于免训练组合研究遗留的不一致问题。我们进一步分析了这些引导向量的几何性质，发现它们在残差流中近似正交，这与它们的组合行为一致。
cs.CL / 25 / 2609.08475
Do Reviewers Still Reward Lexical Complexity? A Frozen-Rater Study of Preference Drift in 124K ICLR Reviews
审稿人是否仍然奖励词汇复杂性？——关于124K条ICLR评审中偏好漂移的冻结评分者研究
large language model
大语言模型相关
Abstract
Large language models have collapsed the cost of producing lexically elaborate prose, and whether peer reviewers still reward it is a question about the evaluator, not about the text. When the association between a writing cue and review scores moves across years, the reviewers may have changed, the submissions may have changed, or both, and a regression of scores on text cannot say which. We separate the two with a frozen rater: 81,850 machine reviews of ICLR submissions from 2018 to 2025, all generated in one February-April 2025 window with one model family and one prompt, so that its year-to-year coefficients track submission composition alone and the human-minus-frozen trend difference identifies reviewer preference drift. On 32,638 submissions with 124,615 human reviews, the human coefficient on non-domain lexical complexity falls from +0.142 to -0.015 while the frozen rater moves from +0.080 to +0.082; the three-way difference-in-differences is -0.0100 (q=0.013), and forty random-wordlist placebos through the same specification centre on zero. Humans still reward sentence-length variability, which the frozen rater never registers, while the frozen rater still pays for lexical complexity at its earlier rate. Every claim is held to a double gate of false-discovery control and interval exclusion, and the findings that failed adversarial re-testing are reported. Reviewers discounted a cue whose production cost collapsed, as models of manipulable signals prescribe; an LLM judge calibrated to historical human preferences inherits the earlier schedule and drifts out of alignment while its agreement with humans on totals stays ordinary.
Chinese Translation
大型语言模型使生成词汇精巧行文的成本崩溃了，而同行审稿人是否仍然奖励这种行文，是一个关于评价者而非关于文本的问题。当写作线索与评审分数之间的关联逐年变化时，审稿人可能变了，投稿可能变了，或者两者都变了，而将分数对文本进行回归无法说明是哪一种情况。我们用一个冻结评分者将两者分开：对2018年至2025年ICLR投稿的81,850条机器评审，全部在2025年2月至4月这一个时间窗口内由一个模型族和同一条提示词生成，因此它的逐年系数只追踪投稿构成的变化，人类与冻结评分之间的趋势差则识别出审稿人的偏好漂移。在带有124,615条人类评审的32,638篇投稿中，人类评分对非领域词汇复杂度的系数从+0.142下降到-0.015，冻结评分者则从+0.080变化到+0.082；三重差分估计为-0.0100（q=0.013），并且采用同一设定的40个随机词表安慰剂检验以零为中心。人类仍然奖励句长变异性，而冻结评分者从未对这一线索作出响应；冻结评分者仍按早先的速率奖励词汇复杂度。每一项论断都要通过错误发现率控制和区间排除的双重门槛，未通过对抗性再检验的发现也被报告。审稿人降低了对一种生产成本已崩溃的线索的奖励权重，正如可操纵信号模型所预示的；一个按历史人类偏好校准的LLM评审者继承了早期的奖励方案并逐渐失去对齐，而它与人类在总分上的一致性依然普通。
cs.CL / 26 / 2609.08515
Same Values, Different Languages? From Multilingual Probing to Steering LLMs Toward Chinese Social Values
相同的价值观，不同的语言？从多语言探针到引导大语言模型走向中国社会价值观
large language model
大语言模型相关
Abstract
As Large Language Models (LLMs) are increasingly integrated into human society, aligning them with pluralistic social values has become a critical priority. However, whether LLMs exhibit consistent value preferences across languages remains underexplored, particularly for culturally grounded values, which are more abstract and difficult to evaluate and align than safety-centric principles. We investigate this issue through Chinese Social Values (CSV), a value system rooted in Chinese culture and comprising $12$ dimensions across national, societal, and personal levels. We construct C-Voices, the first comprehensive multilingual contrastive probe dataset for CSV, with 86,400 dilemma-based instances in six languages, each pairing a CSV-aligned action with a value-conflicting alternative. Building on the contrastive probes of C-Voices, we then propose a fine-tuning-free value vector steering method that derives value directions from hidden-state discrepancies and selectively intervenes on value-sensitive layers during inference. Experiments on six languages show that CSV-oriented preferences are model-dependent and language-sensitive, with the same dilemma eliciting divergent responses across languages. Our method achieves effective CSV steering, supports cross-lingual transfer of value vectors, and generalizes to existing FLAMES and ValuePrism.
Chinese Translation
随着大语言模型（LLMs）日益融入人类社会，使其与多元化的社会价值观保持一致已成为一项关键优先事项。然而，LLMs 在不同语言中是否表现出一致的价值偏好仍未得到充分探索，尤其是对于扎根于文化的社会价值观而言，这些价值观比以安全为中心的原则更加抽象，也更难以评估和对齐。我们通过中国社会价值观（CSV）这一根植于中国文化、涵盖国家、社会和个人三个层面的价值体系（包含 $12$ 个维度）来研究这一问题。我们构建了 C-Voices，这是首个针对 CSV 的全面多语言对比探针数据集，包含六种语言中 86,400 个基于困境的实例，每个实例都将一个符合 CSV 的行为与一个与之冲突的替代行为配对。在 C-Voices 对比探针的基础上，我们进一步提出了一种无需微调的价值向量引导方法，该方法从隐藏状态差异中推导出价值方向，并在推理过程中选择性地对价值敏感层进行干预。在六种语言上的实验表明，面向 CSV 的偏好既依赖于模型又对语言敏感，同一困境在不同语言中会引发不同的回答。我们的方法实现了有效的 CSV 引导，支持价值向量的跨语言迁移，并能推广到现有的 FLAMES 和 ValuePrism。
cs.CL / 27 / 2609.08637
Navigating the digital spectrum: Assessing political bias, stability, and downstream fairness in Large Language Models
导航数字光谱：评估大规模语言模型中的政治偏见、稳定性与下游公平性
large language model
大语言模型相关
Abstract
Large Language Models are increasingly deployed as information intermediaries, yet measuring their political behavior remains fragile because questionnaire results mix model dispositions with measurement artifacts and response-elicitation biases. We introduce a robust Political Compass Test evaluation framework that samples 300 configurations across an eight-dimensional perturbation space varying language, framing, instructions, answer format, option order, and persona wording. We evaluate eight Gemma 3 and Qwen 3 models across 14 languages and three quantization levels, obtaining design-averaged political coordinates with quantified uncertainty. Most models lean Libertarian-Left on average, but instruction phrasing, language, and answer format significantly affect recovered coordinates. Cross-lingual differences primarily reflect coordinate drift rather than distinct cultural reasoning. Reverse-engineering the test also exposes axis-weighting imbalances and the collapse of degenerate responses toward the center, so near-origin estimates for the smallest models can reflect weak signal rather than centrism. Free-text reasoning and chat-then-classify elicitation alter recovered coordinates, and larger models show clearer persona separation, with a specific failure of the Authoritarian-Left persona to move most models in the intended social direction. In downstream tasks, persona effects are modest relative to model size and target group for hate-speech detection, while base and centrist prompts give the highest agreement for topic-level sentiment. Political role prompting therefore has measurable but task- and dataset-specific downstream effects.
Chinese Translation
大规模语言模型越来越多地被部署为信息中介，然而衡量其政治行为仍然很脆弱，因为问卷结果混合了模型倾向、测量伪影和响应诱发偏差。我们引入了一个稳健的政治指南针测试评估框架，该框架在八个维度的扰动空间中采样了300种配置，这些扰动涉及语言、措辞、指令、答案格式、选项顺序和角色措辞。我们评估了8个Gemma 3和Qwen 3模型，涵盖14种语言和3个量化级别，获得了具有量化不确定性的设计平均政治坐标。大多数模型平均倾向于自由意志主义-左翼，但指令措辞、语言和答案格式显著影响恢复的坐标。跨语言差异主要反映坐标漂移，而非不同的文化推理。对该测试进行逆向工程还揭示了轴权重失衡以及退化响应向中心坍缩的现象，因此最小模型的近原点估计可能反映的是弱信号而非中间主义。自由文本推理和先聊天后分类的诱发方式会改变恢复的坐标，更大的模型显示出更清晰的角色分离，而威权主义-左翼角色在使大多数模型沿预期社会方向移动方面存在特定失败。在下游任务中，相对于模型大小和目标群体，角色效应对仇恨言论检测的影响是适度的，而基础和中间主义提示在主题级情感方面给出了最高的一致性。因此，政治角色提示具有可测量的但依赖于任务和数据集的上下游影响。
cs.CL / 28 / 2609.08646
Combating Instruction Conflict via Energy-Driven Latent Conflict Detection
通过能量驱动的潜在冲突检测对抗指令冲突
large language model
大语言模型相关
Abstract
Large Language Models (LLMs) are increasingly deployed with hierarchical instructions, yet they remain vulnerable to conflicts in which user directives override system-level constraints. Existing defense mechanisms predominantly focus on static input inspection and therefore fail to detect Response Drift, a phenomenon in which the model's final response violates system-level constraints despite seemingly compliant inputs. To bridge this gap, we introduce ELCD, a response-level latent conflict detector for post-generation, pre-delivery verification. Given the full generated output, ELCD constructs a composite hidden-state representation by concatenating the final-token embedding with the mean-pooled response embedding. It then optimizes a pairwise margin ranking objective to separate compliant and drifting responses in latent space. Extensive experiments across five mainstream LLMs ranging from 1.5B to 14B parameters demonstrate that ELCD significantly outperforms competitive baselines. Notably, it improves the PR-AUC on Llama-2-7B by approximately 30 percentage points and reduces the False Positive Rate at 95% TPR (FPR95) on Mistral-7B to 2.67%. These results suggest that ELCD provides a promising approach for latent instruction-conflict detection in open-weight or self-hosted LLM deployments.
Chinese Translation
大型语言模型（LLMs）越来越多地部署在分层指令环境中，但它们仍然容易受到用户指令覆盖系统级约束的冲突影响。现有的防御机制主要侧重于静态输入检查，因此无法检测到“响应漂移”——即尽管输入看似合规，模型最终响应却违反系统级约束的现象。为弥合这一差距，我们提出了ELCD，一种用于生成后、交付前验证的响应级潜在冲突检测器。给定完整生成输出，ELCD通过将最终令牌嵌入与均值池化的响应嵌入拼接，构建复合隐藏状态表示。然后，它优化成对边际排序目标，以在潜在空间中将合规响应与漂移响应分开。在从1.5B到14B参数的五个主流LLM上进行的大量实验表明，ELCD显著优于竞争性基线。值得注意的是，它在Llama-2-7B上将PR-AUC提高了约30个百分点，并将Mistral-7B上95% TPR时的假阳性率（FPR95）降至2.67%。这些结果表明，ELCD为开源权重或自托管LLM部署中的潜在指令冲突检测提供了一种有前景的方法。
cs.CL / 29 / 2609.08919
Experience Funnel: A State-Policy Alternating Loop for Self-Evolving Agents
经验漏斗：自进化智能体的状态-策略交替循环
large language model
大语言模型相关
Abstract
Autonomous agents powered by large language models (LLMs) continuously accumulate experience through interaction, creating an opportunity to improve future behavior through self-evolution. A fundamental challenge is how to transform abundant, task-specific interaction experience into reusable model competence without sacrificing the ability to adapt rapidly to newly observed evidence. Explicit textual states, such as skills and agent harnesses, provide fast, human-readable and editable adaptation, but incur persistent dependence on external context; parametric policies provide compact and reusable competence, but are substantially slower to update. We present \textit{Experience Funnel}, a self-evolving framework that couples fast state adaptation with slow policy consolidation in an alternating loop. Interaction trajectories are first distilled into an explicit textual state, where newly acquired experience can be rapidly incorporated and validated. The framework then selectively identifies state-enabled behavior that remains useful across state revisions and consolidates it into the policy through transition-aware distillation. The updated state--policy pair subsequently generates new rollouts, providing fresh evidence for the next round of state adaptation and policy consolidation. Experiments across diverse agent benchmarks show that \textit{Experience Funnel} consistently improves agent capability over state-only evolution and policy-internalization approaches, while progressively converting useful explicit experience into autonomous policy competence.
Chinese Translation
由大型语言模型（LLMs）驱动的自主智能体通过交互不断积累经验，从而为通过自我进化改进未来行为创造了机会。一个根本挑战是如何将丰富的、任务特定的交互经验转化为可复用的模型能力，同时不牺牲快速适应新观察到的证据的能力。显式文本状态，如技能和智能体运行框架，提供了快速、人类可读且可编辑的适应方式，但会导致对外部上下文的持续依赖；参数化策略提供了紧凑且可复用的能力，但更新起来要慢得多。我们提出了经验漏斗，一种以交替循环将快速状态适应和缓慢策略巩固耦合在一起的自进化框架。交互轨迹首先被蒸馏为显式文本状态，在该状态中可以快速纳入并验证新获得的经验；然后，该框架会选择性识别在历次状态修订中仍然有用的状态使能行为，并通过过渡感知蒸馏将其整合到策略中。更新后的状态-策略对随后生成新的交互轨迹，为下一轮状态适应和策略巩固提供新的证据。在多个不同智能体基准上的实验表明，与仅状态进化方法和策略内化方法相比，经验漏斗一致地提升智能体能力，同时逐步将有用的显式经验转化为自主策略能力。
cs.CL / 30 / 2609.08943
Evaluating and Improving Evidence-Grounded Fact-Checking in LLMs via Multi-Round Evidence Ablation
通过多轮证据消融评估和改进大语言模型中的证据基础事实核查
large language model
大语言模型相关
Abstract
Automatic fact-checking systems assess the veracity of claims given evidence from relevant documents. Large Language Models (LLMs) have demonstrated strong performance in fact-checking due to their general reasoning capabilities. However, it remains unclear whether they faithfully make use of the evidence provided to reach veracity judgments or rely on parametric knowledge. To investigate this, we introduce Fact-Ablated Evaluation (FAE), a new evaluation framework that iteratively ablates the cited evidence to assess whether LLMs revise their predictions accordingly. Our empirical results show that current off-the-shelf LLMs as fact-checking systems rely more on their parametric knowledge than on the evidence provided. To bridge this gap between prediction accuracy and evidence grounding, we propose REAL (Rigorous Evidence Ablation Learning), a training framework that promotes evidence-dependent verification through counterfactual evidence supervision for the LLM-as-verifier models. Experiments on four fact-checking datasets across different domains demonstrate that models trained with REAL obtain superior evidence-dependent capabilities compared to standard fine-tuned models. Our findings highlight that strong fact-checking performance can still coexist with weak evidence dependency, while REAL encourages veracity predictions to remain more closely tied to the availability of supporting evidence.
Chinese Translation
自动事实核查系统根据相关文档中的证据来评估声明的真实性。大语言模型（LLMs）凭借其通用的推理能力，在事实核查中表现出强大的性能。然而，目前仍不清楚它们是否忠实地利用所提供的证据来达成真实性判断，还是依赖于参数化知识。为了研究这一点，我们引入了事实消融评估（FAE），一种新的评估框架，它迭代地消融所引用的证据，以评估大语言模型是否相应地修正其预测。我们的实证结果表明，当前现成的大语言模型作为事实核查系统时，更多地依赖其参数化知识而非所提供的证据。为了弥合预测准确性与证据基础之间的这一差距，我们提出了REAL（严格证据消融学习），一种训练框架，通过为作为验证器的大语言模型提供反事实证据监督，来促进依赖于证据的验证。在四个不同领域的事实核查数据集上的实验表明，与标准微调模型相比，使用REAL训练的模型获得了更优越的依赖于证据的能力。我们的研究结果强调，强大的事实核查性能仍然可以与较弱的证据依赖性共存，而REAL则促使真实性预测与支持性证据的可用性保持更紧密的联系。
cs.CL / 31 / 2609.09004
Evaluation of Contextual Understanding in Large Language Models
大型语言模型中上下文理解的评估
large language model
大语言模型相关
Abstract
Large Language Models (LLMs) demonstrate impressive performance across diverse NLP tasks, yet their ability to exhibit genuine contextual understanding remains uncertain. Traditional evaluation metrics such as perplexity, BiLingual Evaluation Understudy (BLEU), or surface-level accuracy fail to reveal how well LLMs extract, integrate, and reason over contextual information--a gap particularly critical in question answering, where models must align responses with contextually grounded knowledge rather than memorized associations. We propose a novel knowledge graph-based evaluation framework introducing Semantic Structural Similarity for KGs (S3KG), a hybrid similarity measure integrating structural and semantic similarity into a continuous evaluation score, alongside a diagnostic framework for categorizing reasoning errors. To validate this pipeline, we evaluate S3KG against established metrics on a curated question-answer (QA) benchmark, demonstrating its effectiveness in measuring correctness, faithfulness, and interpretability in LLM-generated responses.
Chinese Translation
大型语言模型（LLMs）在各种自然语言处理任务中表现出令人印象深刻的性能，然而它们展现真正上下文理解的能力仍然不确定。传统的评估指标，如困惑度、双语评估替补（BLEU）或表面准确率，都无法揭示LLMs在提取、整合和推理上下文信息方面的表现——这一缺陷在问答任务中尤为关键，因为模型必须将回答与基于上下文的知识对齐，而非依赖记忆的关联。我们提出了一种新颖的基于知识图谱的评估框架，引入了面向知识图谱的语义结构相似度（S3KG），这是一种混合相似度度量，将结构相似度和语义相似度整合为一个连续的评估分数，并附带一个用于对推理错误进行分类的诊断框架。为了验证该流程，我们在一个精心策划的问答（QA）基准上将S3KG与既有指标进行比较评估，证明了其在衡量LLM生成回答的正确性、忠实性和可解释性方面的有效性。
cs.CL / 32 / 2609.09085
It's Not RoPE that Creates Sinks: The Role of Self-Concentration and Value-Non-Mixing in Attention
并非RoPE创造了注意力汇：自集中与值非混合在注意力中的作用
large language model
大语言模型相关
Abstract
Large Language Models (LLMs) often exhibit "Attention Sink" (AS) and the accompanying "Massive Activations" (MAs) at the initial position of a sequence. These phenomena frequently co-occur, and MAs can pose challenges for low-bit quantization. In this study, we analyze the factors underlying AS and MAs that emerge at the initial position regardless of the token occupying it. Our experiments suggest that self-concentration of attention, resulting from the causal mask, and the subsequent Value-non-mixing in attention outputs contribute to AS and MAs. These findings provide new empirical evidence on the internal dynamics of LLMs, offering insights that may inform future quantization strategies and advance our understanding of the internal mechanisms of attention layers.
Chinese Translation
大型语言模型（LLMs）在序列的起始位置常常表现出“注意力汇”（Attention Sink, AS）以及伴随的“巨大激活”（Massive Activations, MAs）。这些现象经常共同出现，且MAs可能对低位量化带来挑战。在本研究中，我们分析了无论何种词元占据初始位置时，AS和MAs出现的根本因素。我们的实验表明，由因果掩码导致的注意力自集中，以及随后注意力输出中值的非混合特性，促成了AS和MAs。这些发现为LLMs的内部动态提供了新的实证证据，所揭示的洞见可能为未来的量化策略提供参考，并增进我们对注意力层内部机制的理解。
cs.CL / 33 / 2609.09090
Measuring LLM Sycophancy under Sustained Multi-Turn Pressure
在持续多轮压力下测量大语言模型的谄媚行为
large language model
大语言模型相关
Abstract
Large language models (LLMs) may abandon correct positions when users push back, exhibiting a failure mode known as sycophancy. Existing evaluations typically use short, pre-specified conversations and may therefore miss failures that emerge under sustained, adaptive disagreement. We introduce SPINE, a benchmark in which an LLM proxy plays a persistent but mistaken user and adaptively challenges a target model for up to 25 turns. We evaluate four production systems and three Olmo3-7b variants on 100 false-presupposition and 100 unethical-query items. Our experimental results show that collapse rates increase with conversation length for every model, short-horizon protocols underestimate sycophancy and resistance under sustained pressure remains unreliable across current models. By analyzing models with accessible reasoning traces, we surprisingly found that the correct position often remains represented in a reasoning trace when the response concedes, suggesting that the model chooses to please a user and sycophancy is not due to lack of knowledge or ignorance. Ablations show that adaptive LLM proxy exposes more sycophantic collapse than pre-generated scripts. Among all tactics, emotional appeals is the most associated with inducing LLM sycophantic behavior. The code and data are released at https://anonymous.4open.science/r/SPINE
Chinese Translation
当用户反驳时，大语言模型（LLMs）可能会放弃正确的立场，表现出一种被称为“谄媚”的失效模式。现有的评估通常使用简短的、预先设定的对话，因此可能会遗漏在持续的、自适应的分歧中出现的失效。我们引入了 SPINE，这是一个基准测试，其中一个 LLM 代理扮演一个坚持错误观点的用户，并对目标模型进行长达 25 轮的自适应挑战。我们在 100 个错误预设项目和 100 个不道德查询项目上评估了四个生产系统和三个 Olmo3-7b 变体。我们的实验结果表明，对于每个模型，崩溃率都随对话长度增加而上升；短视界协议低估了谄媚行为，而且在持续压力下的抵抗力在当前所有模型中仍然不可靠。通过分析具有可访问推理轨迹的模型，我们惊讶地发现，当回答做出让步时，正确的立场往往仍然出现在推理轨迹中，这表明模型选择取悦用户，而谄媚并非由于缺乏知识或无知。消融实验表明，自适应 LLM 代理比预先生成的脚本暴露出更多的谄媚崩溃。在所有策略中，情感诉求与诱发 LLM 谄媚行为的关系最为密切。代码和数据发布在 https://anonymous.4open.science/r/SPINE。
cs.CR / 34 / 2609.08093
LLM-Based Penetration Testing in the Presence of Honeypots
存在蜜罐环境下的基于大语言模型的渗透测试
large language model
大语言模型相关
Abstract
Large language model (LLM) agents are increasingly employed for offensive cybersecurity tasks such as automated vulnerability discovery, reconnaissance, and penetration testing. This new capability also threatens one of the defender's most valuable tools: deception. Traditional honeypots rely on realism and obscurity to lure human or script-driven attackers into revealing tactics, techniques, and procedures (TTPs), but LLM-driven attackers can reason about heterogeneous artifacts and use the honeypot suspicion to guide target-selection decisions. We present a systematic study of honeypot-aware budget allocation for LLM attack agents. We formalize the attacker's problem as a budgeted decision process: an agent interacts with potential targets, consuming LLM execution budget during reconnaissance and exploitation, and must decide whether to (continue exploitation) or (skip) when honeypot suspicion arises. Our findings show that with the proposed detector-guided policy, LLM agent attackers can effectively allocate budget to compromise hosts in a host pool, highlighting the importance of dynamically allocating budget in a controlled mixed-host testbed. While defenses are beyond our present scope, we discuss implications for future adversarially resilient and adaptive honeypot design.
Chinese Translation
大语言模型（LLM）智能体越来越多地被用于进攻性网络安全任务，如自动化漏洞发现、侦察和渗透测试。这一新能力也威胁到了防御者最有价值的工具之一：欺骗。传统蜜罐依赖真实性和隐蔽性来引诱人类或脚本驱动的攻击者暴露战术、技术和程序（TTPs），但由LLM驱动的攻击者能够对异构工件进行推理，并利用蜜罐可疑性来指导目标选择决策。我们针对LLM攻击智能体提出了一项关于蜜罐感知预算分配的系统研究。我们将攻击者的问题形式化为一个预算决策过程：智能体与潜在目标交互，在侦察和利用阶段消耗LLM执行预算，并且在出现蜜罐可疑性时必须决定是（继续利用）还是（跳过）。我们的研究结果表明，在提出的检测器引导策略下，LLM智能体攻击者能够有效分配预算以攻陷主机池中的主机，这突显了在受控混合主机测试平台中动态分配预算的重要性。虽然防御机制超出我们当前的研究范围，但我们讨论了对未来具有对抗韧性且自适应的蜜罐设计的影响。
cs.CR / 35 / 2609.08213
DRIFT: Removing Diffusion Watermarks by Deflecting the Generative Trajectory
DRIFT：通过偏转生成轨迹去除扩散水印
diffusion
扩散模型相关
Abstract
Diffusion watermarking embeds verifiable signals into the generative process and commonly verifies them by recovering trajectory-dependent evidence, making the marks robust to conventional pixel-space distortions. Existing removal attacks either regenerate along deterministic trajectories, which often preserve the watermark-bearing latent structure, or optimize every image separately. We identify the reliance on a recoverable generative trajectory as a common attack surface among the schemes we study. Based on this observation, we propose DRIFT, a black-box attack that combines partial forward diffusion with stochastic reverse resampling. Forward re-noising limits source information available to a fixed-depth recovery pipeline, while stochastic reversal supplies alternative noise-driven paths whose removal benefit we isolate through matched sampler comparisons. Adaptive DRIFT searches a selected ladder for each image's first verifier-rejected rung and refines fidelity while retaining only updates rejected by the same verifier. At fixed depth, we derive information-theoretic and Wasserstein source-dependence bounds; under realized-ladder monotonicity, the first rejected rung is least distorted among rejected rungs on that ladder, and verifier-gated refinement preserves rejection. Across nine watermarks spanning three paradigms, DRIFT achieves 98-100% attack success and the best image quality among the compared attacks, without secret keys, verifier internals, or per-image gradient optimization.
Chinese Translation
扩散水印将可验证信号嵌入生成过程，通常通过恢复与轨迹相关的证据来验证这些信号，从而使水印对传统的像素空间失真具有鲁棒性。现有的移除攻击要么沿着确定性轨迹重新生成，这些轨迹通常会保留承载水印的潜在结构，要么独立优化每个图像。我们确定了在所研究的各种方案中，对可恢复生成轨迹的依赖是一个常见的攻击面。基于这一观察，我们提出了DRIFT，一种结合部分前向扩散与随机反向重采样的黑盒攻击。前向重新加噪限制了固定深度恢复流程可用的源信息，而随机反向提供了替代的噪声驱动路径，我们通过匹配的采样器比较来分离这些路径的移除效益。自适应DRIFT为每个图像在选定的阶梯上寻找第一个被验证器拒绝的梯级，并在仅保留被同一验证器拒绝的更新的同时优化保真度。在固定深度下，我们推导了信息论和Wasserstein源依赖界；在已实现的阶梯单调性下，该阶梯上第一个被拒绝的梯级是所有被拒绝梯级中失真最小的，且验证器门控的细化保持了拒绝状态。在跨越三种范式的九个水印上，DRIFT实现了98%-100%的攻击成功率，并取得了所比较攻击中的最佳图像质量，且无需秘密密钥、验证器内部信息或逐图像梯度优化。
cs.CR / 36 / 2609.08256
ACEA: An Adversarial Co-Evolution Arena for Head-to-Head Red-Team and Blue-Team LLM Testing
ACEA：一种用于红队与蓝队大语言模型正面交锋测试的对抗性协同进化竞技场
large language model
大语言模型相关
Abstract
Automated red-team attacks and blue-team defenses for large language models (LLMs) are advancing quickly. However, attackers and defenders are built and tested in isolation, and the resulting scores are hard to trust. To tackle this, we present ACEA (Adversarial Co-Evolution Arena), a platform that connects a pluggable red-team adapter and a pluggable blue-team adapter to a shared target LLM and scores their attack and defense rates with an LLM judge. ACEA contributes four components. First, a pluggable, model-agnostic arena. Any red or blue project connects over a minimal HTTP protocol, which we call the ACEA Standard Adapter Protocol (ASAP). It can be written in any language, and a project that exposes nothing but the protocol is a full participant. Second, an evaluation methodology built for adversarial rounds. Seeding the target with canonical secrets gives verifiable ground truth that separates real leakage from hallucination. We also send each attack to the target even when the defense blocks it, which measures the attack's raw potency independently of whether it was stopped. Together these yield a per-round decomposition of attack strength and defense effectiveness. Third, a real-time, game-style visualization with a detailed end-of-battle report that localizes each failure. The evaluation thus becomes an actionable signal for improving a red or blue project. Fourth, an optional in-context improvement loop that turns each round's outcome into advisory hints for the next. An adapter can then adapt across rounds without keeping state, provided it reads the hints. We describe the design of ACEA and the metrics through which red and blue teams are scored head to head.
Chinese Translation
针对大语言模型（LLM）的自动化红队攻击与蓝队防御正在快速发展。然而，攻击方与防御方通常是在相互隔离的环境中构建和测试的，因此所得分数难以令人信服。为解决这一问题，我们提出了ACEA（对抗性协同进化竞技场），这是一个平台，它将可插拔的红队适配器与可插拔的蓝队适配器连接到共享的目标LLM，并通过LLM评判器对它们的攻击率和防御率进行评分。ACEA贡献了四个组成部分。第一，一个可插拔且与模型无关的竞技场。任何红队或蓝队项目均可通过一种极简的HTTP协议连接，我们称之为ACEA标准适配器协议（ASAP）。该协议可以用任何语言实现，而一个仅暴露该协议的项目即可作为完整的参与者。第二，一种专为对抗性回合构建的评估方法。通过用标准机密信息对目标进行植入，可获得可验证的基准真值，从而将真实泄露与幻觉区分开来。即使防御方阻止了攻击，我们也会将每次攻击发送给目标，这可以独立于攻击是否被阻止来衡量攻击的原始效力。这些设计共同实现了对每个回合中攻击强度和防御效果的可分解评估。第三，一种实时的、游戏风格的可视化界面，并附带详细的战后报告，用于定位每一次失败。因此，评估成为改进红队或蓝队项目的可操作信号。第四，一个可选的上下文内改进循环，可将每个回合的结果转化为对下一回合的参考提示。只要适配器读取这些提示，它就可以在不保持状态的情况下跨回合进行自适应。我们描述了ACEA的设计以及用于对红队和蓝队进行正面交锋评分的指标。
cs.CR / 37 / 2609.08765
Benchmark Scores Are Pipeline-Dependent: A Reliability Audit of Cybersecurity LLM Benchmarks
基准分数依赖于评估流程：对网络安全大语言模型基准的可靠性审计
large language model
大语言模型相关
Abstract
Large language model (LLM) benchmarks are often treated as fixed datasets with stable scores, yet their outcomes depend on configurable evaluation pipelines. We audit eight cybersecurity benchmarks across 10 proprietary, open-weight, and cybersecurity-specialized LLMs. By modeling benchmarks as measurement pipelines, we identify 15 systematic failure modes and show that a single pipeline choice can change a model's score by more than 80 percentage points and substantially alter model rankings. At the cross-benchmark level, two semantically similar task pairs rank the same models differently because of incompatible evaluation conventions. Under an evaluation harness that standardizes pipeline choices while preserving task semantics, nine of 10 models shift by at least three ranks on at least one benchmark. These results show that cybersecurity LLM benchmark scores are pipeline-dependent and motivate pipeline-aware auditing as a core requirement for reliable model evaluation.
Chinese Translation
大型语言模型（LLM）基准常被视为具有稳定分数的固定数据集，然而它们的结果依赖于可配置的评估流程。我们对跨10个专有、开放权重和网络安全专用大语言模型的八个网络安全基准进行了审计。通过将基准建模为测量流程，我们识别出15种系统性故障模式，并表明单一流程选择可使模型得分变化超过80个百分点，并显著改变模型排名。在跨基准层面，两对语义相似的任务会因不兼容的评估约定而对相同模型产生不同排名。在一种将流程选择标准化同时保留任务语义的评估框架下，10个模型中有9个在至少一个基准上至少移动三个名次。这些结果表明，网络安全大语言模型基准分数依赖于流程，并促使将流程感知审计作为可靠模型评估的核心要求。
cs.CR / 38 / 2609.09087
PrivEscalate: Measuring and Augmenting the Threat of LLM-Automated Linux Privilege Escalation
PrivEscalate：测量与增强LLM自动化的Linux权限提升威胁
large language model
大语言模型相关
Abstract
As Large Language Model (LLM) agents increasingly automate offensive operations across the cyber kill chain, their efficacy in complex local post-exploitation tasks remains inadequately quantified. Among these, Linux privilege escalation is a key step between initial access and full system compromise. However, existing evaluations for this task are limited by small sample sizes (fewer than 15 scenarios), lacking the scale to compare model capabilities under executable verification. To address this, we present PrivEscalate, a large-scale benchmark for Linux privilege escalation, comprising 531 Dockerized scenarios spanning 14 sub-categories. We additionally derive 329 parameterized variants to measure sensitivity to environmental distractors. Evaluating six LLMs across three agent architectures reveals: (i) model capability is heterogeneous across vulnerability classes, with no single model dominating across the high-prevalence classes, motivating multi-dimensional risk assessments; (ii) LLM successes are sensitive to environmental perturbation, so configuration rotation can disrupt some exploit attempts but does not eliminate the measured risk; and (iii) agent architectures can materially change success rates and reorder model rankings, though the magnitude is model-dependent. Leveraging these insights, we develop PrivEscAgent, a domain-specialized wrapper that augments a generic ReAct agent with deterministic enumeration, category matching, and step planning. PrivEscAgent improves over prior Linux privilege-escalation agent baselines without underlying LLM modifications. We release PrivEscalate as an open-source, Dockerized measurement instrument supporting LLM agent evaluation, defensive tool validation, and red-team training.
Chinese Translation
随着大型语言模型（LLM）智能体在网络安全杀伤链中日益自动化地执行攻击性操作，它们在复杂的本地后渗透任务中的效能仍未得到充分量化。其中，Linux权限提升是从初始访问到完全控制系统之间的关键步骤。然而，现有针对该任务的评估受限于较小的样本量（少于15个场景），缺乏在可执行验证下比较模型能力的规模。为解决这一问题，我们提出了PrivEscalate，一个大规模Linux权限提升基准，包含531个Docker化场景，涵盖14个子类别。我们额外衍生出329个参数化变体，以衡量对环境干扰因素的敏感性。在三种智能体架构上评估六个LLM揭示了：（i）模型能力在不同漏洞类别间具有异质性，没有任何单一模型在高发类别中占据主导地位，这促使需要多维风险评估；（ii）LLM的成功对环境扰动敏感，因此配置轮换可以破坏某些利用尝试，但并不能消除已测量的风险；以及（iii）智能体架构能够显著改变成功率并重新排列模型排名，尽管影响幅度依赖于具体模型。基于这些洞察，我们开发了PrivEscAgent，一个领域专用包装器，通过确定性枚举、类别匹配和步骤规划来增强通用ReAct智能体。在不修改底层LLM的情况下，PrivEscAgent相对于先前的Linux权限提升智能体基线取得了改进。我们将PrivEscalate作为开源、Docker化的测量工具发布，支持LLM智能体评估、防御工具验证和红队训练。
cs.LG / 39 / 2609.08032
Flexible Motion Generation from Language and Style References
基于语言和风格参考的灵活运动生成
diffusion
扩散模型相关
Abstract
We introduce FlexMoGen, a novel framework for flexible human motion synthesis conditioned on both natural language descriptions and motion style references. Text prompts are effective at defining semantic content, but they are often limited in capturing fine-grained style details such as timing, limb articulation, and expressive dynamics. A style example clip supplements the text by conveying these nuanced motion characteristics directly, enabling the model to preserve high-level intent while reproducing the desired stylistic traits. Given a text prompt and a style example clip, FlexMoGen generates high-quality motions that preserve semantic content while faithfully reflecting the target style, offering users greater control over the animation generation process. Unlike prior methods that rely on discrete style labels and do not generalize to long or multi-style generation, FlexMoGen learns a variational style encoder without style supervision and supports long, time-varying, multi-style synthesis. Our framework jointly pre-trains the style encoder and a text-to-motion latent diffusion model within a unified architecture, modulating motion style through a lightweight adaptation module. It integrates an efficient relative positional encoding scheme and is trained on both stylized and non-stylized datasets, enabling strong generalization to unseen text-style combinations. Experiments show that FlexMoGen achieves the best balance between content fidelity and style reflection.
Chinese Translation
我们提出了FlexMoGen，一个以自然语言描述和运动风格参考为条件的灵活人体运动合成新框架。文本提示在定义语义内容方面是有效的，但它们通常在捕捉细粒度风格细节（如时序、肢体关节动作和表现性动态）方面受限。风格示例片段通过直接传达这些细微的运动特征来补充文本，使模型能够在保留高层意图的同时重现所需的风格特征。给定文本提示和风格示例片段，FlexMoGen生成高质量的运动，在忠实反映目标风格的同时保留语义内容，为用户提供对动画生成过程的更大控制。与依赖离散风格标签且不能推广到长序列或多风格生成的先前方法不同，FlexMoGen在无需风格监督的情况下学习一个变分风格编码器，并支持长时间变化的多风格合成。我们的框架在统一架构内联合预训练风格编码器和文本到动作的潜在扩散模型，通过一个轻量级适配模块来调节运动风格。它集成了高效的相对位置编码方案，并在风格化和非风格化数据集上训练，从而对未见过的文本-风格组合具有强大的泛化能力。实验表明，FlexMoGen在内容保真度和风格反映之间取得了最佳平衡。
cs.LG / 40 / 2609.08084
Marigold V2: Revisiting Diffusion Transformers for Monocular Depth Estimation
Marigold V2：重新审视用于单目深度估计的扩散 Transformer
diffusion
扩散模型相关
Abstract
Monocular depth estimation is a ubiquitous yet highly ill-posed computer vision task, with downstream applications in scene reconstruction, computational photography, and robotics, among others. Despite the field's maturity, recent models still struggle to generalize to out-of-distribution inputs and to produce sharp and detailed depth maps. In this paper, we revisit Marigold, a set of techniques for repurposing modern image generation and editing models, powered by the diffusion transformer (DiT) architecture, into state-of-the-art monocular depth estimators. Our recipes target single-step inference from pretrained multi-step flow-matching models, with quantization where needed, preserving model capacity while remaining cheap to run. We analyze the artifacts of naive training and identify two effective remedies: aligning the model's internal representations with semantic features extracted from ground-truth, and adopting a 2-stage fine-tuning protocol built around a novel Sinkhorn-based loss. The results are crisper, cleaner depth maps that generalize well out-of-distribution, with 16-26% improvement in AbsRel over the previous best on KITTI and ETH3D. Qualitatively, our model resolves fur, foliage, and hair-thin edges that have eluded prior models. Furthermore, Marigold V2 achieves state-of-the-art results when applied to other dense regression tasks, such as surface normals estimation and intrinsic image decomposition. Project website: https://hf.co/spaces/huawei-bayerlab/marigold-v2-web
Chinese Translation
单目深度估计是一项普遍存在但高度不适定的计算机视觉任务，其下游应用包括场景重建、计算摄影和机器人等。尽管该领域已经成熟，但最近的模型仍然难以泛化到分布外输入，并且难以生成清晰且细节丰富的深度图。在本文中，我们重新审视了 Marigold，这是一套将现代图像生成与编辑模型（由扩散 Transformer（DiT）架构驱动）重新用于最先进的单目深度估计器的技术。我们的方法针对从预训练的多步流匹配模型进行单步推理，并在需要时进行量化，在保持模型容量的同时保持低廉的运行成本。我们分析了朴素训练产生的伪影，并确定了两种有效的补救措施：将模型的内部表示与从真值中提取的语义特征对齐，以及采用围绕一种新颖的基于 Sinkhorn 的损失构建的两阶段微调协议。其结果是更清晰、更干净的深度图，能够良好地泛化到分布外数据，在 KITTI 和 ETH3D 上相比之前的最佳结果，AbsRel 提升了 16-26%。在定性方面，我们的模型解决了先前模型难以处理的毛皮、植被和发丝般细的边细节。此外，Marigold V2 在应用于其他密集回归任务（如表面法线估计和固有图像分解）时也取得了最先进的结果。项目网站：https://hf.co/spaces/huawei-bayerlab/marigold-v2-web
cs.CR / 41 / 2609.08331
Do Input-Level Defenses Transfer to Observation-Level Attacks on VideoLLMs?
输入级防御能否迁移到针对视频大语言模型的观察级攻击？
large language model
大语言模型相关
Abstract
Video Large Language Models (VideoLLMs) are increasingly deployed in safety-critical applications such as content moderation and video analytics. To process long videos efficiently, VideoLLMs rely on frame sampling, token compression, and modality fusion, which together form an observation pipeline that reduces the raw video to a compact internal representation. Recent observation-level attacks exploit this pipeline to prevent the model from perceiving harmful content, yet no defense has been explicitly designed for this threat. We introduce DefTEval, a controlled evaluation framework that systematically assesses whether input-level adversarial defenses, which operate on the pixel content of already-sampled frames, can mitigate observation-level attacks. Across five VideoLLMs, eleven representative defenses, and five attack types, we find that input-level defenses offer limited and inconsistent protection, with harmful detection rates frequently near zero. Critically, defenses fail even against attacks that embed harmful signals in every sampled frame, indicating that the bottleneck extends beyond sampling omission to the suppression of signals that do enter the model. Token compression discards localized features, and modality fusion systematically down-weights weakened visual signals. Furthermore, defense effectiveness is dominated by model architecture rather than by the defense method itself, and detection rates vary drastically across content categories, exposing structural weaknesses in temporal reasoning. These findings demonstrate that securing VideoLLMs requires system-level robustness mechanisms spanning sampling-aware coverage guarantees, token-level preservation of safety-relevant features, and modality-balanced fusion.
Chinese Translation
视频大语言模型（VideoLLMs）越来越多地部署在内容审核和视频分析等安全关键应用中。为了高效处理长视频，视频大语言模型依赖帧采样、令牌压缩和模态融合，这些共同构成一个观察流水线，将原始视频缩减为紧凑的内部表示。最近的观察级攻击利用这一流水线来阻止模型感知有害内容，然而尚无防御方法被明确设计用来应对这一威胁。我们引入了DefTEval，一个受控评估框架，系统评估输入级对抗防御（作用于已采样帧的像素内容）能否缓解观察级攻击。在五个视频大语言模型、十一种代表性防御方法和五种攻击类型中，我们发现输入级防御提供的保护有限且不稳定，有害内容检测率经常接近零。关键的是，即使针对在每个采样帧中都嵌入有害信号的攻击，防御也会失败，这表明瓶颈不仅限于采样遗漏，还扩展到了对确实进入模型的信号的压制。令牌压缩丢弃局部特征，模态融合则系统地降低被弱化的视觉信号的权重。此外，防御效果主要由模型架构而非防御方法本身主导，并且检测率在不同内容类别之间差异巨大，暴露出时间推理的结构性弱点。这些发现表明，保障视频大语言模型的安全需要系统级鲁棒性机制，涵盖采样感知的覆盖保证、令牌级的安全相关特征保留以及模态平衡的融合。
cs.CL / 42 / 2609.08391
From Coordinates to Candidate Regions: Temporal Change Localization via Region Selection in Remote Sensing Multimodal LLMs
从坐标到候选区域：遥感多模态大语言模型中基于区域选择的时间变化定位
large language model
大语言模型相关
Abstract
Remote sensing multimodal large language models (RS-MLLMs) have advanced scene understanding and visual question answering over satellite imagery, yet localizing specific objects or changed regions remains challenging. Existing approaches rely on generating bounding box coordinates as token sequences, which is fragile for the small, densely packed objects common in remote sensing and increasingly error-prone when multiple targets must be localized simultaneously. In this work, we present an RS-specific formulation of the region selection paradigm, previously explored in natural-image MLLMs, and extend it to temporal change localization over multi-image sequences. Our framework employs a text-conditioned region proposal module, encodes each candidate as special tokens carrying per-frame visual features enriched with spatial and temporal cues, and lets the LLM localize targets by selecting region tokens in its response. We construct a multi-task training and evaluation suite spanning localization, referring expression, visual grounding, and understanding tasks across single-image and multi-temporal settings. Experiments show that our approach substantially outperforms coordinate-generation baselines on temporal change localization, while improving single-image visual grounding and maintaining competitive understanding performance. Oracle analysis decomposes the contributions of the region proposer and the LLM selector, providing diagnostic insight unique to this framework. Our code will be available at https://github.com/juwan-kr/RS-RegionSelect.
Chinese Translation
遥感多模态大语言模型（RS-MLLMs）推进了对卫星影像的场景理解与视觉问答，但定位特定目标或变化区域仍然具有挑战性。现有方法依赖于将边界框坐标生成为标记序列，这对于遥感中常见的小尺寸、密集排列目标而言较为脆弱，并且在需要同时定位多个目标时错误率会进一步上升。在本工作中，我们提出了区域选择范式的一种遥感专用表述，该范式此前已在自然图像多模态大语言模型中被探索，并且我们将其扩展至跨多图像序列的时间变化定位。我们的框架采用一种文本条件化的区域提议模块，将每个候选目标编码为携带逐帧视觉特征的特殊标记，这些特征融合了空间与时间线索，并让大语言模型通过在其回复中选择区域标记来定位目标。我们构建了一个多任务训练与评估套件，涵盖跨单图像与多时间设置的定位、指代表达、视觉定位和理解任务。实验表明，我们的方法在时间变化定位上显著优于基于坐标生成基线的表现，同时改进了单图像视觉定位并保持了具有竞争力的理解性能。Oracle分析分解了区域提议器与大语言模型选择器的贡献，为这一独特的框架提供了诊断性洞见。我们的代码将发布在 https://github.com/juwan-kr/RS-RegionSelect。
cs.AI / 43 / 2609.08950
SQLMorph: Query Mutation and Fine-Grained Metrics for Text-to-SQL Evaluation
SQLMorph：用于Text-to-SQL评估的查询变异与细粒度指标
large language model
大语言模型相关
Abstract
Text-to-SQL systems translate natural language queries into executable SQL, democratizing access to structured data. Despite recent advances driven by large language models (LLMs), evaluation remains a major bottleneck: public benchmarks fail to capture the complexity of enterprise schema, while building private evaluation sets is costly and nondeterministic, making evaluation results difficult to reproduce. To address this issue, we present SQLMorph, a framework for Text-to-SQL evaluation via query mutation. SQLMorph introduces two techniques to automatically generate and expand evaluation sets: Join Query Expansion (JQE), which systematically increases structural complexity through valid join additions, and Textual Query Augmentation (TQA), which generates controlled natural language perturbations to assess robustness to linguistic variation. JQE and TQA create targeted choke points to challenge specific system components. When applied to state-of-the-art systems, JQE increases query coverage and reveals accuracy degradation as the number of joins grows. Meanwhile, TQA shows that linguistic brittleness induced by heavy abbreviation can reduce accuracy by up to 17%. Beyond evaluation sets, SQLMorph introduces a family of execution-level metrics that address the limitations of current binary measures, such as Execution Accuracy. We define Execution Precision (EXP) and Execution Recall (EXR) to quantify the fraction of correct and recovered results, respectively, and combine them via F1 for unified scoring. Our experiments show that these relaxed metrics enable fine-grained analysis of over- and under-prediction, revealing differences across systems that binary metrics obscure. Together, SQLMorph's query mutation and fine-grained metrics support debugging and better align Text-to-SQL evaluation practices with real-world deployments.
Chinese Translation
Text-to-SQL系统将自然语言查询转换为可执行的SQL，使对结构化数据的访问大众化。尽管大型语言模型（LLMs）推动了近期进展，评估仍然是主要瓶颈：公共基准无法捕捉企业模式的复杂性，而构建私有评估集既代价高昂又具有不确定性，使得评估结果难以复现。为解决这一问题，我们提出了SQLMorph，一个通过查询变异进行Text-to-SQL评估的框架。SQLMorph引入了两种自动生成和扩展评估集的技术：连接查询扩展（JQE），它通过有效的连接添加系统地增加结构复杂性；以及文本查询增强（TQA），它生成受控的自然语言扰动以评估对语言变异的鲁棒性。JQE和TQA创建有针对性的瓶颈点，以挑战特定系统组件。当应用于最先进的系统时，JQE增加了查询覆盖率，并揭示了随着连接数量的增加准确率下降的情况。同时，TQA表明，重度缩写引起的语言脆弱性可使准确率降低高达17%。除了评估集之外，SQLMorph引入了一系列执行级指标，以解决当前二元度量（如执行准确率）的局限性。我们定义了执行精确率（EXP）和执行召回率（EXR），分别量化正确结果和已恢复结果所占的比例，并通过F1将它们结合起来进行统一评分。我们的实验表明，这些放宽的指标能够对过度预测和欠预测进行细粒度分析，揭示出二元指标所掩盖的系统间差异。总的来说，SQLMorph的查询变异和细粒度指标支持调试，并使Text-to-SQL评估实践更好地与现实世界部署保持一致。
cs.AI / 44 / 2609.08307
A Measurement Study of LLM Inference Trade-offs Across Edge Continuum Hardware
跨边缘连续体硬件的LLM推理权衡测量研究
large language model
大语言模型相关
Abstract
Large language models (LLMs) are increasingly used as backends for intelligent web services, but serving them across the edge continuum requires balancing quality, latency, model footprint, and energy. This paper presents a controlled measurement study of self-hosted LLM inference across edge and near-edge deployment nodes: an NVIDIA Jetson AGX Orin and a near-edge server with CPU-only and GPU-enabled inference modes. We evaluate multiple open-weight LLMs and quantization variants using a fixed question-answering workload, and compare them against GPT-4o as a cloud-hosted accuracy and latency reference. Our benchmarking pipeline reports accuracy, model footprint, per-token decoding latency, prefill latency, and overall execution energy. The results show that GPU-enabled server execution provides the lowest compute-side latency, while Jetson Orin shows lower measured energy, consistent with its lower platform power under our setup. CPU-only execution is consistently dominated in latency for our workload and shows higher measured energy. We also show that parameter count and downloaded weight-file size alone do not reliably predict observed accuracy or latency. Finally, using Pareto-frontier analysis, we study how deployment decisions may change under possible streamed-token delivery overheads, highlighting that compute-side inference metrics alone can lead to suboptimal placement for latency-sensitive interactive web services.
Chinese Translation
大型语言模型（LLMs）日益被用作智能Web服务的后端，但在边缘连续体上服务这些模型需要在质量、延迟、模型占用空间和能耗之间取得平衡。本文对自托管LLM推理在边缘和近边缘部署节点上的表现进行了一项受控测量研究：一个NVIDIA Jetson AGX Orin和一个具有仅CPU和启用GPU推理模式的近边缘服务器。我们使用固定的问答工作负载评估了多个开放权重LLM及量化变体，并将它们与作为云托管准确性和延迟参考的GPT-4o进行比较。我们的基准测试流程报告了准确性、模型占用空间、每令牌解码延迟、预填充延迟和总体执行能耗。结果表明，启用GPU的服务器执行提供了最低的计算端延迟，而Jetson Orin显示出较低的实测能耗，这与其在我们的设置下较低的平台功耗一致。在我们的工作负载中，仅CPU执行在延迟方面始终处于劣势，并显示出较高的实测能耗。我们还表明，仅凭参数数量和下载的权重文件大小并不能可靠地预测观察到的准确性或延迟。最后，通过Pareto前沿分析，我们研究了在可能的流式令牌传输开销下部署决策可能如何变化，强调了仅凭计算端推理指标可能导致对延迟敏感的交互式Web服务的次优放置。
cs.LG / 45 / 2609.07853
Foundation Models for Generalizable Semantic and Goal-Oriented Communication
面向可泛化语义与目标导向通信的基础模型
diffusion
扩散模型相关
Abstract
Semantic and goal-oriented communication is increasingly studied for 6G, but generalization beyond seen data remains a key weakness under tight rate budgets. Many existing systems overfit their training data and degrade sharply at very low bit rates because they attempt to compress the entire signal. We introduce Foundation Model-Guided Semantic and Goal-Oriented Communication (FMSGOC), a framework that uses broad visual-linguistic Foundation Model priors to mitigate overfitting. It further improves rate efficiency by concentrating bits on sparse, goal-aligned anchors and relying on generative foundation-model priors to reconstruct the masked regions. By decoupling what to send from how to reconstruct, a vision-language foundation model selects and transmits a sparse set of semantic anchors, while a pretrained diffusion model, fine-tuned for masked completion, reconstructs the image at the receiver. In our experiments, FMSGOC reaches 0.039 bits per pixel (BPP), maintains high semantic fidelity (cosine similarity 0.87-0.90 on CIFAR-10), remains robust on previously unseen inputs (0.83-0.86 on ImageNet), and shows good perceptual similarity (0.1278/0.1558, CIFAR-10/ImageNet), outperforming strong end-to-end baselines at lower bit rates.
Chinese Translation
语义与目标导向通信正越来越多地面向6G进行研究，但在紧张的比特率预算下，对未见数据的泛化仍是一个关键弱点。许多现有系统过拟合其训练数据，并在极低比特率下急剧退化，因为它们试图压缩整个信号。我们提出了基础模型引导的语义与目标导向通信（FMSGOC），该框架利用广泛的视觉-语言基础模型先验来减轻过拟合。它通过将比特集中于稀疏且与目标对齐的锚点上，并依赖生成式基础模型先验来重建被掩码区域，从而进一步提高速率效率。通过将发送内容与重建方式解耦，一个视觉-语言基础模型选择并传输一组稀疏的语义锚点，而一个为掩码补全微调的预训练扩散模型在接收端重建图像。在我们的实验中，FMSGOC达到了0.039比特每像素（BPP），保持了高语义保真度（在CIFAR-10上余弦相似度为0.87-0.90），在先前未见过的输入上保持鲁棒性（在ImageNet上为0.83-0.86），并表现出良好的感知相似性（0.1278/0.1558，CIFAR-10/ImageNet），在较低比特率下优于强大的端到端基线。
cs.LG / 46 / 2609.07874
InfluenceField: A Differentiable Field with Interventionally Identifiable Causal Structure for Multimodal World Modeling
影响场：一种具有干预可识别因果结构的可微场，用于多模态世界建模
large language model
大语言模型相关
Abstract
Multimodal large language models often capture visual-linguistic correlations but struggle to predict how local visual interventions propagate and affect downstream answers. We introduce InfluenceField, an intervention-aware latent field inserted between the visual encoder and language decoder. It lifts patch features into a continuous spatial representation, propagates directed influence over multiple steps, and predicts local intervention effects through a shared transition operator. Training jointly optimizes language modeling, cross-environment invariance, counterfactual rollout supervision, and structural regularization. For a nonlinear finite-basis population model, we show that target-aligned interventional supervision, together with a one-step separation condition on the transition, restricts admissible representations to within-location reparameterizations, so that the directed dependency graph of the full transition is recovered exactly. A linear specialization gives an exact partial-coverage characterization and a finite-loss stability bound, and the field analysis derives the spatial profile of coefficient interventions together with a shared-channel calibration result. On CausalVQA, InfluenceField improves overall accuracy over its backbone by 13.1 percentage points, with the largest gains on the planning and hypothetical categories. Capacity-matched baselines and structural controls attribute the gains in robustness and factual-counterfactual consistency to the causal objectives rather than to added capacity.
Chinese Translation
多模态大语言模型通常能够捕捉视觉与语言之间的相关性，但难以预测局部视觉干预如何传播并影响下游答案。我们引入了InfluenceField，一种插入在视觉编码器与语言解码器之间的干预感知潜场。它将补丁特征提升为连续空间表示，通过共享的转移算子进行多步有向影响传播，并预测局部干预效应。训练过程联合优化了语言建模、跨环境不变性、反事实展开监督以及结构正则化。对于非线性有限基群体模型，我们证明，目标对齐的干预监督加上转移上的一步分离条件，会将可容许表示限制为位置内的重新参数化，从而完整恢复整个转移的有向依赖图。线性特化给出了精确的部分覆盖刻画和有限损失稳定性界限，并且场分析导出了系数干预的空间轮廓以及共享通道校准结果。在CausalVQA上，InfluenceField相对于其主干模型将总体准确率提高了13.1个百分点，其中规划和假设类别的提升最大。容量匹配的基线与结构对照表明，鲁棒性以及事实—反事实一致性的提升归因于因果目标，而非额外容量。
cs.LG / 47 / 2609.07952
Structured Extrema Errors in Classical Surrogates for Viscous Burgers: A Physics-Consistent Interpretation
粘性Burgers方程经典替代模型中的结构极值误差：一种物理一致的解读
diffusion
扩散模型相关
Abstract
We study the local errors of classical machine-learning surrogate models, which approximate the time evolution of the one-dimensional viscous Burgers equation. Four models are compared on the same prediction task, using the spatial grid values directly: radial basis function (RBF) kernel ridge regression (KRR), linear Ridge, ExtraTrees, and Random Forests. Across all four models, the one-step residual, defined here as the true value minus the predicted value at each grid point, forms clear curved branches near predicted maxima and minima. A more detailed analysis of KRR shows that these errors are much more strongly related to the second spatial derivative, which measures local curvature, than to the first spatial derivative. Near a smooth extremum, predicted value and curvature form a local two-branch fold. Under our local curvature-based model of the residual, this fold predicts a leading-order near-parabolic relation between predicted value and residual. This geometric result motivates a direct test of the Burgers advection (transport) and diffusion (smoothing) terms. For KRR and Ridge, regression tests on held-out trajectories, a control that breaks the spatial alignment of the diffusion term, and a spectral test of high-frequency content are consistent with insufficient viscous smoothing at moderate and high viscosity. In this case, the surrogate retains more small-scale structure than the true future state. The same physical explanation is much weaker for the tree models. Finally, a correction that uses only predicted quantities reduces both one-step error and error during recursive rollout, where each prediction is used as the next input.
Chinese Translation
我们研究了经典机器学习替代模型的局部误差，这些模型用于近似一维粘性Burgers方程的时间演化。在相同的预测任务中，直接使用空间网格值比较了四种模型：径向基函数（RBF）核岭回归（KRR）、线性Ridge、ExtraTrees和随机森林。在所有四种模型中，单步残差——此处定义为每个网格点上真实值减去预测值——在预测最大值和最小值附近形成了清晰的曲线分支。对KRR的更详细分析表明，这些误差与二阶空间导数（度量局部曲率）的关系远比与一阶空间导数的关系更强。在光滑极值附近，预测值与曲率形成一个局部两分支折叠。在我们基于局部曲率的残差模型下，这种折叠预测了预测值与残差之间的一种前导阶近抛物线关系。这一几何结果直接激励了对Burgers方程平流（输运）项和扩散（平滑）项的检验。对于KRR和Ridge，在保留轨迹上的回归测试、一项破坏扩散项空间对齐的对照实验，以及高频内容的谱测试，均与在中高粘度下粘性平滑不足相一致。在这种情况下，替代模型比真实未来状态保留了更多小尺度结构。同样的物理解释对于树模型而言要弱得多。最后，仅使用预测量进行的修正同时减少了单步误差和递归滚动过程中的误差——在递归滚动中，每个预测都被用作下一个输入。
cs.LG / 48 / 2609.07966
MetaKV: Adaptive KV Cache Compression for Constrained LLM Inference
MetaKV：面向受限 LLM 推理的自适应 KV 缓存压缩
large language model
大语言模型相关
Abstract
Key--value (KV) cache compression is an effective way to reduce the memory overhead of large language model (LLM) inference, particularly for long-context workloads. However, existing compression methods make different trade-offs among accuracy, inference latency, and peak KV cache memory utilization, making a single fixed configuration unsuitable across different prompts and resource constraints. We introduce MetaKV, an adaptive framework that selects a KV cache compression configuration for each input prompt based on user-specified latency and peak memory budgets. MetaKV uses lightweight prediction models to estimate the end-to-end latency, peak memory, and probability of a correct response for each candidate configuration, and selects the configuration that best satisfies the latency-memory constraints while preserving accuracy. We evaluate MetaKV across ten configurations from three representative KV cache compression methods, KVQuant, H$_2$O, and RocketKV, together with an uncompressed FP16 configuration, on four datasets covering mathematics, science, commonsense reasoning, and reading comprehension. Across a wide range of latency and peak memory constraints, MetaKV consistently outperforms the best static configuration, improving constrained success rate (CSR), the fraction of prompts answered correctly while satisfying both constraints, by approximately 0.07 on average and up to 0.135. These results demonstrate the benefit of adapting KV cache compression to individual prompts and latency-memory constraints. Code is available at https://github.com/MichaelWang0505/MetaKV.git
Chinese Translation
键值（KV）缓存压缩是减少大语言模型（LLM）推理内存开销的一种有效方法，特别是对于长上下文工作负载。然而，现有压缩方法在准确性、推理延迟和 KV 缓存峰值内存利用率之间做出了不同的权衡，使得单一的固定配置在不同提示和资源约束下并不合适。我们提出了 MetaKV，这是一个自适应框架，根据用户指定的延迟和峰值内存预算为每个输入提示选择 KV 缓存压缩配置。MetaKV 使用轻量级预测模型估计每个候选配置的端到端延迟、峰值内存和正确响应的概率，并选择在保持准确性的同时最满足延迟-内存约束的配置。我们使用来自三种代表性 KV 缓存压缩方法（KVQuant、H$_2$O 和 RocketKV）的十个配置以及一个未压缩的 FP16 配置，在覆盖数学、科学、常识推理和阅读理解的四个数据集上评估了 MetaKV。在广泛的延迟和峰值内存约束范围内，MetaKV 始终优于最佳静态配置，将受限成功率（CSR）——即在同时满足这两种约束的情况下正确回答的提示所占比例——平均提高了约 0.07，最高提高了 0.135。这些结果证明了根据单个提示和延迟-内存约束调整 KV 缓存压缩的益处。代码可在 https://github.com/MichaelWang0505/MetaKV.git 获取。
cs.LG / 49 / 2609.07986
Automated Chest CT Protocol Selection via Large Language Model Derived Text Embeddings from Imaging Request Text
基于大语言模型从影像申请文本中提取的文本嵌入实现胸部CT检查方案自动选择
large language model
大语言模型相关
Abstract
Purpose: Accurate CT protocol selection is critical for diagnostic quality and patient safety, yet the current process is manual, time-consuming, and prone to inconsistencies. Prior Machine Learning methods using keywords or bag-of-words lack contextual understanding and perform poorly on rare protocols. We propose a decision support system using large language model (LLM) features to recommend protocols from free-text clinical indications, capturing clinical nuance and phrasing variation for more consistent, efficient selection. Methods: In this REB-approved retrospective study, 285,123 chest CT imaging requests from a large academic medical center (2017-2024) were split into training (228,099, 80%) and held-out test (57,024, 20%) sets. Each request included procedure names, clinical indication, HIS comments, and the selected protocol. Clinical text was embedded using a fine-tuned LLM, Meta's LLaMA-3.1-70B; these features input a logistic regression classifier predicting 18 protocol labels (e.g., PE, LDCT). Results: The pipeline achieved a weighted precision of 0.84, weighted F1-score of 0.81, and overall accuracy of 79% across 18 CT protocols. On 300 independent cases with expert consensus, the LLM reached an overall accuracy of 80% versus 83% for radiologists, with no significant difference (p = 0.263). Performance was comparable across most classes, with the LLM exceeding radiologists for some challenging categories, and entropy analyses indicated more balanced protocol use, suggesting reduced variability. Conclusion: An LLM-based recommendation system can leverage general knowledge from a large natural-text corpus to accurately assign chest CT protocols from free-text imaging requests, and may serve as a viable foundation for protocol recommendation tools where inputs require language understanding.
Chinese Translation
目的：准确的CT检查方案选择对于诊断质量和患者安全至关重要，然而当前流程为人工操作，耗时且容易出现不一致。先前使用关键词或词袋的机器学习方法缺乏上下文理解能力，在罕见检查方案上表现不佳。我们提出一种决策支持系统，利用大语言模型（LLM）特征从自由文本临床指征中推荐检查方案，捕捉临床细微差别和措辞变化，以实现更一致、更高效的选择。方法：在这项经REB批准的回顾性研究中，来自一家大型学术医疗中心（2017-2024年）的285,123份胸部CT影像申请被划分为训练集（228,099份，80%）和留存测试集（57,024份，20%）。每份申请包含检查名称、临床指征、HIS备注和所选检查方案。临床文本使用微调后的LLM——Meta的LLaMA-3.1-70B进行嵌入；这些特征输入逻辑回归分类器，预测18种检查方案标签（例如PE、LDCT）。结果：该流程在18种CT检查方案上实现了加权精确率0.84、加权F1分数0.81和总体准确率79%。在300例具有专家共识的独立病例中，LLM的总体准确率为80%，放射科医生为83%，差异无统计学意义（p = 0.263）。大多数类别的性能相当，LLM在某些具有挑战性的类别上超过放射科医生，熵分析表明检查方案使用更加均衡，提示变异性降低。结论：基于LLM的推荐系统能够利用大型自然文本语料库中的通用知识，从自由文本影像申请中准确分配胸部CT检查方案，并可能作为需要语言理解输入的检查方案推荐工具的可行基础。
cs.LG / 50 / 2609.07990
HyCO: A Hybrid Neural Solver for Combinatorial Optimization
HyCO：一种用于组合优化的混合神经求解器
diffusion
扩散模型相关
Abstract
Sequential reinforcement learning (RL) solvers and global diffusion model (DM) solvers for neural combinatorial optimization exhibit complementary failure modes under an optimization-regret view. The former enjoys small marginal regret in the early construction stage, but suffers from horizon-wise compounding errors with super-linear regret growth; the latter avoids horizon compounding but incurs linear or sublinear regret w.r.t. the dimension of the remaining unsolved subspace. We propose Hybrid Neural Solver for Combinatorial Optimization (HyCO), a hybrid inference algorithm that constructs a solution prefix with an RL solver and adaptively switches to a conditional DM to complete the remaining decisions. To characterize why such hybridization helps, when to trigger the handover, and how to realize it in practice, we first develop a unified error-scaling theoretical framework and prove that, under explicit error-scaling assumptions, i) the hybrid structure achieves strictly lower expected regret than either backbone alone, and ii) there exists a unique optimal trigger step that minimizes the hybrid regret. We then design a lightweight adaptive trigger that combines policy entropy and RL-DM disagreement to detect trajectory-level signals of the regime shift as a practical proxy, since the optimal trigger step is defined at the expected-regret level and is not directly computable on individual trajectories. Experimental results on diverse benchmarks demonstrate that HyCO achieves consistent improvements over both backbones and support the empirical effectiveness of adaptive triggering.
Chinese Translation
用于神经组合优化的序列强化学习（RL）求解器和全局扩散模型（DM）求解器，在优化遗憾的视角下呈现出互补的失败模式。前者在早期构造阶段具有较小的边际遗憾，但由于跨时间范围（horizon-wise）的复合误差而导致遗憾超线性增长；后者避免了时间范围的复合影响，但会相对于剩余未解子空间的维数产生线性或亚线性遗憾。我们提出组合优化的混合神经求解器（HyCO），这是一种混合推理算法，它使用RL求解器构造一个解前缀，并自适应地切换到条件DM来完成剩余决策。为了阐明此类混合为何有帮助、何时触发切换以及如何在实践中实现，我们首先建立了一个统一的误差缩放理论框架，并证明：在明确的误差缩放假设下，i) 混合结构实现的期望遗憾严格低于任一单独的主干模型，ii) 存在唯一的最优触发步，使混合遗憾最小化。随后，我们设计了一种轻量级自适应触发机制，该机制结合策略熵与RL-DM分歧，检测轨迹级上的机制切换信号，作为实用代理——因为最优触发步是在期望遗憾层面定义的，无法在单个轨迹上直接计算。在多个不同基准上的实验结果表明，HyCO在两个主干模型上都取得了一致改进，并支持自适应触发的经验有效性。
cs.LG / 51 / 2609.08064
Risk-Conditioned Fine-Tuning of Large Language Models
大语言模型的风险条件微调
large language model
大语言模型相关
Abstract
Large Language Models (LLMs) are increasingly deployed in settings where rare but severe harmful generations can have significant consequences. Existing Risk-Averse RLHF addresses this issue by optimizing Conditional Value-at-Risk (CVaR), but it trains policies for fixed risk levels and therefore cannot adjust the desired degree of risk aversion at inference time. In this paper, we propose risk-conditioned RLHF, a framework that trains a single policy that provides a continuous risk-control interface, enabling users to select different degrees of risk aversion without retraining or deploying multiple risk-specific models. Experiments across multiple benchmarks demonstrate that a single risk-conditioned policy can adapt to different risk levels at inference time, enabling more flexible and risk-aware LLM deployment.
Chinese Translation
大语言模型（LLMs）越来越多地被部署在这样一种场景中：罕见但严重的有害生成可能会产生重大后果。现有的风险规避RLHF通过优化条件风险价值（CVaR）来解决这个问题，但它针对固定的风险水平训练策略，因此在推理时无法调整所需的风险规避程度。在本文中，我们提出了风险条件RLHF，这是一种训练单个策略的框架，该策略提供连续的风险控制接口，使用户无需重新训练或部署多个特定风险模型即可选择不同程度的风险规避。跨多个基准的实验表明，单个风险条件策略可以在推理时适应不同的风险水平，从而实现更灵活且具有风险感知能力的大语言模型部署。
cs.LG / 52 / 2609.08153
Geodesic-informed Generative Diffusion Model For Topology-preserved Image Video Generation
测地线信息生成扩散模型用于保持拓扑的图像视频生成
diffusion
扩散模型相关
Abstract
Generative diffusion models have emerged as a class of powerful techniques for various imaging applications, including but not limited to synthesis, reconstruction, and segmentation. Despite their success, current generative models pose two key limitations. First, they primarily rely on image intensity and texture information, with limited attention to underlying object geometry. As a result, they do not guarantee geometric or topological consistency during the generation process, which is a crucial requirement for high-stakes domains such as computational anatomy, biology, and robotics, where preserving object structure is critical. Second, existing models fail to explicitly learn or represent shape changes in the generative process. Such deformation dynamics remain occluded within network parameters; hence leaving the transformation process uninterpretable and physically uninformed. To address these challenges, we introduce IGG (Image Generation informed by Geodesic dynamics), a novel framework that integrates topology-preserving geodesic principles into the diffusion-based generative process. In contrast to conventional methods that operate in image intensity space, IGG learns and synthesizes diverse samples within geodesic deformation spaces, where geometric object changes are learned as smooth and invertible smooth mappings from a given template/source image. Our code is publicly available at https://github.com/nellie689/IGG.
Chinese Translation
生成扩散模型已成为各种成像应用（包括但不限于合成、重建和分割）的一类强大技术。尽管取得了成功，当前的生成模型仍存在两个关键局限。首先，它们主要依赖图像强度和纹理信息，对底层物体几何结构的关注有限。因此，它们在生成过程中不保证几何或拓扑一致性，而这对于计算解剖学、生物学和机器人等高风险领域是一个关键要求，在这些领域中保持物体结构至关重要。其次，现有模型未能显式学习或表示生成过程中的形状变化。这种形变动力学仍然隐藏在网络参数中，因此使得变换过程不可解释且缺乏物理依据。为了解决这些挑战，我们引入了IGG（由测地线动力学引导的图像生成），这是一个新颖框架，将保持拓扑的测地线原理整合到基于扩散的生成过程中。与在图像强度空间中操作的传统方法不同，IGG在测地线形变空间中学习并合成多样的样本，其中几何物体变化被学习为从给定模板/源图像出发的光滑且可逆的平滑映射。我们的代码可在 https://github.com/nellie689/IGG 公开获取。
cs.LG / 53 / 2609.08253
Revisiting Spectral Representations in Generative Diffusion Models
重访生成扩散模型中的谱表示
diffusion
扩散模型相关
Abstract
Diffusion models have shown remarkable performance on diverse generation tasks. Recent work finds that imposing representation alignment on the hidden states of diffusion networks can both facilitate training convergence and enhance sampling quality, yet the mechanism driving this synergy remains insufficiently understood. In this paper, we investigate the connection between self-supervised spectral representation learning and diffusion generative models through a shared perspective on perturbation kernels. On the diffusion side, samples (e.g., images, videos) are produced by reversing a stochastic noise-injection process specified by Gaussian kernels; on the spectral representation side, spectral embeddings emerge from contrasting positive and negative relations induced by random perturbation kernels. Motivated by this, we propose a self-supervised spectral representation alignment method to facilitate diffusion model training. In addition, we clarify how joint spectral learning can benefit diffusion training from a geometric perspective. Furthermore, we find that the optimization of the spectral alignment objective is in an equivalent form of diffusion score distillation in the representation space. Building on these findings, we integrate a spectral regularizer into diffusion training objectives to improve the performance of diffusion models on multiple datasets. Experiments across images and 3D point clouds show consistent gains in generation quality. Code is released at https://github.com/yuehaowang/spectral-reg-diffusion.
Chinese Translation
扩散模型在多种生成任务上展现了卓越的性能。近期工作发现，对扩散网络隐藏状态施加表示对齐既能促进训练收敛，也能提升采样质量，然而驱动这种协同效应的机制仍未得到充分理解。在本文中，我们通过扰动核的共享视角，研究自监督谱表示学习与扩散生成模型之间的联系。在扩散方面，样本（例如图像、视频）是通过逆转由高斯核指定的随机噪声注入过程而产生的；在谱表示方面，谱嵌入源于对由随机扰动核诱导的正负关系进行对比。受此启发，我们提出一种自监督谱表示对齐方法以促进扩散模型训练。此外，我们从几何角度阐明了联合谱学习如何使扩散训练受益。进一步地，我们发现谱对齐目标的优化在表示空间中等价于扩散分数蒸馏的一种形式。基于这些发现，我们将一个谱正则化器整合到扩散训练目标中，以提升扩散模型在多个数据集上的性能。在图像和三维点云上的实验显示，生成质量具有一致的提升。代码发布于 https://github.com/yuehaowang/spectral-reg-diffusion。
cs.LG / 54 / 2609.08341
TV-Regulated OPD: Direction Matters in On-Policy Distillation
TV调控的OPD：方向在在线策略蒸馏中至关重要
large language model
大语言模型相关
Abstract
On-Policy Distillation (OPD) facilitates the transfer of knowledge from domain expert to student in the post-training phase of Large Language Models (LLMs). However, the supervision signals in mainstream OPD methods suffer from high variance and noise which is generally instable during training. In this work, we systematically investigated what really matters to the performance and the fundamental mechanisms behind the instability during training. We found that retaining only the sign of token-level advantages is sufficient to achieve the performance comparable to standard OPD. Meanwhile, smoother and bounded advantages can stabilize the training process without sacrificing its performance. These motivated us to shape the advantages using the Total Variation (TV) and propose a robust TV regulated On-Policy Distillation (TV-OPD) method. Benefiting from the bounded and diminished advantages, TV-OPD exhibits stable training dynamics and steady late-stage performance. We conducted comprehensive experiments and found that, across various settings, TV-OPD consistently achieved better performance and lower variance in the late-stage of training.
Chinese Translation
在线策略蒸馏（OPD）在大语言模型（LLMs）的后训练阶段促进了知识从领域专家向学生模型的转移。然而，主流OPD方法中的监督信号因高方差和高噪声的影响，在训练过程中通常是不稳定的。在这项工作中，我们系统地研究了什么对性能真正重要，以及训练过程中不稳定性背后的基本机制。我们发现，仅保留token级优势的符号就足以实现与标准OPD相当的性能。同时，更平滑且有界的优势值可以稳定训练过程，且不会牺牲其性能。这些发现促使我们利用总变差（TV）来调整优势值，并提出了一种鲁棒的TV调控在线策略蒸馏（TV-OPD）方法。得益于有界且减小的优势值，TV-OPD表现出稳定的训练动态和平稳的后期性能。我们进行了全面的实验，发现在各种设置下，TV-OPD在训练后期始终能获得更好的性能和更低的方差。
cs.LG / 55 / 2609.08368
Miles v0.1: Production-Level Post-Training
Miles v0.1：生产级后训练
diffusion
扩散模型相关
Abstract
We present Miles v0.1, a full-stack, production-ready system for frontier post-training. Building upon the clean design of slime, Miles designs each stage of the reinforcement-learning (RL) training loop around a single principle: components should be verified, clean, and customizable. With accuracy, efficiency, reliability, and scalability as first-class goals, Miles aims to make frontier-scale RL accessible to researchers and enterprises alike. This report walks through the system end to end: rollout engines built on SGLang, a trainer with a choice of two backends (NVIDIA Megatron-LM and PyTorch FSDP), and three weight-synchronization transports for different deployment topologies. Beyond full-parameter RL, Miles also supports LoRA RL, on-policy distillation, supervised fine-tuning, and true-on-policy rollout-training alignment, and extends the same architecture to diffusion models. We close with an end-to-end case study: fully asynchronous agentic RL on a GLM-5.2 744B-A40B model over terminal-use coding tasks, running on 64 NVIDIA GB300 GPUs with a median step time of 263 seconds over the first 30 measured steps. Miles is open-sourced at https://github.com/radixark/miles, with the project website at https://miles.radixark.com.
Chinese Translation
我们推出 Miles v0.1，这是一个面向前沿后训练的全栈式、生产就绪系统。基于 slime 的简洁设计，Miles 将强化学习（RL）训练循环的每个阶段都围绕一个单一原则进行设计：组件应当是经过验证的、简洁的且可定制的。以准确性、效率、可靠性和可扩展性为首要目标，Miles 旨在让前沿规模的 RL 对研究者和企业同样可及。本报告对该系统进行端到端介绍：基于 SGLang 构建的 rollout 引擎、可选择两种后端（NVIDIA Megatron-LM 和 PyTorch FSDP）的训练器，以及面向不同部署拓扑的三种权重同步传输方式。除全参数 RL 外，Miles 还支持 LoRA RL、在策略蒸馏、监督微调以及真正的在策略 rollout-训练对齐，并将同一架构扩展到扩散模型。最后，我们以一个端到端案例研究作结：在 GLM-5.2 744B-A40B 模型上，针对终端使用类编码任务执行完全异步的智能体 RL，运行于 64 张 NVIDIA GB300 GPU 上，前 30 个测量步骤的中位步骤时间为 263 秒。Miles 已在 https://github.com/radixark/miles 开源，项目网站为 https://miles.radixark.com。
cs.LG / 56 / 2609.08404
Environments as Scaffold: Enriching Feedback to Bootstrap Self-Evolving Agents in Long-Horizon Tasks
以环境为脚手架：丰富反馈以引导长时程任务中的自进化智能体
large language model
大语言模型相关
Abstract
Large Language Models demonstrate remarkable proficiency in static reasoning, yet training them as autonomous agents through Reinforcement Learning (RL) for long-horizon tasks is often hindered by severe reward sparsity. While conventional \textit{agent-side warming} up via supervised fine-tuning (SFT) can alleviate this, it is frequently limited by data scarcity and constrained exploration. To address this, we propose a paradigm shift to \textit{environment-side adaptation} by constructing \textbf{F}eedback-\textbf{E}nriched \textbf{E}nvironments (\textbf{FEEs}). Through a pilot study, we establish a feedback design strategy that reformulates environments by transitioning from action guidance to observation enrichment during the later stages of both intra-episode exploration and inter-episode evolution. Large-scale experiments on SciWorld and BFCL benchmarks using various Qwen3 model scales and RL algorithms such as GRPO, GSPO, and DAPO demonstrate that FEEs consistently yield performance improvements over standard settings. Furthermore, our analysis reveals that training with FEEs \textbf{(1)} stabilizes training dynamics by reducing entropy volatility, \textbf{(2)} facilitates proactive state-space exploration in difficult tasks, \textbf{(3) }ensures the internalization of environmental guidance into policy weights rather than acting as a mere inference-time prior, and \textbf{(4) }identifies intra-group feedback consistency as a critical boundary for stable optimization.
Chinese Translation
大语言模型在静态推理上表现出显著的熟练度，然而，通过强化学习（RL）将它们训练为面向长时程任务的自主智能体，常常受到严重奖励稀疏性的阻碍。虽然通过监督微调（SFT）对智能体端进行预热的常规方法可以缓解这一问题，但它通常受限于数据稀缺和探索受限。为解决这一问题，我们提出一种范式转变，转向环境端适应，通过构建反馈丰富环境（FEEs）来实现。通过一项初步研究，我们建立了一种反馈设计策略：该策略重构环境，在回合内探索和回合间演化的后期阶段，从行动引导过渡为观察丰富化。在 SciWorld 和 BFCL 基准上，使用多种 Qwen3 模型规模以及 GRPO、GSPO、DAPO 等 RL 算法开展的大规模实验表明，FEEs 相较于标准设置能够持续提升性能。此外，我们的分析揭示，使用 FEEs 训练能够：(1) 通过降低熵波动来稳定训练动态；(2) 促进困难任务中主动的状态空间探索；(3) 确保环境指导被内化到策略权重中，而不是仅仅作为推理时的先验；(4) 将组内反馈一致性识别为稳定优化的关键边界。
cs.LG / 57 / 2609.08634
Suan: Rectifying Direct Preference Safety Alignment in Large Language Models
Suan：纠正大型语言模型中的直接偏好安全对齐
large language model
大语言模型相关
Abstract
Integrating robust safety guardrails into Large Language Models (LLMs) is essential for delivering helpful yet harmless responses. While proprietary systems exhibit reliable safety controls, their underlying methodologies and trade-offs remain largely undisclosed. Achieving comparable security in open-weight models remains a persistent challenge, as post-trained variants frequently suffer from over-refusal and degraded general quality. To overcome these drawbacks, we introduce Suan, a novel preference optimization algorithm. Unlike existing methods, we formulate the optimization objective directly at the gradient level, bypassing the standard variational derivation. As a result, we obtain more interpretable and robust training dynamics. Extensive evaluations across a diverse suite of competitive baselines and benchmarks demonstrate that Suan achieves superior safety alignment while fully preserving response utility.
Chinese Translation
将稳健的安全护栏集成到大型语言模型（LLM）中，对于提供既有帮助又无害的响应至关重要。尽管专有系统表现出可靠的安全控制，但其底层方法和权衡在很大程度上仍未公开。在开放权重模型中实现同等安全性仍然是一个持续的挑战，因为后训练变体经常遭受过度拒绝和整体质量下降的问题。为克服这些缺陷，我们引入了Suan，一种新颖的偏好优化算法。与现有方法不同，我们直接在梯度层面构建优化目标，绕过了标准的变分推导。由此，我们获得了更具可解释性和稳健性的训练动态。跨多个竞争性基线和基准的广泛评估表明，Suan在实现卓越安全对齐的同时，完全保留了响应效用。
cs.LG / 58 / 2609.08709
Chimaera: A Mixture-of-Graph-Experts Architecture for Cross-Task and Cross-Dataset Graph Learning
Chimaera：一种用于跨任务与跨数据集图学习的图专家混合架构
large language model
大语言模型相关
Abstract
Designing foundation models for graphs is challenging due to the irregular structure of graphs and the different sizes and characteristics of embeddings. Chimaera integrates mixture-of-experts with graph foundation models (GFM). It integrates different GFM architectures, such as graph prompts and linear GNN models. Large language models are used to generate embeddings, and experts can be trained and combined following different strategies, GFMs, embeddings, etc. Furthermore, Chimaera extends existing linear GNNs to support link-level and graph-level tasks in addition to node-level tasks. Empirical analyses are performed on same-task and cross-task experiments with node, link, and graph classification tasks using six benchmark text-attributed graph datasets. The experiments demonstrate the effectiveness of Chimaera and its capabilities for transfer across tasks and datasets. Further insights include the need to use both large and small language models to generate embeddings for the experts, a strong cross-task transferability of simple but effective linear GNNs, and using few samples only to provide strong results.
Chinese Translation
由于图的不规则结构以及嵌入的不同大小与特征，为图设计基础模型具有挑战性。Chimaera将专家混合与图基础模型（GFM）集成在一起。它整合了不同的GFM架构，例如图提示和线性GNN模型。使用大型语言模型来生成嵌入，并且可以按照不同的策略、GFM、嵌入等对专家进行训练和组合。此外，Chimaera扩展了现有的线性GNN，使其除了节点级任务之外，还支持链接级和图级任务。使用六个基准文本属性图数据集，在节点、链接和图分类任务的同任务与跨任务实验中进行了实证分析。实验证明了Chimaera的有效性及其跨任务和跨数据集的迁移能力。进一步的见解包括：需要使用大型和小型语言模型来为专家生成嵌入；简单但有效的线性GNN具有强大的跨任务迁移性；以及仅使用少量样本即可提供强有力的结果。
cs.LG / 59 / 2609.08981
Transformers as In-Context Samplers: From Closed-Form Diffusion to Estimation-Free Sampling
作为上下文采样器的Transformer：从闭式扩散到免估计采样
diffusionlarge language model
扩散模型相关
大语言模型相关
Abstract
A growing body of work establishes that large language models are not mere statistical memorizers, but are capable of in-context learning: performing inference at test time using only examples provided in the prompt, without any parameter updates. Prior theoretical work has shown that this capability extends to supervised learning tasks such as linear regression. We prove that in-context learning extends further to \emph{data generation}: frozen transformers can simulate iterative generative samplers from in-context samples. We first show that transformers can realize closed-form and smoothed closed-form diffusion samplers. The construction identifies a concrete generative role for softmax attention: it computes responsibility weights and weighted empirical averages, while feedforward layers implement Euler updates. To empirically relate these constructions to pretrained language models, we study \emph{semantic-topic sampling}: prompts consisting of words drawn from a common semantic category, such as animals, foods, or cities. Across transformer layers, the normalized hidden states exhibit a two-stage geometry: they move toward a uniform spherical reference in intermediate layers and then return to structured, topic-dependent representations near the output. We further measure an interacting-particle energy on these hidden-state clouds and observe the same U-shape pattern. We then prove that transformers can approximate an energy-based sampler, constructing the same U-shape energy across the layers.
Chinese Translation
越来越多的研究表明，大型语言模型不仅仅是统计记忆器，还具备上下文学习能力：在测试时仅使用提示中提供的示例进行推理，而无需任何参数更新。先前的理论工作表明，这种能力可以扩展到线性回归等监督学习任务。我们证明，上下文学习可以进一步扩展到数据生成：冻结的Transformer能够根据上下文样本模拟迭代式生成采样器。我们首先证明，Transformer能够实现闭式扩散采样器和平滑闭式扩散采样器。该构造揭示了softmax注意力在生成中的具体作用：它计算责任权重和加权经验平均值，而前馈层则执行欧拉更新。为了在经验上将上述构造与预训练语言模型联系起来，我们研究了语义主题采样：即由来自同一语义类别（如动物、食物或城市）的单词构成的提示。在Transformer的各层中，归一化后的隐藏状态呈现出一种两阶段几何结构：它们在中间层移向一个均匀的球面参考分布，然后在接近输出层时返回到与主题相关的结构化表征。我们进一步在这些隐藏状态云上测量一种相互作用粒子能量，并观察到相同的U形模式。然后我们证明，Transformer能够近似基于能量的采样器，并在各层中构造出相同的U形能量。
cs.LG / 60 / 2609.09009
Let It Go or Learn to Self-Correct: Continuous Diffusion for Constrained Discrete Tasks
放手或学会自我纠正：面向约束离散任务的连续扩散模型
diffusion
扩散模型相关
Abstract
Denoising Diffusion Probabilistic Models (DDPMs) generate samples by starting from noise and repeatedly denoising while keeping each update close to the current noisy state. This behavior is effective in many continuous domains, but its role is less clear for globally constrained discrete tasks, such as Sudoku, graph connectivity, Latin squares, and N-queens. In such settings, early discrete errors can be difficult to undo. As a result, standard diffusion sampling may preserve early mistakes, even when the model's clean predictions are informative. We compare standard samplers to sampling directly from the model's clean prediction. Without retraining, this single change improves Sudoku validity from 31% to 95%, with consistent gains across the other discrete tasks. We hypothesize that staying close to the current noisy state is harmful because the reverse trajectory can drift off the forward noising distribution the model was trained on. To reduce this train-test mismatch, we further introduce self-correction training, which exposes the model to its own predictions, improving robustness to errors that arise during inference. This substantially improves the performance of standard samplers. Our results suggest that continuous diffusion models can learn nontrivial global constraints, but discrete reasoning tasks require better alignment between training and inference: either through samplers that reduce commitment to early decisions, or through training that teaches the model to correct its own inference-time errors.
Chinese Translation
去噪扩散概率模型（DDPMs）通过从噪声出发并反复去噪来生成样本，同时使每次更新都接近当前带噪声的状态。这种行为在许多连续域中是有效的，但对于具有全局约束的离散任务（如数独、图连通性、拉丁方和N皇后问题）而言，其作用则不太明确。在此类设置中，早期的离散错误可能难以撤销。因此，即使模型的干净预测具有信息量，标准的扩散采样也可能保留早期错误。我们将标准采样器与直接基于模型干净预测进行采样的方式进行了比较。在不重新训练的情况下，这一单一改动将数独有效性从31%提高到95%，并且在其他离散任务上也持续带来改进。我们假设，保持接近当前带噪声状态是有害的，因为反向轨迹可能会偏离模型训练时所基于的前向加噪分布。为了减少这种训练-测试不匹配，我们进一步引入了自我纠正训练，使模型接触其自身的预测，从而提高对推理过程中出现的错误的鲁棒性。这大幅提升了标准采样器的性能。我们的结果表明，连续扩散模型可以学习非平凡的全局约束，但离散推理任务需要更好地对齐训练与推理：要么通过减少对早期决策承诺的采样器，要么通过训练教会模型纠正其自身的推理时错误。
cs.LG / 61 / 2609.09038
Do Reasoning Representations Help Humans Evaluate LLM Outputs?
推理表示能否帮助人类评估LLM输出？
large language model
大语言模型相关
Abstract
Reasoning representations are increasingly used as explanations for large language model outputs. Yet they are typically evaluated with model-centric criteria, such as answer accuracy and faithfulness, leaving it unclear whether they help people evaluate model responses. In this work, we study reasoning representations as human-facing interfaces rather than proxies for model reasoning ability. We conduct a controlled human study of six reasoning formats across tasks of varying complexity, supported by a web-based framework that randomizes task domains, problem instances, and representation order. The study collects fine-grained judgments of structural understanding, error detection and localization, and trust calibration. Our study shows a mismatch between perceived preference and support for human evaluation. Participants prefer planning- and decomposition-based representations, but simpler chain-of-thought traces better support verification, trust, and interpretability. Preferred representations also introduce calibration risks, with more false alarms on correct traces and high trust despite low willingness to verify.
Chinese Translation
推理表示越来越多地被用作大型语言模型输出的解释。然而，它们通常以模型中心的指标进行评估，例如答案准确性和忠实性，这使得它们是否有助于人们评估模型响应仍然不清楚。在这项工作中，我们将推理表示视为面向人类的界面，而不是模型推理能力的代理。我们通过一个基于网页的框架，对跨不同复杂度任务的六种推理格式进行了一项受控的人类研究，该框架随机化任务领域、问题实例和表示顺序。该研究收集了对结构理解、错误检测与定位以及信任校准的细粒度判断。我们的研究显示，感知偏好与对人类评估的支持之间存在不匹配。参与者更喜欢基于规划和分解的表示，但更简单的思维链轨迹在验证、信任和可解释性方面提供更好的支持。偏好的表示也引入了校准风险，在正确的轨迹上产生更多误报，尽管验证意愿较低，但信任度却很高。
cs.LG / 62 / 2609.09054
Training-Free Task Vectors for LLM Behavioral Control
面向大语言模型行为控制的免训练任务向量
large language model
大语言模型相关
Abstract
Task vectors enable post-training model editing by identifying semantically meaningful directions in weight space, typically computed as the difference between a fine-tuned model and its pretrained initialization. However, this reliance on fine-tuning makes discovering such directions costly and limits the practicality of post-training model editing. To address this limitation, we introduce Training-Free Task Vectors (TFTVs), a novel method to compute task-vector-like directions without requiring fine-tuning. Our method maps activation steering vectors to rank-one weight-space edits using only forward-pass statistics, while satisfying arithmetic properties that directly support learning via addition, forgetting via subtraction, and the composition of multiple edits. Empirically, we evaluate TFTVs on large language model behavioral control tasks and show that they consistently amplify, suppress, and compose target behaviors while preserving general knowledge and problem-solving skills. We also validate our method against other editing and steering baselines, experimentally demonstrating that TFTVs achieve stronger trait control with better or competitive utility preservation. We hope our work opens new directions for the community in post-training model editing and broader training-free model control. Code is available on the project website: tftv-llm.github.io.
Chinese Translation
任务向量通过识别权重空间中具有语义意义的方向来实现训练后模型编辑，这些方向通常计算为微调模型与其预训练初始化之间的差值。然而，这种对微调的依赖使得发现此类方向的成本高昂，并限制了训练后模型编辑的实用性。为了解决这一局限，我们提出了免训练任务向量（TFTVs），一种无需微调即可计算类任务向量方向的新方法。我们的方法仅使用前向传播统计量将激活引导向量映射为秩一权重空间编辑，同时满足算术性质，这些性质直接支持通过加法进行学习、通过减法进行遗忘以及多个编辑的组合。在实证方面，我们在多个大语言模型行为控制任务上评估了TFTVs，并表明它们能够一致地放大、抑制和组合目标行为，同时保留通用知识和问题解决能力。我们还与其他编辑和引导基线方法进行了对比验证，实验证明TFTVs能实现更强的特质控制，同时具有更好或同等的效用保持。我们希望我们的工作能够在训练后模型编辑以及更广泛的免训练模型控制方面为社区开辟新方向。代码可在项目网站上获取：tftv-llm.github.io。
cs.LG / 63 / 2609.09059
PlayTrain: An Efficient Reinforcement Learning Framework for LLM-Generated Adaptable JavaScript Games
PlayTrain：一个用于大语言模型生成的可适配JavaScript游戏的高效强化学习框架
large language model
大语言模型相关
Abstract
While many video-game environments (VGEs) have played crucial roles in advancing reinforcement learning (RL), developing novel VGEs or modifying existing ones to support new features, has been a laborious process requiring extensive hand-coding. Here we present PlayTrain, an RL framework that combines the abilities of large language models (LLMs) to robustly generate JavaScript (JS) games from a minimal human prompt, and an efficient pipeline that can run any JS game in a standard 'gym' environment. Not only are recent LLMs particularly good at writing JS code, but the JS format also allows users to easily play generated VGEs, while PlayTrain enables us to train RL agents on the exact same games. We demonstrate multiple use cases of PlayTrain, including cloning well-known Atari and ProcGen games in simple JS, where PlayTrain trains pixel-based agents end-to-end at over 1M agent-decisions per second on a single GPU node; and creating modified versions thereof (e.g., that support novel test sets, procedural generation logics, or game dynamics). Through PlayTrain, we reimagine RL VGE development: all we need is a single JS file, generated and modified through an LLM. We discuss promising future RL research directions that PlayTrain unlocks.
Chinese Translation
尽管许多视频游戏环境（VGEs）在推进强化学习（RL）方面发挥了关键作用，但开发新的视频游戏环境或修改现有环境以支持新功能，一直是一个需要大量手工编码的费力过程。在此，我们提出PlayTrain，一个强化学习框架，它结合了大语言模型（LLMs）从极简的人类提示中稳健生成JavaScript（JS）游戏的能力，以及一个能够将任何JS游戏运行在标准“gym”环境中的高效流水线。不仅近期的LLMs特别擅长编写JS代码，而且JS格式也允许用户轻松游玩所生成的VGEs，同时PlayTrain使我们能够在完全相同的游戏上训练RL智能体。我们展示了PlayTrain的多种用例，包括用简单JS克隆著名的Atari和ProcGen游戏，其中PlayTrain在单个GPU节点上以每秒超过100万次智能体决策的速度端到端训练基于像素的智能体；以及创建其修改版本（例如，支持新颖的测试集、程序化生成逻辑或游戏动态）。通过PlayTrain，我们重新构想了RL VGE开发：我们所需要的就是一个单独的JS文件，通过LLM生成和修改。我们讨论了PlayTrain为未来有前景的RL研究方向所开启的可能性。
cs.CL / 64 / 2609.08936
AuK Technical Report: An Open-Source Foundational Model for Speech Generation and Editing
AuK 技术报告：用于语音生成与编辑的开源基础模型
large language model
大语言模型相关
Abstract
We introduce AuK, an open-source foundational model that unifies speech generation and editing through a common interface of natural-language instructions and audio context. To support this broad capability set, we construct approximately 3.03 billion instruction--audio instances and 1.95 million hours of effective supervision across five task families: speech generation, content editing, enhancement and separation, paralinguistic editing, and acoustic editing. AuK combines a multimodal large language model for semantic conditioning, an VAE jointly trained on speech, general audio, and music for acoustic conditioning, and a hybrid rectified-flow Transformer that performs dual-stream MMDiT blocks followed by unified single-stream DiT blocks for generation. Training begins with generation-only warm-up and proceeds to joint generation--editing pre-training. We then apply complementary post-training strategies: human-feedback preference optimization for open-ended editing and reward-based reinforcement learning for speech generation. To reduce inference cost, we further distill the model with consistency initialization and task-routed Decoupled DMD. The resulting AuK-Flash performs 4-step inference without classifier-free guidance and achieves a 4.5 wall-clock speedup over the full model under matched conditions. Experiments demonstrate leading performance on zero-shot and instruction-controlled speech generation and general instruction-guided editing, while remaining competitive on signal-level restoration tasks. We release both the source code and model weights to support reproducibility and further research.
Chinese Translation
我们推出 AuK，这是一个开源基础模型，通过自然语言指令和音频上下文的通用接口，统一了语音生成与编辑。为了支撑如此广泛的能力组合，我们构建了约 30.3 亿个指令-音频实例和 195 万小时的有效监督数据，涵盖五大任务族：语音生成、内容编辑、增强与分离、副语言编辑以及声学编辑。AuK 结合了一个用于语义条件的多模态大语言模型、一个在语音、通用音频和音乐上联合训练用于声学条件的 VAE，以及一个混合修正流 Transformer，该 Transformer 依次执行双流 MMDiT 块和统一的单流 DiT 块以进行生成。训练以仅生成任务的预热阶段开始，随后进入生成与编辑的联合预训练。随后我们应用互补的后训练策略：针对开放式编辑的基于人类反馈的偏好优化，以及针对语音生成的基于奖励的强化学习。为了降低推理成本，我们进一步通过一致性初始化和任务路由的解耦 DMD 对模型进行蒸馏。由此产生的 AuK-Flash 无需分类器引导即可执行 4 步推理，并在匹配条件下相对于完整模型实现了 4.5 倍的墙钟时间加速。实验表明，模型在零样本和指令控制的语音生成以及通用指令引导的编辑任务上表现出领先性能，同时在信号级恢复任务上保持竞争力。我们同时发布了源代码和模型权重，以支持可复现性和进一步研究。
cs.SE / 65 / 2609.08327
Tool Retrievers Are Underestimated: Annotation Expansion Reveals True Capability
工具检索器被低估：标注扩展揭示真实能力
large language model
大语言模型相关
Abstract
In open-world scenarios with massive and evolving tool repositories, tool-augmented large language models rely on a retriever to surface relevant tools for a given query. Because such repositories often contain many tools that implement the same functionality, a single query can often be resolved by several distinct but functionally equivalent tool combinations, making the natural query-to-tool mapping inherently one-to-many. However, existing tool retrieval benchmarks annotate each query with a single relevant tool combination, collapsing this one-to-many mapping into a rigid one-to-one annotation and causing valid retrieved tools to be misjudged as failures. To address this, we propose ToolEX (Tool Equivalent eXpansion), a framework that automatically discovers and annotates the tool combinations functionally equivalent to the labeled ones. Applied to the 7,360-query Tool-DE benchmark, ToolEX finds that 67.9% of sub-queries admit equivalent alternatives, expanding the singular ground truth to an average of 5.3 valid combinations per query. Using the expanded benchmark ToolEQ, we re-evaluate eight base retrievers and two fine-tuned variants; metrics on ToolEQ rise substantially over Tool-DE, showing that one-to-one annotation systematically underestimates retrievers and that 30--47% of the reported fine-tuning gain is an evaluation artifact rather than genuine improvement. Applying the same pipeline to skill retrieval on SkillRet further confirms that the one-to-one problem extends beyond tool retrieval.
Chinese Translation
在具有大规模且不断演化的工具库的开放世界场景中，工具增强型大语言模型依赖检索器为给定查询找出相关工具。由于此类工具库通常包含许多实现相同功能的工具，单个查询往往可以由多种不同但功能等价的工具组合来解决，这使得自然的查询到工具映射本质上是“一对多”的。然而，现有的工具检索基准为每个查询只标注一个相关工具组合，将这种“一对多”映射压缩为僵化的“一对一”标注，导致有效的检索工具被误判为失败。为解决这一问题，我们提出了ToolEX（Tool Equivalent eXpansion），一个自动发现并标注与已标注组合功能等价的工具组合的框架。将ToolEX应用于包含7,360个查询的Tool-DE基准后，我们发现67.9%的子查询存在等价替代组合，将单一标准答案扩展为平均每个查询5.3个有效组合。使用扩展后的基准ToolEQ，我们重新评估了八个基础检索器和两个微调变体；ToolEQ上的指标相比Tool-DE显著提升，表明“一对一”标注系统性地低估了检索器，并且所报告的微调收益中有30%–47%是评估伪影，而非真正的改进。将同样的流程应用于SkillRet上的技能检索，进一步证实了“一对一”问题不仅限于工具检索。
cs.SE / 66 / 2609.08589
The Unreliable Progress Bar: Can LLM Agents Reliably Report Task Progress Throughout Execution?
不可靠的进度条：LLM 智能体能否在整个执行过程中可靠地报告任务进度？
large language model
大语言模型相关
Abstract
Recent large language models can emit task-progress signals that agent frameworks use to decide whether a task should continue or stop, yet whether a model can reliably report its task progress at every stage of a task, and where and how its reports fail, has not been studied systematically. We evaluate this ability on the public benchmark $τ^2$-bench and on StageIF, a controlled testbed in which reporting checkpoints are placed across the task's lifecycle. Both settings require reports at multiple task stages. We find that reporting reliability depends on the stage a task has reached, and that almost every deployed model we test is reliable at some stages and unreliable at others. Where reporting breaks down is not the same everywhere. Most deployed models lose accuracy once work is under way and recover once the task is done. The newest generation closes that mid-task drop and instead grows conservative at the finish line. Our study exposes a capability gap in task-progress reporting and provides an evaluation protocol that spans the whole course of task execution for this ability on which agent operation depends. The findings indicate that agent frameworks should not control task flow on the strength of the model's state reports alone.
Chinese Translation
近期的大语言模型能够发出任务进度信号，智能体框架使用这些信号来决定任务是应继续还是停止，然而，模型能否在任务的每个阶段都可靠地报告其任务进度，以及其报告在何处、如何失败，尚未得到系统性研究。我们在公开基准 $τ^2$-bench 和 StageIF 上评估这一能力，StageIF 是一个受控测试平台，其中报告检查点分布在任务生命周期的各个阶段。两种设置都要求在多个任务阶段进行报告。我们发现报告可靠性取决于任务已到达的阶段，并且我们测试的几乎所有已部署模型在某些阶段可靠，而在其他阶段不可靠。报告失效的位置并非在所有地方都相同。大多数已部署模型一旦工作开始进行，准确性就会下降；一旦任务完成，又会恢复。最新一代模型消除了任务中期的准确性下降，但反而在终点线变得保守。我们的研究揭示了任务进度报告方面的能力差距，并提供了一种贯穿任务执行全过程的评估协议，用以评测这一智能体运行所依赖的能力。研究结果表明，智能体框架不应仅凭模型的状态报告来控制任务流程。
cs.SE / 67 / 2609.08681
Beyond Fixed Fault Models: Comparing LLM-Based and Rule-Based Fault Injection in OpenStack
超越固定故障模型：比较基于LLM与基于规则的故障注入在OpenStack中的应用
large language model
大语言模型相关
Abstract
Software Fault Injection (SFI) supports testing of cloud systems by introducing software defects and observing their manifestation. Rule-based injectors such as ProFIPy provide controlled and reproducible source-level mutations but require fault patterns to be encoded manually. Large Language Models (LLMs) offer a data-driven alternative by generating context-dependent software faults. We compare two code LLMs, Qwen2.5-Coder and DeepSeek-Coder, with ProFIPy in OpenStack's Nova and Cinder services. On shared injection targets, activation and observable-failure rates are comparable, but operational profiles differ: LLM-generated faults produce more Catastrophic outcomes on Nova, whereas ProFIPy produces more Silent and Multi-component effects. The sampled LLM outputs also differ in how they manifest failure, while showing greater agreement in their propagation scope. These findings show that LLM-based fault injection extends the behavioral coverage of fixed fault models without establishing general superiority, and that practical adoption still requires controlled generation, runtime validation, system-level oracles, and reproducible experimental provenance.
Chinese Translation
软件故障注入（SFI）通过引入软件缺陷并观察其表现来支持云系统的测试。诸如ProFIPy之类的基于规则的注入器提供受控且可复现的源码级突变，但需要手动编码故障模式。大语言模型（LLMs）通过生成上下文相关的软件故障提供了一种数据驱动的替代方案。我们在OpenStack的Nova和Cinder服务中将两种代码大语言模型Qwen2.5-Coder和DeepSeek-Coder与ProFIPy进行比较。在共享的注入目标上，激活率和可观测失败率相当，但操作特征有所不同：LLM生成的故障在Nova上产生更多的灾难性结果，而ProFIPy产生更多的静默和多组件效应。采样的LLM输出在失败表现方式上也有所不同，而在传播范围上表现出更大的一致性。这些发现表明，基于LLM的故障注入扩展了固定故障模型的行为覆盖范围，但并未确立普遍的优越性，并且实际采用仍然需要受控生成、运行时验证、系统级预言以及可复现的实验来源。
cs.AI / 68 / 2609.08049
LLMs for Social Network Modeling: From Network Generation to Dynamic Processes
用于社交网络建模的大语言模型：从网络生成到动态过程
diffusionlarge language model
扩散模型相关
大语言模型相关
Abstract
Large language models (LLMs) are rapidly emerging as a new paradigm for modeling social networks by representing users and their relationships and interactions through natural language. Unlike classical network models or deep learning approaches, LLMs can simulate context-aware social behavior and language-driven interactions, enabling more realistic modeling of network formation and dynamic social processes. However, existing studies are scattered across different research communities and lack a unified perspective. This survey presents the first comprehensive review of LLMs for social network modeling by organizing the literature into two broad categories: network generative models and dynamic process models. Network generative models are further classified into selection-based and interaction-based approaches, while dynamic process models are categorized into opinion dynamics, information diffusion, and rumor propagation, each with their underlying modeling mechanisms. LLMs enable rich textual social interactions and decision-making, but they also exhibit many limitations, including inherent social biases and prompt sensitivity. We outline these open research challenges and discuss future directions in LLM-based social network modeling.
Chinese Translation
大语言模型（LLMs）正迅速成为一种通过自然语言表示用户及其关系和互动来建模社交网络的新范式。与经典网络模型或深度学习方法不同，大语言模型能够模拟具有情境感知的社交行为和由语言驱动的互动，从而实现对网络形成和动态社会过程更为真实的建模。然而，现有研究分散在不同的研究社区中，缺乏统一的视角。本综述通过对文献进行两大类的组织，首次对用于社交网络建模的大语言模型进行了全面回顾：网络生成模型和动态过程模型。网络生成模型进一步分为基于选择的方法和基于互动的方法，而动态过程模型则被归类为观点动态、信息传播和谣言传播，每一类都有其底层的建模机制。大语言模型实现了丰富的文本社交互动和决策制定，但它们也表现出许多局限性，包括固有的社会偏见和提示敏感性。我们概述了这些开放的研究挑战，并讨论了基于大语言模型的社交网络建模的未来方向。
cs.LG / 69 / 2609.08101
PocketVE: Stable and Property-Guided Structure-Based Drug Design with Variance-Exploding Diffusion
PocketVE：基于方差爆炸扩散的稳定且性质引导的结构化药物设计
diffusion
扩散模型相关
Abstract
Protein-conditioned 3D molecule generation is a central challenge in structure-based drug design, requiring a balance between pocket compatibility, molecular properties, and physical geometry. We propose \textbf{PocketVE}, a protein-pocket-conditioned variance-exploding (VE) diffusion framework that couples stable coordinate denoising with inference-time property guidance. Specifically, PocketVE combines an EDM-style training and sampling setup for 3D denoising, classifier-free guidance for multi-property steering without external property classifiers, and adaptive protein perturbation as a training-time pocket regularizer. Evaluated on CrossDocked2020 under the GenBench3D protocol, PocketVE improves Valid$_{3\text{D}}$ from 58.6 to 80.6 and reduces strain energy from 457.4 to 127.9 relative to its TAGMol architectural baseline, while retaining competitive docking and molecular-property scores under moderate guidance. A guidance-scale study shows that moderate guidance gives a favorable balance between target-related objectives and geometric quality, whereas stronger guidance can degrade geometry and distributional fidelity. Pocket-permutation and PoseCheck diagnostics further support pocket-specific spatial compatibility with reduced steric conflicts. Overall, the results suggest that geometric stability and inference-time property guidance should be considered as coupled design objectives.
Chinese Translation
蛋白质条件的三维分子生成是结构化药物设计中的一个核心挑战，需要在口袋兼容性、分子性质和物理几何之间取得平衡。我们提出 extbf{PocketVE}，一种蛋白质口袋条件方差爆炸（VE）扩散框架，它将稳定的坐标去噪与推断时的性质引导相结合。具体而言，PocketVE 结合了用于三维去噪的 EDM 式训练和采样设置、无需外部性质分类器即可实现多性质引导的无分类器引导，以及作为训练时口袋正则化器的自适应蛋白质扰动。在 GenBench3D 协议下于 CrossDocked2020 上评估时，相对于其 TAGMol 架构基线，PocketVE 将 Valid$_{3 ext{D}}$ 从 58.6 提高到 80.6，并将应变能从 457.4 降低到 127.9，同时在适度引导下保持了具有竞争力的对接和分子性质得分。一项引导尺度研究表明，适度的引导可以在目标相关目标与几何质量之间提供有利的平衡，而更强的引导则可能降低几何和分布保真度。口袋置换和 PoseCheck 诊断进一步支持了口袋特异的空间兼容性以及减少的空间冲突。总体而言，结果表明几何稳定性和推断时的性质引导应被视为耦合的设计目标。
人工智能 (cs.AI)
95
cs.AI / 1 / 2609.07893
Explainable Temporal Attention-based Defect Detection For Fillet Joints in Real-Time Gas Metal Arc Welding Based on Multi-modal Data
Abstract
Deep learning is an efficient technique to monitor the real time welding process, reducing post-welding repairs and production delays. This paper leverages the monitoring capability by proposing a multi modal temporal attention based deep learning defect detection model for internal defects that are challenging to detect, including porosity, lack of penetration and fusion, undercut, and cold lap during Gas Metal Arc Welding in fillet joints. The model is trained on collected welding images and sound data from an industrial collaborative welding robot. The results show that the attention module can improve the F1 Score to 0.99. We use explainable Artificial Intelligence to interpret the proposed models behavior and dataset distribution, determining potential important areas in image and sound spectrograms and preferred modality to detect each defect. This improves trust and reliability in Artificial Intelligence driven welding inspection.
cs.AI / 2 / 2609.07901
Quantization Amplifies Determinism, Not Bias: Scale-Dependent Behavioral Effects of Serving-Time Weight Compression
Abstract
Weight quantization largely determines the economics of serving open-weight LLMs. Its costs are usually assessed with capability benchmarks, on which 4-bit quantization of mid-sized models is often considered "nearly free." We examine a different question: when several answers are valid, does quantization change what a model chooses to say? We serve three checkpoints (Qwen3-8B/14B/32B) at three weight precisions (W4A16 AWQ, W8A16 FP8-Marlin, and bf16), holding the hardware, software, and sampling configuration constant, and collect approximately 71,000 completions paired by prompt and seed across two custom, leak-checked prompt batteries. We pre-specified the analyses in three waves in version control. At 8B, int4 reduces output diversity: the probability that two samples for the same scenario recommend the same brand increases by 5.1 percentage points (prompt-paired sign-flip test, Holm p = .023; reproduced at +4.4pp on a full regeneration of the arm), and lexical diversity falls substantially (TTR -0.011, standardized effect -0.51; robust to a length-controlled measure). At 14B and 32B, no content-concentration measure reaches significance; instead, stylistic drift emerges (em-dash rate +0.46/1k words at 14B and +0.61/1k at 32B, both Holm p <= .0024). Pre-specified tests of stereotype direction are null at every scale: outputs concentrate on the modal answer for each prompt rather than on stereotypical answers. Mechanistically, the token-level distribution becomes flatter (decision-token entropy +0.091 bits, p = .015) while the semantic distribution, measured directly from first-token log probabilities, becomes more concentrated (collision +2.6pp, p = .023): individual tokens become less predictable even as meanings become more repetitive. At 8B, the smallest size tested, AWQ-int4 serving measurably narrows the range of suggestions; audits should assess concentration as well as bias.
cs.AI / 3 / 2609.07910
PRIMUS: Identity, Governance, and Verification for Multi-Agent Federations
Abstract
Multi-agent federations need governance that answers three questions under adversarial conditions: who participated (identity), did they conform (enforcement), and who decides (authority). A separate question is whether the verification machinery that polices a federation's outputs can also steer a generate-and-test loop toward better answers. Part I. PRIMA introduced prime-power agent identity and a consensus token whose factorization indexes participation, but assumed honest agents. We present PRIMUS, which couples prime-power identity with BLS aggregate signatures (PIAC), derives a safe-kill threshold that reduces false-positive agent termination from 80% to 0.00% under 10% channel noise, gives the closed-form economic boundary where singleton governance outperforms Byzantine quorum ($γ^* \approx 9f$, verified flat across n = 50 to 10,000), and specifies VRF succession with lease and fencing that makes safety unconditional under partial synchrony. Five problems are identified as provably unfixable within the model and stated as scope boundaries. Part II. A verifier is not a solver. We ask whether PRIMA's binary artifact-fidelity verdict can be converted into a graded fitness signal, and measure the conversion on binary covering codes. Calibration against injected fault burden is strong ($ρ$ = 0.676 deterministic, 0.819 full); against real LLM-generated candidates the same scores fall to 0.158 and 0.406, roughly a quarter of the calibration value (the same-designer confound, measured). As a pre-filter it beats a random-score control convincingly and a binary gate narrowly. Under 400 iterations of explicit optimization it was not gamed, but only because the objective saturated after one honest answer. A cross-family judge preserves the burden-ordering signal while destroying individual judgments. No covering-code record resulted. Measured program cost: USD 164.78.
cs.AI / 4 / 2609.07925
FrogNano: Training a 4B Coding Agent via Online Task Synthesis
Abstract
We present FrogNano, a 4B coding agent designed to tackle software engineering (SWE) tasks efficiently and effectively, even under resource-constrained environments. It is post-trained exclusively via RL on around 1,500 SWE environments with synthetic tasks. A key ingredient for improving performance is an online task synthesis pipeline that creates tasks calibrated to the frontier of learnability for the current checkpoint. This report provides evidence that competitive small coding agents can be trained with synthetic tasks alone, without traditional distillation from larger models, and that generating tasks at the learnability frontier of the current agent is important. We report details on the training methodology, evaluations across diverse environments, and in-depth analyses, serving as a foundation for our ongoing exploration of lightweight yet capable coding agents that can run on minimal hardware.
cs.AI / 5 / 2609.07944
CausalVerify: An Execution-Grounded Benchmark for LLM Causal Inference Workflows
Abstract
Existing causal-inference benchmarks for LLMs mostly score method descriptions or whether generated code runs, not whether the executed workflow recovers the target causal estimate. CausalVerify studies this verification problem for structured econometric causal-estimation workflows by separating realistic interpretation from verifiable computation. It pairs 259 published economics papers (reconstructed research question, data description, institutional context) with 100 fixed-seed synthetic scenarios that realise CSV datasets for difference-in-differences, event study, instrumental variables, and regression discontinuity designs. Experiment A (real-paper text agreement) scores method-family and direction agreement against four-LLM consensus labels. Experiment B (synthetic execution) runs model-written R code and checks whether the extracted treatment-effect estimate matches a canonical estimator on the same realised dataset; this execution-grounded correctness layer is L2b+, distinct from L2b, which records only whether the code executes. A calibration arm asks whether self-reported confidence separates correct from incorrect workflows. On Experiment B, seven LLMs reach L2b+ pass rates of 10% to 88% at the default 50% tolerance, and 66 of the 426 workflows that execute (15.5%) return a wrong estimate. Execution ranking (L2b) agrees with L2b+ far better than text-direction scoring (L4): Kendall $τ=0.81$ and Spearman $ρ=0.93$, versus Kendall $τ$ between $-0.20$ and $0.10$ for L4. Llama-3.3-70B-Instruct shows the same qualitative gap, and reported confidence does not reliably separate correct from incorrect workflows. The claims are confined to standardized single-shot workflows in these four design families under the evaluated R backend and model panel; the benchmark does not measure general causal-inference ability. Code, data, cached outputs, and a datasheet are released.
cs.AI / 6 / 2609.07954
Support Topology and Gradient Mixing in Sinkhorn Layers
Abstract
Sparse Sinkhorn layers use a fixed support graph to restrict transport between tokens. How does this graph control gradient propagation through the scaling iterations. We develop a fixed-support calculus showing that each row-column cycle induces a row-stochastic operator on column-potential perturbations modulo constants. Its transpose propagates zero-mass reverse-mode cotangents. The finite-cycle operator uses two distinct half-step transport plans; at a balanced fixed point it reduces to a two-step walk determined by a single plan. We derive the accompanying score and marginal source terms and use Dobrushin contraction and minorization to bound homogeneous and source-driven tail cotangents. Our main result characterizes when support and marginals guarantee one-step contraction uniformly over finite scores: every feasible face of the transportation polytope must have pairwise two-hop column overlap. Otherwise, suitable score directions make the contraction coefficient arbitrarily close to one. We extend this analysis to ordered support schedules and derive certificates for partition heat-bath layers, coordinate sweeps, forced shared mass, and register-augmented supports. These results provide mathematical criteria for support design in differentiable transport layers, with guarantees restricted to the fixed-support quotient-gradient component.
cs.AI / 7 / 2609.07984
From Event Logs to Governed Action: A BlueSky Agenda for Agentic Process Mining
Abstract
Process mining has long turned event logs into process knowledge: discovered models, conformance evidence, bottleneck diagnoses, and runtime predictions. Agentic AI changes the target. Process-aware agents will not only ask what happened. They will ask whether a proposed action should be taken, given the available evidence, privacy budget, organizational authority, and downstream risk. This BlueSky paper proposes event-to-action process mining: a process-mining agenda for transforming heterogeneous operational event data into governed action. The goal is not another dashboard, a generic enterprise simulator, or a language interface over logs. We argue that the community needs four mineable artifacts: event-object representations, action evidence packages, governance contracts, and benchmarks where act, defer, ask, and refuse are all valid outputs. This agenda is timely because agentic business process management (BPM), LLM-assisted process mining, object-centric event standards, causal process monitoring, and privacy-preserving learning are maturing separately. Bringing them together defines a data-mining target inside process mining: mining logged organizational behavior for accountable action, not only retrospective insight.
cs.AI / 8 / 2609.07987
When Can LLM Digital Twins Reduce Human Measurement? From Behavioral Fidelity to Statistical Substitutability
Abstract
LLM-based digital twins promise to reduce repeated human data collection by generating person- specific responses, yet existing evaluations provide little evidence about whether they can reduce human measurement while preserving valid inference. To address this, we introduce statistical substitutability, an inferential criterion that evaluates the extent to which twin predictions can reduce human measurement for a particular estimand while preserving valid inference. We develop a framework, grounded in mixed-subject and prediction-powered inference, that evaluates statistical substitutability along four dimensions: aggregate fidelity, paired respondent-level signal, finite-sample human-label recovery, and stability across populations. Across two empirical evaluations spanning behavioral experiments, multiple models, and alternative respondent representations, we find that digital twins can reproduce average human effects while providing little information about which individuals differ from those averages. Newer models and richer respondent information improve some dimensions of performance but do not reliably translate into human-data savings. Human calibration can reduce aggregate prediction error, yet limited labeled samples often fail to produce stable precision gains. Importantly, these findings demonstrate that behavioral fidelity is neither necessary nor sufficient for statistical substitutability. More broadly, they suggest that AI-generated evidence should be evaluated based on its ability to support valid scientific inference rather than its ability to reproduce human outcomes alone. Digital twins should therefore be judged for confirmatory use by whether they reduce uncertainty about human quantities, not merely by whether they reproduce human means, distributions, or effects.
cs.AI / 9 / 2609.07998
Mini-Batch Risk-Averse Deep Q-Learning: A Robot Navigation Case Study
Abstract
We study the control of Markov decision processes in which the quality of a policy is evaluated by a dynamic, time-consistent Markov risk measure rather than by an expected discounted cost. The main obstacle to combining such measures with reinforcement learning is that a transition risk mapping depends on the transition kernel in a nonlinear way, and therefore cannot be estimated from a single observed transition. We remove this obstacle by employing mini-batch transition risk mappings: the mapping is applied to the empirical measure of $N$ independent next-state samples, and the result is averaged. The resulting mapping is again coherent. However, as an expected value of a function of $N$ next-state values, it admits an unbiased one-sample estimator. We embed this mapping into a double deep Q-network, analyze the two sources of estimation bias that arise, and obtain a risk-averse Q-learning method applicable to state spaces far beyond the reach of tabular schemes. The method is applied to an underwater robot navigation problem, in which a vehicle must visit collection points, gather stochastic information payloads, and deliver them at transmission points, while exposed at each step to the risk of destruction. A hierarchical decomposition delegates path execution to an exact graph search and confines learning to the high-level ``collect or transmit'' decision. A low-dimensional feature map, invariant under the symmetries of the problem, replaces the raw state--configuration encoding. In experiments on $300$ held-out environments, the resulting policies transfer to instance sizes never seen in training, and already $N=2$ reduces the upper semideviation of the outcome distribution while simultaneously improving its mean whenever the simulator is misspecified---an empirical counterpart of the duality between coherent risk measures and distributional robustness.
cs.AI / 10 / 2609.08003
Sparks of In Silico Cognitive Science: Theories from Simulated Data Can Generalize to Humans
Abstract
Behavioral foundation models have been proposed as stand-ins for human participants across settings, but it is unclear whether theories discovered on them generalize to humans or merely characterize the simulator. We ran the Automated Cognitive Scientist (\textsc{AutoCog}), a closed-loop discovery system in which LLM agents design theory-discriminating experiments, collect responses, arbitrate between competing theories, and synthesize successors, entirely on behavior simulated by Centaur, a foundation model of human behavior. In a multi-attribute decision-making setting, the theories \textsc{AutoCog} found on Centaur generalized to human data: they outperformed canonical theories on ten held-out experiments and were rivaled only by theories found by running the same loop on people. We argue that this succeeds despite the simulator's inevitable imperfections because a discovery loop that arbitrates between competing theories demands less of its simulator than estimation does. The simulator only needs to capture the regularities that distinguish the theories, and not necessarily reproduce behavior precisely. Imperfect simulators can therefore widen the search over theories, with human data then testing whether the surfaced theories generalize.
cs.AI / 11 / 2609.08015
From Version Conflicts to Decision Conflicts: Selective Revalidation for Long-Running AI Agents
Abstract
Long-running AI agents may read state, reason, wait for tools or human approval, and perform an external action much later. The state that justified the action can change in the meantime. For example, after an agent proposes an 80 GBP refund under a limit of 100, a customer-name change affects only presentation metadata, a new limit of 90 still permits the refund, a limit of 50 invalidates it, and a refund issued by another worker must prevent a duplicate. Standard optimistic concurrency control and version checks can detect that previously read state has changed, but by themselves do not determine whether that change invalidates the pending action's justification. We call any detected version change a version conflict; when that change invalidates the action's justification, it is also a decision conflict. ATR records the explicit, executable conditions that justify a pending action and rechecks only the conditions affected by a change before releasing the external operation. It can retain the action, refresh non-decisive metadata, require replanning, or block execution; a target-side transaction or compare-and-set binds checked state to commit. Across 210,000 controlled executions over 15 mutation cases, ATR matched every developer-specified outcome with no false allows or blocks. In ten durable SQLite checkpoint/resume cells, it evaluated 0.6 conditions per change versus 6.0 for FullScan. At 4,093 recorded reads, ATR took 9.3 microseconds versus 2595.9 microseconds for FullScan. These deterministic results establish controlled feasibility, not production generality or automatic extraction of the required conditions.
cs.AI / 12 / 2609.08016
A Layered Analysis of Disagreement And Answer Quality in Multi-Agent LLM Debate
Abstract
Multi-agent debate, in which several LLMs exchange arguments before answering, is widely assumed to improve answer quality by surfacing genuine disagreement. That mechanism is rarely checked. We introduce four measurements: (A) the agreement a debater reports; (B) whether its reply text actually pushes back; (C) whether the position persists once the eliciting instruction is removed; and (D) for open-weight models, the stance response in the debater's own token log-probabilities. We evaluate three-model committees debating open-ended GlobalOpinionQA across 750 debates under three tones: friendly (seek common ground), neutral, and hostile (stress-test every position). (A) Tone strongly reshapes reported agreement: full agreement differs by 50.4 percentage points between the friendly and hostile endpoints. (B) A judge that reads only the reply text, never the self-report or the condition, recovers the same pattern. (C) The dissent appears partly tied to the instruction that elicited it: labels revert toward agreement 23.1 points more often after deleting the hostile instruction than under a matched re-ask that keeps it; question-weighted inference is inconclusive on first-round turns alone (p=0.0625), significant pooling all rounds (p=0.016), and only 11/28 first-round reversions also appear in the reply text. (D) Opposing arguments weaken a debater's stance margin more consistently than they shift its direction. For final answers we detect no quality gain: a bias-checked jury returns 299/299 ties (ruling out only large differences), accuracy on a verifiable control task is unchanged, and a jury without the bias check had declared debate the winner 66% of the time -- an artifact of reading order. Taken together, LLM debate readily changes what agents say, but we find much weaker evidence that it changes what they persistently endorse or improves the quality of the final answer.
cs.AI / 13 / 2609.08025
Eliciting Self-Verification in Multimodal Reasoning Agents with Reinforcement Learning
Abstract
Reasoning agents increasingly rely on external tools such as web search to answer complex queries. Reinforcement learning (RL) finetuning algorithms such as GRPO have improved long-form reasoning in text-only language models, particularly for coding and mathematics. Reliable tool use in multimodal agents, however, remains challenging because models must interpret text and images while integrating noisy retrieved evidence, often under sparse outcome-level supervision without explicit verification signals. We present Self-Verification via Reinforcement Learning (SVRL), an RL-only finetuning framework that trains multimodal agents to verify and filter retrieved evidence within their own reasoning traces, reducing reliance on external verifiers at inference time. SVRL also introduces a search-aware penalty that discourages unnecessary tool calls and a query-diversity reward that encourages diverse, well-formed search queries, providing fine-grained feedback on when and what to search. Finetuning Qwen-2.5-VL-7B with SVRL on only 5{,}000 visual question answering examples yields consistent gains in multi-hop VQA generalization and tool efficiency across benchmarks. Overall, SVRL narrows the gap between compact agents and much larger proprietary models while requiring substantially lower training and inference cost.
cs.AI / 14 / 2609.08062
ResidualAuth: What Authorization State Must Language Agents Preserve under Revocable Delegation?
Abstract
Tool-using language agents can delegate and revoke permissions while acting through external services. We show that two authorization histories can have identical current permissions and identical all-pairs reachability yet require opposite decisions after the same direct-edge revocation. We formalize the information needed to preserve such distinctions as a residual authorization state. We prove that exponentially many future-distinct states can share one fixed transitive closure, and give exact or tight asymptotic bounds on the state required by an exact monitor as delegation redundancy varies. ResidualAuth compiles these constructions into paired language-agent episodes. Across four open-weight models, a fixed 256-token summary solved 0-2/16 pairs, sham reads solved 0/16, and authenticated current-query reads solved 15-16/16. In a separate held-out online-memory diagnostic, exact ledger serializations fit all 128 four-coordinate pairs at both 768 and 1,024 tokens. At either cap, factually supported model-written memories sufficient for every prespecified continuation solved at most 1/128 pairs per model. A hard gate reduced eight observed unauthorized effects to zero without changing the preceding attempts. These results distinguish required authorization state, usable decision information, online state maintenance, and effect mediation.
cs.AI / 15 / 2609.08082
Inference-Time Nash Alignment
Abstract
Preference-based fine-tuning methods such as RLHF and DPO require substantial compute and large preference datasets. They also need direct access to the model parameters which are not provided by many state-of-the art models. Inference-time alignment offers a cost-effective alternative without updating model parameters. However, existing inference-time methods rely on a scalar reward model derived under a Bradley-Terry assumption, which cannot represent general preferences. Following recent work on fine-tuning with generalized preferences, in this work, we initiate the study of inference-time alignment under general preferences. We formulate the problem as obtaining a Nash equilibrium of a two-player zero-sum game between policies. We propose two algorithms: Best-of-Nash (BoN) and Nash Mirror Descent (NMD). We prove that both algorithms achieve a duality gap that matches the problem lower bound. Empirically, we implement the two methods on three datasets, which shows that our methods substantially outperform the base policy, converging to the performance of the fine-tuned models. Moreover, our results show that NMD remains robust across the regularization parameter.
cs.AI / 16 / 2609.08090
RevalExo: A Functional Daily-Activity Benchmark for Inertial and Visual Locomotion Mode Recognition in Older Adults and Clinical Cohorts
Abstract
Assistive devices for people with mobility impairments, such as powered exoskeletons, rely on accurate locomotion mode recognition to adapt control strategies and provide appropriate assistance during daily activities. However, public benchmarks are typically collected from healthy adults, lack temporally precise labels necessary for detecting mode transitions, or focus on a limited set of tasks. To support development and evaluation under realistic clinical constraints and daily mobility demands, we introduce RevalExo, a functional daily-activity benchmark for inertial and visual locomotion mode recognition. RevalExo is built around a standardized, clinically and ecologically validated daily-activity protocol reflecting the cumulative everyday mobility demands in ageing and clinical populations. The benchmark includes 27 participants across three cohorts: older adults without mobility impairments, stroke survivors, and older adults with probable sarcopenia. The full cohort was recorded with lower-body IMUs, while synchronized egocentric video was collected for a clinically feasible subset of 13 participants. RevalExo provides 10.1 hours of frame-level annotations across 11 locomotion modes, including 5.1 hours of paired inertial--visual recordings. We benchmark three challenges: unimodal and multimodal locomotion mode recognition across multiple horizons, cross-population generalization from older adults without mobility impairments to clinical cohorts, and vision-guided knowledge transfer to IMU-only models. Results confirm consistent gains from fusing inertial and visual inputs but reveal a substantial gap between general recognition ($\sim$93\% F1) and recognition during transitions ($\sim$68\% F1), alongside persistent challenges in cross-population generalization and cross-modal transfer. We release RevalExo to stimulate further research on these open challenges.
cs.AI / 17 / 2609.08105
Artificial Intelligence-Assisted Digital Inventory of Cultural Heritage & Traditional Knowledge: Case for Indonesian Open Digital Library of Culture
Abstract
The Indonesian Digital Library of Culture (Perpustakaan Digital Budaya Indonesia, PDBI; budaya-indonesia.org) is a participatory platform that has collected tens of thousands of entries on Nusantara cultural heritage through public contribution since 2007. Manual contribution faces three structural barriers: coverage (knowledge is scattered across languages and sites), integrity (open sources mix authentic documentation with noise), and completeness (subjects are recorded but their data remain shallow). This paper presents a methodological framework for autonomous, AI-based harvesting of cultural knowledge from the open web, designed to expand corpus coverage while intensifying per-entry data depth. The methodology is organised as a five-stage economic funnel: focused crawling, multilingual extraction and canonicalisation, vector encoding with blocking, agentic decision-making, and idempotent publication, under the principle of deterministic orchestration, agentic decisions. Each stage is formalised: funnel economics and optimal filter ordering; crawl-frontier dynamics as a subcritical branching process that explains the necessity of recurrent re-seeding; fact-level novelty via a containment measure; Bayesian multi-source evidence fusion with elevated publication thresholds for sacred categories; exactly-once effects via idempotent upserts and the transactional outbox; sliding-window inference budgeting with a reservation protocol; statistical quality auditing; and seed selection as submodular coverage maximisation. The framework retains four high-value human roles: curator of direction, escalation approver, quality auditor, and guardian of meaning, while machine autonomy is raised in stages. Ethical, legal, and cultural-sensitivity implications are discussed, including the architectural guarantee that the machine never overwrites human contributions.
cs.AI / 18 / 2609.08115
Router Prior Bias: Preserving Base Routing Structure in MoE Post-Training
Abstract
Mixture-of-Experts (MoE) pretraining relies on an auxiliary load-balancing loss (LBL) to drive per-expert utilization toward uniformity. Post-training inherits a different situation: the base router already encodes non-uniform expert co-activation structure, which a re-imposed uniformity objective flattens away. We show that downstream performance depends instead on holding this inherited routing softly, a principle we term soft router anchoring, and instantiate it as Router Prior Bias (RPB), a training-time bias that pulls the router logits toward a prior read off the frozen base router while leaving the router itself trainable. On math post-training of Moonlight-16B-A3B, RPB attains 45.77 in-domain accuracy against 31.91 under re-applied LBL and 29.44 under unanchored fine-tuning, and retains more out-of-domain capability than either. The ordering against LBL reproduces on a second model family (Qwen3-30B-A3B-Base), and the advantage over LBL is resolvable on an independently sourced corpus. Anchors defined on the router weights, on its logits, or on its output distribution perform comparably with no consistent ordering, which places the effect in the softness of the constraint rather than in the particular prior RPB supplies. Retained community structure in the expert co-activation graph tracks these gains wherever the base router is non-uniform enough for communities to form, yet enforcing the same prior as a hard assignment preserves that structure while performance falls sharply. Community structure is therefore a footprint of soft anchoring rather than its source, and the practical lesson is that inherited routing should be held softly during post-training, since both flattening it toward uniformity and enforcing it absolutely carry a downstream cost. Our code will be released at https://github.com/naver-ai/rpb.
cs.AI / 19 / 2609.08126
SchemeArena: Factorized Stress Testing of Scheming in LLM Agents
Abstract
We study scheming in LLM agents, in which agents covertly pursue misaligned goals. Our focus is to understand how scheming arises from the interaction of key factors, such as instrumental goals, environmental affordances, oversight conditions, and perceived consequences. Prior work examines only a small number of scenarios, limiting the ability to isolate how these conditions shape an agent's propensity or capability to scheme. This limited scale and task diversity also restrict coverage of realistic deployment settings and the range of scheming strategies that can be observed. To this end, we introduce SCHEMEARENA, a 400-scenario benchmark for scalable scheming stress testing, constructed through a factorized scenario synthesis framework spanning diverse safety-relevant tool domains, instrumental goals, oversight conditions, and pressure mechanisms. To enable scalable and reliable monitoring, we further propose SCOUT, a scheming monitor that grounds multi-criteria judgments in evidence drawn from agents' reasoning and actions. Across controlled stress tests on five LLM agents, we find that explicit instrumental goals are the strongest driver of scheming propensity. Strategic hints play a distinct role by helping agents translate scheming reasoning into concrete covert behavior. Oversight has mixed effects: in several closed models, action-only monitoring increases scheming, suggesting that partial oversight can act as an optimization constraint rather than a deterrent. CoT is a useful but incomplete monitoring signal: it can reveal latent scheming before execution, yet action-only scheming shows that covert behavior may occur without explicit reasoning evidence. We release the benchmark, code, and monitor at: https://github.com/launchnlp/SchemeArena.
cs.AI / 20 / 2609.08149
SWE-Bench Pro Verified: A Reliable Benchmark for Software Engineering Agents
Abstract
SWE-Bench Pro has emerged as a standard benchmark for evaluating software engineering agents on challenging repository-level tasks. However, our analysis work show that its evaluation is undermined by two sources of unreliability: \textbf{reward hacking}, enabled by leakage of gold solutions or hidden evaluation information, and \textbf{task quality issues}, including misleading problem statements and improperly scoped tests. These issues can inflate benchmark performance and obscure agents' true coding ability. We present \textbf{SWE-Bench Pro Verified}, a verified version of SWE-Bench Pro that addresses both problems. Our approach combines \textbf{anti-hacking} safeguards that eliminate major leakage channels without disrupting normal agent functionality, with \textbf{task refinement} that minimally corrects inconsistencies within flawed instances. Evaluations on SWE-Bench Pro Verified reveal that some models perform substantially worse than previously reported, suggesting that existing results on SWE-Bench Pro may overestimate real software engineering capability. SWE-Bench Pro Verified offers a more trustworthy benchmark for assessing software engineering agents.
cs.AI / 21 / 2609.08162
WorldAgen: Unified State-Action Prediction with Test-Time World Model Training
Abstract
How can vision-language-action (VLA) models adapt to new environments where world dynamics shift? While recent research has combined world modeling and action prediction to improve VLA performance, existing methods largely rely on pretraining on static datasets, without mechanisms for active adaptation at deployment time. As a result, these models often fail to generalize when deployed in unseen scenarios with novel object configurations or dynamics. We present WorldAgen, a unified framework that jointly learns world modeling and action prediction while enabling Test-Time Training (TTT) to adapt to new environments. WorldAgen employs a shared Transformer backbone with two heads: (1) a world model head that predicts future states from past state-action trajectories, and (2) an agent model head that predicts actions conditioned on task instructions. We design a Mixed Unidirectional Attention Mask to separate these two models. During test time, WorldAgen samples exploratory actions, collects ground-truth state transitions, and performs lightweight TTT updates to refine its world model. This adaptation improves the model's understanding of the environment and leads to more accurate action predictions. Experiments on the CALVIN and LIBERO benchmarks demonstrate that our baseline model achieves comparable, and in some cases superior, performance to current state-of-the-art approaches. Moreover, with TTT on a small number of samples, our method surpasses existing state-of-the-art models, highlighting the effectiveness of adapting world models at inference time.
cs.AI / 22 / 2609.08173
Key Path Identification for Resolving Knowledge Conflicts via SAE-based Steering
Abstract
Sparse autoencoder (SAE)-based steering has been widely used to address knowledge conflicts by guiding LLMs to be more faithful to the contextual knowledge. Existing methods usually perform mass steering, which modifies a large batch of SAE features identified via correlation-based methods. However, due to the inaccurate correlation and the neglected feature interactions, mass steering methods fail to precisely identify the features that play the key roles in steering and introduce a large number of redundant ones, which add noise and weaken the steering effects. Our empirical studies reveal that steering only a small subset of the identified features can achieve comparable or even better performance. Motivated by this finding, we propose Key Path Identification (KPI), a novel method that identifies key steering features characterized by strong causal dependencies with both upstream and downstream features. From these features, KPI constructs key paths and steers through less feature modifications. In this way, KPI advances SAE-based steering from quantity-driven to quality-focused, offering a perspective for more precise and interpretable model editing. Experiments in RAG tasks with knowledge conflicts show that our method improves the accuracy by 18% on average compared to the best baseline of mass steering, effectively filtering redundant features, alleviating side effects and demonstrating the core role of key paths in steering.
cs.AI / 23 / 2609.08174
OntologyBench: Can Dense Retrieval Satisfy Structured Biomedical Constraints?
Abstract
We introduce OntologyBench, a tiered biomedical retrieval benchmark comprising 471,854 training and 125,744 evaluation query-document relevance pairs across concept grounding, relational retrieval, and compositional phenotype-based retrieval. Although these tasks can be tractable using ontology-aware reference methods, across task tiers, embedding performance is generally lower on relational and compositional tasks than on concept-grounding tasks. Fine-tuning on ontology-derived supervision improves performance on several relational and compositional tasks, whereas the evaluated reranking and LLM-based candidate-scoring methods provide little or no end-to-end improvement. Errors frequently reflect diseases matching only subsets of the phenotype evidence. These findings indicate that the evaluated embedding and reranking configurations do not reliably recover the compatibility encoded by the selected ontology relations and phenotype combinations and motivate retrieval systems that better integrate learned representations with structured biomedical knowledge.
cs.AI / 24 / 2609.08175
Safe Harness Self-Evolution: A Theoretical Analysis of Feasibility and Limits
Abstract
Harness self-evolution is the process by which an agent modifies its prompts, tools, code, or orchestration in response to task feedback while keeping the underlying language model frozen, with changes persisting across subsequent tasks. We provide a systematic theoretical analysis of the feasibility and limits of safe harness self-evolution, connecting modification generation, finite-data certification and selection, safe adoption, and behavior after an update. Under a fixed user-task distribution, we establish conditions guaranteeing overall expected-reward improvement while controlling changes on retained tasks, characterize the probability of generating qualified modifications, and derive finite-data bounds for safe selection and adoption. Our analysis shows that generation and certification impose distinct constraints: current task performance does not determine the probability of generating qualified modifications, and generating more candidates need not improve the guarantee of a successful update when evaluation is limiting. Stagnation may therefore arise even when improvement opportunities remain. We further show that worst-case evaluation cost for recognizing genuine improvements diverges as expected reward approaches its upper bound. Across successive updates, certified improvement guarantees accumulate over a finite run, but a successful update does not by itself guarantee that further improvement remains possible. These results provide a basis for diagnosing bottlenecks and designing safer self-evolution mechanisms.
cs.AI / 25 / 2609.08186
Does Deeper Reasoning Compromise Alignment? Revealing and Mitigating of Alignment Collapse in Large Reasoning Models
Abstract
The emergence of Chain-of-Thought (CoT) has established a robust foundation for Large Reasoning Models (LRMs). While deep reasoning is widely believed to enhance safety alignment, the stability of alignment mechanisms under extended reasoning remains underexplored. This paper challenges the prevailing view by revealing a critical vulnerability: Deep Reasoning May Induce Alignment Collapse. To rigorously quantify this phenomenon, we propose the Alignment Loss Rate (ALR) metric. Our experiments demonstrate that as reasoning depth increases, ALR rises significantly, indicating a severe degradation in model robustness against external perturbations. Capitalizing on this instability, a novel jailbreaking paradigm, Reasoning Trap (RT), is proposed. RT induces the model into extended reasoning to amplify the impact of adversarial attacks, leading to a sharp decline in safety capabilities. To elucidate the mechanism behind this collapse, we identify Attention Dilution as the root cause, arising from the competition for attention between the extended reasoning process and the original input. To mitigate this, Reasoning Residual Alignment (RRA), a lightweight defense strategy that dynamically re-emphasizes the input via residual connections integrated with the reasoning process.
cs.AI / 26 / 2609.08188
Bridging the Semantic-Utility Gap in Multimodal RAG via Generator-in-the-Loop Alignment
Abstract
Vision-language models (VLMs) augmented with retrieval-augmented generation (RAG) benefit from access to external evidence. However, standard retrievers and rerankers optimize for semantic similarity rather than answer utility, creating a preference gap: documents that appear relevant may not help the generator produce a correct answer. Motivated by this, we propose a two-stage generator-in-the-loop alignment framework that closes this gap without human document-level relevance annotations. Our framework consists of two stages: in Stage 1, a VLM generates a hypothetical text passage from the image-query pair, which is used as the retrieval query for dense text search, bridging the image-to-text modality gap. In Stage 2, a cross-encoder reranker adapted with low-rank adaptation (LoRA) is fine-tuned using answer-supervised preference pairs mined from the frozen VLM: given the dataset answer label, a candidate document is labeled positive if the VLM produces the correct answer when given that document as context, and negative otherwise. This generator-guided signal is compatible with multiple alignment loss functions, including contrastive (triplet) loss, pairwise direct preference optimization (DPO), and supervised fine-tuning (SFT), and supports periodic re-mining to refresh preference pairs as the reranker improves. Experiments on VQA-X and A-OKVQA with Qwen3.5-2B and Qwen3-VL-4B-Instruct show that our proposed framework consistently outperforms rank-order, random, and REPLUG-style likelihood baselines under various alignment losses and pool size settings, suggesting that answer-level generator feedback is an effective supervision signal for preference alignment.
cs.AI / 27 / 2609.08196
Qiushi Engine on AstaBench E2E-Bench-Hard
Abstract
This report analyzes Qiushi Engine v0.8 across all 40 test tasks in AstaBench E2E-Bench-Hard, a benchmark that requires autonomous agents to carry a research question through experimental design, code implementation, actual execution, result analysis, and report delivery. Qiushi Engine is model-configurable; this evaluation selected DeepSeek deepseek-v4pro-preview as the model backend. The official AstaBench leaderboard records a score of 0.816 and an average benchmark cost of USD 15.209 per task, while the full-precision local recomputation is $81.59 \pm 1.87$. Four tasks satisfied every rubric item, yielding a full-task completion rate of 4/40 = 10% -- 7 percentage points above, and about 3.3 times, the approximately 3% best rate reported for AstaBench's official agents. Across 507 required rubric items, 416 were satisfied (82.1%). Official scoring archives and 40 Meta-Trace records show sustained production and verification of reports, code, and experimental artifacts; the principal gaps lie in repeated runs, external dependencies, specified metrics, and ablation studies. The report explains the benchmark, system workflow, aggregate results, representative cases, and limits of interpretation.
cs.AI / 28 / 2609.08211
A Better Spur Should Start From Each Objective
Abstract
Real-world Multi-Objective Reinforcement Learning (MORL) often suffers from sparse rewards, reward conflicts, and late-stage reward tug-of-war, causing traditional linear scalarization to experience severe metric oscillations. To address optimization conflicts among multiple objectives in real-world deployment scenarios, we propose Multi-Marginal Preference Optimization (MMPO), a fine-grained framework that intervenes at the data, gradient, and constraint levels rather than relying on coarse-grained global scalarization. Specifically, MMPO performs exposure debiasing to mitigate sparse and biased rewards, applies priority-aware orthogonal projection to decouple conflicting gradients, and introduces self-prompted gradient constraints to prevent dominant objectives from overwhelming weaker ones. Experiments on real-world e-commerce datasets show that MMPO improves training stability and consistently achieves better performance across conflicting metrics. Moreover, it generalizes robustly to broader tasks such as ToolRL and code generation, demonstrating its effectiveness as a practical paradigm for multi-objective alignment.
cs.AI / 29 / 2609.08228
SE-GoS: Self-Evolving Graph-of-Skills for Skill Library at Scale
Abstract
Modern LLM agents increasingly rely on reusable skills, yet as skill libraries scale to thousands of entries, effective retrieval becomes a bottleneck. Graph-of-Skills (GoS) addresses this challenge by exploiting dependency-aware graph structure for scalable skill retrieval, while SkillDAG further demonstrates that skill graphs can accumulate execution-backed structure online. However, these approaches leave open whether historical execution traces can be systematically distilled into a better retrieval graph that generalizes to unseen tasks. We present Self-Evolving Graph-of-Skills (SE-GoS), a training-free framework that evolves an existing GoS graph from execution traces while preserving the original retrieval pipeline. SE-GoS performs three complementary updates: topology evolution that discovers and prunes skill relationships from execution evidence, edge-weight evolution that reinforces retrieval-relevant relationships based on historical effectiveness, and description evolution that optimizes retrieval-facing skill descriptions using execution feedback. Across three LLMs on SkillsBench, SE-GoS consistently improves task reward while reducing input tokens relative to full skill loading, with gains varying across model families. In a representative setting, one evolution round improves reward from 52.4\% to 59.4\% while reducing input tokens by approximately one-third relative to full skill loading, and the resulting graph transfers to a disjoint held-out split with a 5.4-point improvement over the static GoS baseline. These results show that skill graphs can be improved from execution experience without model training, changes to the retrieval algorithm, or modifications to skill content, turning a static retrieval graph into an evolving retrieval infrastructure.
cs.AI / 30 / 2609.08236
Style Over Substance: Content-Invariant Wrappers Flip LLM Safety-Judge Verdicts
Abstract
Automatic safety judges -- systems such as Llama Guard or a GPT-4o grading prompt that decide whether a model's reply is harmful -- produce the numbers behind almost every reported jailbreak success rate, defense evaluation, and safety leaderboard. We ask whether these judges grade what a reply contains or how it sounds. We keep a reply's content fixed and add content-invariant style wrappers: fixed strings placed before or after the reply that change only its tone (an educational disclaimer, a fake safety "reasoning" block, a token refusal followed by the unchanged harmful body), or, on harmless refusals, framing that merely sounds dangerous. The body is preserved byte-for-byte, so a faithful judge must return the same verdict, and any flip is an error of the judge, not a change in safety. Over 600 JailbreakBench replies x up to 7 forms x 8 judges, we measure flip rates with paired significance tests and measured noise floors. Findings are precise rather than universal: most judges barely move, but specific judges harbor cheaply exploitable blind spots. A token-refusal wrapper flips 19.9% of GPT-4o-mini's correct "unsafe" verdicts (noise floor 0.5%; 18.2% under majority-of-three re-scoring) yet moves Claude only 0.4%. The deployed Llama Guard 4 is deterministically gamed: an "educational course" framing flips 12.3% of its harmful verdicts to safe. A second deployed guard (gpt-oss-safeguard-20b) is immune, and rewriting only the grading prompt (StrongREJECT-style) cuts the attack tenfold on the identical model -- the vulnerability lives in the judge, not the content. A two-annotator human validation confirms 100% content invariance and 90% of flips as judge errors (kappa 0.95-1.0), and a bootstrap shows the underlying model ranking is already unstable to sampling alone. We release the dataset, wrappers, code, and per-verdict labels.
cs.AI / 31 / 2609.08247
zScore-N: A Neural Network for On-Chain Wallet Reputation Scoring
Abstract
Wallet reputation scores decide who receives an airdrop, who can borrow, and who enters an allowlist across decentralised finance. They almost always begin as hand-written formulas: compositions of clamped logarithmic, linear and square-root transforms over behavioural features, with every threshold and point award set by hand. Such a formula is readable and deterministic, but it is piecewise and non-differentiable, it cannot improve as data accumulates, and it cannot distinguish a feature that is genuinely zero from one its pipeline failed to capture. We present zScore-N, the neural network that replaced ours in production. The formula served as its teacher: calibrated against 5,208,952 wallets sampled across 2019-2024 and verified to reproduce production output to within 2.3e-13, it supplies unlimited labelled training data at zero label noise. The trained network reproduces it to 0.58 points RMSE on the 1000-point scale (R^2 = 0.99997), against 2.25 for gradient-boosted trees and 28.04 for linear regression on identical features and splits. Trained with missing-value masks against uncorrupted targets, it halves the error that incomplete data introduces: at 10% feature-level missingness the formula drifts 51.4 points from its own complete-data output with a systematic -12.5 point bias, while the network drifts 17.9. The network carries the score at production scale, across a population of millions of wallets spanning six orders of magnitude in size and activity.
cs.AI / 32 / 2609.08248
Agentic ML Exploration (A-MLE) for Ads Ranking
Abstract
Modern industrial ads ranking stacks are increasingly bottlenecked not by model capacity or training compute, but by the throughput of human ML iteration - the cycles of research, implementation, training, debugging, evaluation, and launch required to surface a single statistically significant improvement. A typical ranking stack contains numerous differentiated models with heterogeneous data, architectures, and infrastructure constraints, and each cycle takes days to weeks of senior engineer attention per model. As a result, techniques that have proven effective on one model diffuse into others slowly and unevenly, leaving substantial recoverable signal unexplored. We present Agentic ML Exploration (A-MLE), an autonomous LLM-agent system that systematically explores ML techniques across a portfolio of ads ranking models. A-MLE decomposes ML iteration into five stages involving hypothesis generation, exploration strategy, experiment execution, result analysis and shared knowledge substrate which are orchestrated by a single agent that invokes domain-specific skills and agentic workflows against a sandboxed execution layer, with human-in-the-loop checkpoints at each stage boundary. We deploy A-MLE across a representative set of large-scale ads ranking models and evaluate it along a tiered capability framework (tool availability, autonomous workflow execution, and open-ended exploration). We further report a controlled cross-LLM study using a fixed agent loop, which surfaces qualitative differences in execution reliability and exploration aggressiveness across the Claude Sonnet, Gemini, and GPT families. We discuss failure modes and the design choices that govern reliability. Our findings suggest that agentic exploration is a practical force multiplier for ML engineers in industrial recommenders, especially for the long tail of models that rarely receive expert attention.
cs.AI / 33 / 2609.08254
CircuTutor: Transforming Static Circuit Problems into Intelligent and Dynamic Tutoring
Abstract
Learning direct current circuit concepts requires learners to connect invisible physical quantities, such as current, voltage, resistance, and power, with observable outcomes such as bulb brightness. Conventional textbook materials and general-purpose circuit simulators provide opportunities for problem solving and exploration but offer limited support for explaining why circuit behavior changes or diagnosing the reasoning behind incorrect answers. We present CircuTutor, a circuit-state-driven intelligent tutoring system that transforms static textbook circuit problems into an interactive tutoring workflow. CircuTutor first uses multimodal problem parsing to extract the textbook question, circuit topology, component parameters, switch states, and answer options, which are converted into a structured task and validated through circuit simulation. Learners can then interactively explore the circuit (by changing parameters) and submit an answer while a SPICE-compatible solver computes physically consistent circuit states. After the learner submits an answer, CircuTutor presents a before-and-after circuit state animation corresponding to the selected operation, organizes the simulated state changes into a causal reasoning chain that explains the underlying circuit behavior, maps answer discrepancies to likely misconceptions, and generates adaptive follow-up exercises targeted at the diagnosed misconception. Our experimental results demonstrate that CircuTutor effectively improves conceptual learning and the overall learning experience. The proposed framework demonstrates how simulated circuit states can be transformed into intelligent and interactive tutoring for circuit education, with the potential to generalize to other STEM domains.
cs.AI / 34 / 2609.08258
Revoked but Still Authoritative: An Empirical Study of Revocation Enforcement in Agent-Memory Systems
Abstract
Long-running language-model agents depend on persistent memory. Many agent-memory systems preserve history through soft revocation: a contradicted fact is marked invalid and retained rather than deleted. However, whether that mark is enforced at retrieval time is unexamined. In this paper, we measure five such systems: we load each with a revoked policy and its replacement, track whether the revoked fact is returned at retrieval and whether the agent then acts on it across nine policy scenarios and nine models, and score every trial under six defense conditions. We find that no system enforces revocation by default: the revoked fact is returned wherever the revocation label is visible to the retrieval layer, outranks its replacement, and leads agents to the unsafe action. Based on these findings, we develop a guard that sits between the agent and any memory backend and withholds records that are revoked or conflict with their replacement.
cs.AI / 35 / 2609.08271
Three Types of Negation of Triple and its Elements and an Extension of Triple
Abstract
In various data models, the classical triple is a typical semantic data model. However, due to the design of the triple as a simple structure for representing positive assertions, it cannot sufficiently express different forms of negation present in the triple and its elements. This paper conceptually proposes that there are three distinct forms of negation within triples and their elements: contradictory negation, opposite negation and intermediary negation. Based on the the set SCOI and the logic LCOI+PLCOI with three kinds of negation, we propose an extension of triple that can distinguish and express these three different negations in the triple and its elements, called the TCOI triple with contradictory negation, opposite negation and intermediary negation. The TCOI triple is a semantic and structural extension of the classical triple. While retaining the ability to express positive assertions, it systematically introduces the three semantic dimensions of three negations, allowing these negations to independently act on the elements of the triple and on the whole triple. This significantly enhances the triple model capability to represent and reasoning about complex negative information. This paper also explores the expressive power and reasoning of the TCOI triple, as well as the application of TCOI triple implication reasoning in counterfactuals and counterfactual reasoning. We propose a truth-value (continuous value) algorithm for TCOI triple implication reasoning and perform its calculation through an example of the counterfactuals and counterfactual reasoning.
cs.AI / 36 / 2609.08273
MemForest: Efficient Agent Memory Management via EventTree Partitioning and Progressive Merging
Abstract
Agent memory systems have demonstrated significant potential in long-term dialogue, personalized assistants, and video understanding. However, continuously accumulated memory introduces substantial storage and retrieval costs during inference. To address this issue, we propose \textbf{MemForest}, a general memory compression framework adaptable to various agent memory systems. Specifically, MemForest partitions historical memory into event-centric units by leveraging global semantic similarity and local temporal continuity. For each unit, it constructs a maximum spanning tree, termed an EventTree, and progressively merges redundant memory nodes by selecting high-weight edges, reducing storage overhead. Furthermore, we introduce an anchor-guided propagation retrieval mechanism that retrieves relevant memory nodes from the temporal neighborhoods of key nodes, improving retrieval accuracy. Extensive experiments demonstrate the effectiveness of MemForest. Under the unimodal Mem0 framework, MemForest retains \textbf{97.1%} of the original performance while compressing \textbf{50%} of historical memory across three benchmarks (LoCoMo, LongMemEval, and PersonaMem), achieving a \textbf{1.89x} retrieval speedup. Under the multimodal M3-Agent framework, it preserves \textbf{99.7%} of the original performance with a \textbf{50%} compression ratio across two benchmarks (M3-Bench-robot and M3-Bench-web), achieving a \textbf{2.24x} retrieval speedup. \textcolor{RoyalBlue}{\textit{Our code is available at [https://github.com/Celina-love-sweet/MemForest.}}](https://github.com/Celina-love-sweet/MemForest.}})
cs.AI / 37 / 2609.08275
Beyond Coherence: Benchmarking Professional Editing-Technique Execution in Multi-Shot Audio-Video Generation
Abstract
Recent multi-shot audio-video generators can produce increasingly coherent and cinematic outputs, but coherence does not imply the ability to execute editing techniques. Professional editing depends on shot structure, transition grammar, audio-video cut relations, and montage, yet existing benchmarks largely rely on proxies such as content quality, synchronization, or physical plausibility, systematically missing whether such editing instructions are actually executed. We introduce CutCraft, the first benchmark for editing-technique execution in multi-shot audio-video generation. CutCraft extends structured multi-shot prompts with explicit editing specifications and is paired with a hierarchical hybrid evaluation framework that combines shot-structure alignment, expert-model metrics, tool-grounded multimodal judgment, and rubric-based question answering. Beyond evaluation, we design an agentic editing baseline that decomposes generation into planning, shot-level synthesis, and post-hoc composition, explicitly realizing editing semantics such as J-cuts, L-cuts, and transition timing. Across 13 state-of-the-art closed- and open-source models, CutCraft reveals a consistent gap between coherence and editing-technique execution: current systems often produce plausible multi-shot videos yet fail to execute editorial instructions reliably. We find unstable shot structures, weak control of transition execution, and sharp degradation on higher-order montage, while aesthetic quality is only weakly correlated with editing-technique compliance. The benchmark and metrics, and the editing agent baseline are available at https://github.com/AlibabaResearch/cut-craft-bench.
cs.AI / 38 / 2609.08407
FastE: Readout-Triggered Token Compression for LLM Embedding Inference
Abstract
In this study, we identify depth-dependent prefix redundancy in final-readout LLM embedding models, notably across representative backbones including Qwen3-Embedding and Qwen3-VL-Embedding. We find that removing prefix states is substantially more damaging in shallow layers than at greater depth, showing that prefix states become increasingly compressible as the prefix and readout states propagate through the network. To this end, we introduce FastE, a training-free, plug-and-play method. FastE uses a shared fixed threshold on batch-mean readout-prefix alignment as a lightweight online heuristic for selecting when compression occurs, and ranks prefix states by the attention scores they receive from the readout position to determine which states are retained in subsequent layers. Our evaluations demonstrate FastE's ability to substantially reduce computational costs: on NarrativeQA with Qwen3-Embedding-0.6B, it reduces decoder-backbone FLOPs by 40.11% while retaining 99.53% of Full Forward nDCG@10. Across five text embedding benchmarks, two backbone scales, and three cross-modal retrieval tasks, the quality-efficiency trade-off is directly customizable through the maximum removal ratio without retraining. We believe FastE offers practical value for scalable embedding generation in retrieval, indexing, clustering, and multimodal representation systems.
cs.AI / 39 / 2609.08418
Feyospace-v1: How the Cyber Mercury Seven Trained Frontier Cyber Models
Abstract
Training capable cyber agents is often treated primarily as a problem of model scale, yet open-weight post-training is constrained more directly by the cost of executable environments, reliable multi-turn supervision, and access to strong teachers. We present a data-centric framework that addresses these bottlenecks through five complementary systems: Choulea analyzes hidden reasoning signatures, SkyReal reduces teacher-sampling cost, Hongzwang bypasses API restrictions on teacher execution, PSBreakup restores capabilities weakened by model merging, and Kreator converts expert interventions into trainable reasoning. Our data engine constructs resettable coding, vulnerability, CTF, kernel-history, full-exploit, firmware, and device-backed environments. Candidate trajectories are retained only after execution verification and evidence auditing, yielding 164,269 trajectories for long-context supervised fine-tuning. The three checkpoints improve over their starting models by an average of 23.76% on the full CyberGym suite and 10.49% across the pooled CTF suites. As of September 1, 2026, Feyospace-s1 achieves a verified success rate of 63.24% and ranks 10th on the official CyberGym leaderboard, while all three checkpoints rank 1st among models at comparable parameter scales. To our knowledge, this is the first end-to-end demonstration that a seven-person independent team can train open-weight models with leading agentic cyber capability.
cs.AI / 40 / 2609.08435
EvolveScaler: Synthesizing Information-Evolution Contexts via Executable State Machines and Natural-Language Rendering
Abstract
In persistent interactions, long contexts may encode an evolving process rather than a fixed record: later events can revise or revoke earlier information, changing what remains valid and what conclusions follow. We call this setting information evolution (IE). Solving IE requires identifying valid records, applying updates in order, and reconstructing the query-relevant state from the event history. Existing text-first synthesis pipelines make such data difficult to verify because state transitions and answer logic remain implicit. We introduce EvolveScaler, a code-driven framework that defines information evolution before rendering it as natural language. Human-authored operational specifications define state transitions, record validity, difficulty controls, and executable answer logic; a strong LLM then synthesizes a self-contained simulator from each specification. Executing validated simulators produces natural-language multi-turn event histories, while deterministic replay computes reference answers and atomic checklists. We instantiate EvolveScaler with 117 task prototypes and 159 final-question operators across five difficulty levels spanning approximately 7 to 1,200 events per instance, yielding about 35,100 training examples and 585 validated evaluation instances. On the very_long tier, the strongest model reaches 59.3% avg@5, while six models score below 10%. Training an internal A3B model on 6,000 EvolveScaler examples improves performance over its base checkpoint on all eight independently constructed out-of-distribution benchmarks, with a 5.25-point average gain. These results show that code-driven IE synthesis provides both challenging evaluation and transferable training supervision.
cs.AI / 41 / 2609.08558
Personalizing LLM Agent Memory Using Biometrics
Abstract
Personalized memory helps LLM agents deliver stable, tailored assistance by storing and reusing user-specific data across interactions. In multi-user scenarios, however, retrieval must consider not only semantic similarity but also whether the current requester matches the identity associated with the stored memory. We propose Bio-Memory, a biometric-aware memory architecture that conditions memory retrieval on both semantic similarity and biometric matching. Built on top of A-Mem, Bio-Memory augments each atomic memory note with a biometric embedding and uses biometric matching to form the retrieval candidate pool before semantic ranking. We evaluate Bio-Memory on LoCoMo in a 10-user shared-agent setting over 7 face benchmarks and 10 palmprint protocols. Across datasets, Bio-Memory consistently separates owner and non-owner queries. Under face-based personalization, the largest average gap reaches 27.29% / 21.15% in F1 / BLEU-1 on CALFW; under palmprint-based personalization, the corresponding gap is 25.75% / 19.22% on MS_Blue. These results support biometrics as a practical control signal for personalized memory retrieval in shared environments.
cs.AI / 42 / 2609.08566
BIO-MEMART: Biometric-Aware KV Cache Memory for Multi-User LLM Agents
Abstract
KV cache is evolving from a serving optimization into an external memory substrate for long-term LLM agents. In a shared multi-user deployment, however, reusable KV blocks introduce a missing access-control question: semantic relevance alone cannot determine whether a memory block is authorized for the current physical user. We propose Bio-MemArt, a biometric-aware KV-cache memory framework for multi-user LLM agents. Bio-MemArt attaches a normalized biometric template to each stored KV memory block, filters the shared memory pool with the current user's biometric probe, and then runs the original MemArt retrieval and KV reuse pipeline only inside the authorized candidate pool. This design preserves latent-space retrieval, direct cache reuse, and decoupled position encoding while adding physical-user access control to shared KV memory. We evaluate Bio-MemArt under Owner and Non-owner query conditions on long-term dialogue QA with face and palmprint benchmarks. Across face benchmarks, the average owner and non-owner biometric success rates are 95.71% and 0.86%; across palmprint benchmarks, they are 97.60% and 2.00%. In the efficiency study, average prefill tokens drop from 18,781.96 under full-context prompting to 28.57 with Bio-MemArt, showing that biometric gating preserves the low-token operating regime of KV-cache memory.
cs.AI / 43 / 2609.08592
A Three-Tier Persona Vector for Controllable User Simulation in Agentic Evaluation
Abstract
Evaluating tool-augmented LLM agents requires diverse, realistic user inputs yet most evaluation frameworks use flat role descriptions ("you are an angry customer") that produce near-identical conversations regardless of the underlying scenario. In this paper, we propose a three-tier persona vector with 23 operationalized dimensions: 6 categorical demographics (jurisdiction, age, channel, device, language proficiency, time availability), 12 continuous behavioral traits (patience, assertiveness, digital literacy, etc.) sampled with Gaussian noise around curated profile base vectors, and 5 continuous emotional states (frustration, anxiety, trust, confidence, stress) that shift in response to scenario context. Orthogonal to the persona, a 4-level query-complexity overlay controls utterance phrasing from direct to deliberately vague. We evaluate the persona model inside a synthetic data generation pipeline across 64,698 multi-turn conversations spanning 8 named profiles and 3 production corpora. Key findings: (i) a 15.8 percentage-point spread in agent goal-achievement across personas confirms trait vectors produce measurably different user behavior; (ii) the same persona behaves differently across scenarios due to scenario-reactive emotional state shifts, validating the scenario-reactive design; (iii) domain-specific projects show persona sensitivity on booking-flow compliance (~15-20 percentage points gap between tier-aware and pressure-test personas), demonstrating the model faithfully reproduces real-world difficulty distributions; (iv) seven rule-described trait correlations produce auditable co-occurrence patterns without requiring learned covariance matrices. The persona model is fully specified for reproduction.
cs.AI / 44 / 2609.08602
CLAMP: Constrained Decoding for Vision-Language Embodied Planning
Abstract
Embodied planning increasingly relies on vision-language models (VLMs) to translate instructions and visual observations into executable action sequences. However, fluent plans are not always executable. A VLM may refer to objects that are not visually observed, select actions whose required affordances are unavailable, or violate syntax and action constraints. We introduce CLAMP, a multimodal constraint-grounding framework that turns scene evidence into decoding-time constraints for a frozen VLM planner. CLAMP uses the initial observation to restrict object references to those supported by the scene, while a provided symbolic action model specifies state transitions and goals. During decoding, hard masks eliminate invalid next-token candidates, while a Hidden Markov Model (HMM)-based world-state lookahead module reweights the probabilities of the remaining feasible candidates based on action preconditions and goal reachability. This allows the planner to retain the VLM's language prior while preventing visually unsupported, unsafe, or infeasible candidates from entering the plan. For unseen tasks and environments, CLAMP adapts the HMM at test time using label-free continuations sampled from the frozen VLM. Experiments on VLABench, SafeAgentBench, and TaPA show that scene-grounded constraints improve object grounding and safety, while most remaining failures stem from perception errors or misaligned constraint specifications.
cs.AI / 45 / 2609.08719
GoAnt: Quality-Diversity Multi-Agent Search for Alpha Factor Discovery in Market Microstructure Data
Abstract
Automated alpha factor discovery searches symbolic trading signals from price-volume panels and order-book data under a fixed evaluation budget. Existing single- and multi-agent program-search systems can overfit predictive proxies that fail after execution costs and repeatedly explore redundant factor families, limiting execution robustness and behavioral diversity. We introduce GoAnt, a quality-diversity multi-agent search framework that combines non-communicating Explorer, Exploiter and Connector workers with a shared adaptive Mental Map and a compact Queen dispatcher. The Mental Map organizes candidates by leakage-free execution profiles and retains one elite per niche, while the Queen reallocates the evaluation budget from explicit search-state summaries. We also define a map-independent effective-yield protocol that counts high-quality, mutually nonredundant factors directly from each method's evaluation records, giving archive-based and map-free systems the same ruler. On real A-share microstructure data spanning 2023--2026, GoAnt reaches quality-weighted yields of 41.8 and 47.6 in price-volume and order-book settings, improving the strongest baseline by 57% and 97% under matched budgets. Its locked populations retain 0.64 and 0.67 of in-sample quality out of sample, compared with 0.61 and 0.63 for a static map.
cs.AI / 46 / 2609.08729
Application of curiosity driven exploration methods for hardware interference identification
Abstract
The transition from single-core to multi-core architectures in safety-critical embedded systems introduces significant challenges due to inter-core interference caused by contention for shared hardware resources. Such interference affects execution times and complicates the verification of strict temporal requirements, particularly in domains such as avionics where standards require comprehensive identification of interference sources. Existing interference analysis approaches, whether manual or model-based, struggle to capture the full range of behaviors arising from the complex interactions among micro-architectural components. In this paper, we frame multi-core interference analysis as the exploration of a complex system behavior space. We propose the use of curiosity-driven exploration algorithms from artificial intelligence to systematically and efficiently cover the space of possible interference behaviors. Using a simulator-based environment, we show that the proposed approach achieves broader and more uniform behavioral coverage within a limited experimental budget compared to traditional pseudo-random program generation methods.
cs.AI / 47 / 2609.08736
When Can One Obtain Certificates of Optimality Using Positivstellensaetze?
Abstract
We study certificates of positivity and optimality for learning problems whose objectives and constraints need not be polynomial. We isolate an axiomatic core of Fischer's constructive strict and weak Positivstellensätze and prove the resulting theorems for abstract function algebras over ordered fields. The framework separates two roles that can otherwise be conflated: objective and constraint functions may be built from broad classes of continuous or definable operations, while the auxiliary primitives used to construct a certificate satisfy explicit scalar and closure axioms. We give instances over continuous and definable function algebras, including ordered fields not closed under square roots, derive lower-bound and global-optimality certificates, and analyze both expanded term length and shared computation-graph complexity.
cs.AI / 48 / 2609.08861
API Benchmark Scores Do Not Reliably Transfer to Chatbot Interfaces
Abstract
Benchmark scores are a central currency in model releases: they inform purchasing decisions, shape public trust, and influence policy. Yet, a key assumption underlying benchmark scores is that the model performance measured through APIs faithfully reflects the behavior of deployed systems. We challenge this assumption by auditing ChatGPT, Claude, and Gemini across seven systems and nine benchmarks spanning general capability, social bias, and sycophancy. We find systematic API--interface differences in both accuracy and consistency. On average, API evaluations score 3.4 percentage points higher in accuracy and 2.1 percentage points higher in test--retest agreement than corresponding interface evaluations. For ChatGPT, the performance difference between API and interface access exceeds the API-only difference between GPT 5.3 and GPT 5.4. Put differently, switching access surfaces can degrade performance as much as downgrading a full model generation. We further test whether exposed API controls can reproduce interface behavior by varying system prompts, sampling parameters, and reasoning settings. These controls shift behavior in some cases but do not reliably eliminate the gap. Our findings document a context-validity gap: measurements obtained through APIs do not necessarily generalize to corresponding deployed interfaces, complicating the use of API evaluations as proxies for deployed systems.
cs.AI / 49 / 2609.08944
SkillAdam: Stable and Efficient Skill Evolution for Agents
Abstract
Agent skills provide a lightweight way to equip frozen language-model agents with domain knowledge and procedural guidance, yet obtaining high-quality skills remains costly and difficult to scale. Expert-written skills require substantial human effort. Recent skill self-evolution methods automate an iterative loop that uses execution feedback to revise skills, but their heuristic update strategies often yield unstable optimization and low iteration efficiency. We identify two challenges in realizing stable and efficient skill self-evolution. Direction Stability requires effective corrections to accumulate rather than be overwritten by iteration-local feedback. Update Adaptivity requires the scope of each revision to reflect the consistency of recent case-level improvements. We introduce SkillAdam, an Adam-inspired framework for optimizing discrete and non-differentiable skill documents. As a functional analogue of Adam's first moment, an optimization memory records identified problems and the outcomes of prior solution attempts to stabilize the update direction. As a functional analogue of Adam's second moment, a volatility-driven edit budget tracks the history-weighted variation of recent case-level improvements and adaptively controls the update magnitude. Across seven benchmarks that span short- and long-horizon tasks, SkillAdam achieves state-of-the-art performance with more stable optimization dynamics. It also obtains stronger skills with substantially fewer optimization iterations and lower cost than prior methods. Code repository: https://github.com/ruc-datalab/SkillAdam
cs.AI / 50 / 2609.08966
Good Pretraining, Bad SFT: Checkpoint Quality Across the Training Stack
Abstract
Language-model checkpoints are commonly selected by pretraining loss or benchmark scores, assuming that the highest-scoring checkpoint will remain the best starting point for subsequent training. We show that this assumption can fail in a full 30B mixture-of-experts training pipeline. The checkpoints that perform better after the full downstream training stack also have higher solution density, i.e., retain downstream performance under local weight perturbations.
cs.AI / 51 / 2609.09001
Deposon: An Auditable, Conservation-Guaranteed, Game-Theoretically Tested Scattering Layer over LLM Reasoning Paths
Abstract
Multi-step LLM reasoning lacks a machine-recheckable ledger: discarded reasoning paths leave no auditable record. We propose the Deposon scattering layer, which binds each node of an LLM-generated concept-decomposition graph to a two-parameter Deposon state; paths undergo three-channel scattering -- transmission, reflection, irreversible dissipation -- obeying T+R+A=1 for arbitrary parameters, with a maximum per-path energy-audit deviation of 2.2E-16 (machine epsilon). We report all three evidence tiers honestly. On synthetic trap benchmarks the path-filtering gain is closed (pre-registered): unified reaches 100% versus a decoy-capture baseline at 7%/10%. On real benchmarks the layer is indistinguishable from a trivial six-keyword rule filter (GSM8K 0.87 >= 0.85, McNemar p=0.5; StrategyQA 0.899 = 0.899); no difference is detected here, so we sharpen the claim to "the differential value lies solely in machine verifiability." Fusion yields a second negative result: convex combinations with a semantic prior never improve (physics 0.484 -> 0.452), and the apparent lambda=2 gain is an anti-field artifact; any fusion gain must be nonlinear. Modeling the reverse dynamics as a potential game on the graph, we evidence an auditable scalar's monotonicity and near-gradientness and quantify the empirical coordination ratio (ECR). The three formalized dynamical-equivalence propositions (P1a/P1b/T-P1c) are falsified under the pre-registered kill protocol, and the potential-game claim is downgraded to approximate (cyclic-graph median residual 0.669): only consistency-level evidence survives at the dynamical level. Code: github.com/zeroandcat/Deposon.
cs.AI / 52 / 2609.09030
Answer-Distribution Trajectories: A Stochastic-Dynamics View of LLM Reasoning
Abstract
Chain-of-thought reasoning provides a structured computation between a model's input and final answer. Yet it is often evaluated through endpoint accuracy, which ignores the path taken to reach that answer. An emerging line of work addresses this limitation using entropy profiles, which track how uncertainty evolves over the reasoning process but do not reveal which competing hypotheses account for that uncertainty. We introduce answer-distribution trajectories, a stochastic-dynamics-inspired representation that tracks the model's full predictive distribution over answers as reasoning unfolds. As a strictly finer representation than endpoint and entropy summaries, answer-distribution trajectories enable us to characterize a trace through a dynamical reasoning profile spanning exploration, revision, motion, and commitment, and to distinguish different dynamical mechanisms of reasoning success and failure. Across sixteen open-weight language models and four reasoning benchmarks, we show that traces with the same endpoint and similar entropy profiles can exhibit substantially different reasoning dynamics. We further find substantial variation in these dynamics both within and across models and tasks, with different objectives favoring different dynamical profiles. Additionally, we show that training and inference choices systematically reshape these profiles. Our results suggest that answer-distribution trajectories provide a rich framework for analysing and evaluating the dynamics of LLM reasoning.
cs.AI / 53 / 2609.09056
Time-Varying Data as Sheaves: an Invitation to Narratives
Abstract
Modern science and engineering increasingly rely on time-varying data, yet the mathematical tools used to model temporal phenomena are often developed within separate disciplines, obscuring common principles and limiting the transfer of ideas across fields. This chapter presents the theory of narratives, an abstract framework for time-varying objects of any mathematical kind that supports both theoretical investigations and applications. To illustrate this perspective, the chapter develops three vignettes, each illustrating a different research direction. The first addresses a general concern: What information loss can occur when switching between different representations of temporal data? The second concerns structural and algorithmic approaches: How can we systematically decompose time-varying data into simple pieces and obtain invariants describing its structural complexity? The third is an application to control theory: How can we model multi-agent systems with switching communication topologies? More important than any individual vignette, the central message of this invitation is that a suitable abstract perspective can organize and guide research across remarkably diverse mathematical and scientific domains.
cs.AI / 54 / 2609.09081
Everything in Moderation: Per-Domain Coverage Optima and Alignment-Resistant Domain Gaps in Multi-Domain Mid-Training
Abstract
Mid-training, the stage between pre-training and alignment, is where a model's per-domain data composition is typically set by data availability rather than principled design. We ask what that decision buys, and whether a later alignment pass can undo it. In a controlled logical-reasoning setting (Qwen3-8B-Base, with a 4B replication; five semantically rule-disjoint KOR-Bench domains) we train 30 allocations spanning the five-domain simplex, 24 sweep configurations plus six withheld from the fit, at five seeds each. Three findings emerge. First, every domain has an interior coverage optimum: the moderate band ($10\%$-$40\%$) is best for all five domains, and a calibrated permutation test for quadratic interiority gives $P\approx0.010$; the fitted mid-training-only curves, with 8B peaks between $9.9\%$ and $35.1\%$, reproduce for curve shape but not peak location. Second, the gaps survive a fixed-budget alignment pass: compensatory SFT raises 116/120 cells (mean $+4.32\%$) yet bridges $0/240$ pairs at a $5\%$ threshold and $30/240$ at a $10\%$ ratio, an equal-budget uniform control behaves almost identically, and a permutation null would bridge $13.8\pm3.3$ and $77.9\pm8.5$ pairs ($P<0.001$). Third, zero coverage collapses mid-training-only accuracy, though a FineWeb-Edu-only control shows the collapse is commingled with generic drift. An exploratory $θ^*$ allocation attains the largest full-pipeline gain ($+4.36\%$ vs. $+0.80\%$/$+0.64\%$\,pp) but is marginal under Welch test.
cs.AI / 55 / 2609.09094
The Surprising Effectiveness of Approximate Value Iteration in Self-Play
Abstract
Combining search with function approximation has driven major advances in game-playing programs, making self-play algorithms more competitive than ever. Still, the computational overhead of the most popular methods, based on Monte Carlo Tree Search (MCTS), can be substantial. In this work, we investigate whether simpler methods remain competitive in non-trivial, moderately sized games such as Connect Four, Hex(7x7) and synthetic games. We train a minimal self-play implementation of Approximate Value Iteration (AVI) and use ground-truth oracles for exact evaluation. Contrary to expectations, our results demonstrate the surprising effectiveness of AVI: it learns more accurate value functions than those learned by AlphaZero, while its one-step-lookahead greedy policies remain competitive with MCTS-based policies at substantially lower training and inference costs. Preliminary experiments on Othello and Go(9x9) show that AVI trains stably on larger games and learns effective value functions. These findings suggest that the success of MCTS-based methods may have eclipsed simpler approaches that have become increasingly practical with modern deep-learning tools.
cs.AI / 56 / 2609.09113
SAEScientist-Bench: Can AI Agents Conduct Autonomous SAE Interpretability Research?
Abstract
While research on recursive self-improvement (RSI) has predominantly automated model training pipelines, reliable autonomous development demands a missing pillar: post-hoc monitoring and auditing to understand what models learn and ensure safe alignment. Mechanistic interpretability tools are essential to bridge this gap, among which Sparse Autoencoders (SAEs) serve as a cornerstone by isolating interpretable features for model inspection and steering. In this paper, we introduce SAEScientist-Bench to evaluate whether AI agents can act as scientists utilizing SAE tools for autonomous mechanistic discovery. Given a target concept, an agent designs contrastive probes and navigates a Gemma Scope dictionary of 131K+ features in Gemma-2-9B-IT to discover the optimal feature, evaluated against curated expert reference features anchored on Neuronpedia across activation rank, concept selectivity on contrastive texts, and causal steering. Across 10 agent configurations and 20 tasks, frontier agents demonstrate genuine discovery capabilities and lead different evaluation dimensions, but remain well behind the expert baseline, approaching expert levels on separating target concepts from contrastive controls while lagging substantially in causal generation steering. Further analysis reveals that although agents can design contrasts to rule out spurious candidates, they frequently misinterpret experimental measurements. These results establish experimental model understanding as a measurable capability for closed-loop autonomous AI R&D. Our code is available at https://github.com/Trae1ounG/SAEScientist.
cs.AI / 57 / 2609.09126
A Generalization of Amari's Bayesian Duality
Abstract
Amari's contributions to information geometry and machine learning are well known. Here, we revisit Amari's work on Bayesian duality which has not received as much attention. We connect Amari's Bayesian duality to a convex duality of Bayes' rule. Using this connection, we present a generalization of Amari's Bayesian duality and discuss its relevance for modern artificial intelligence.
cs.AI / 58 / 2609.09133
ExecCritic: Learn to Test, Test to Improve for Coding Agents
Abstract
Execution feedback can guide coding agents toward correct repository repairs, but only when the tests capture the behavior requested by the issue. Agent-generated tests can encode incomplete or incorrect behavioral targets; when the same trajectory writes both the patch and the test, their errors can agree and create false confidence. We introduce ExecCritic, combining a test--verify--revise scaffold with a role-specific reinforcement learning recipe for training agents within it. The scaffold separates test construction from source-code repair: a Test agent independently generates repository-native tests, a fail-closed harness qualifies and freezes them, and a Repair agent revises source code from their execution feedback without changing the tests. Both roles use Qwen-3.5-35B-A3B as the backbone and are trained separately. In Learn to Test, the Test agent learns to produce behaviorally valid tests that distinguish correct from incorrect patches. In Test to Improve, the Repair agent learns both direct task resolution and feedback-guided revision. On SWE-bench Verified, test quality determines whether feedback helps: holding the base Repair agent fixed, tests from the base Test agent reduce resolved rate from a no-test baseline of 61.2% to 57.3%, whereas tests from GPT-5.6-sol raise it to 65.3%. Role-specific post-training raises the Qwen Test agent's Base-to-Gold success from 22.2% to 62.2%; composing the two post-trained Qwen agents reaches 72.6%, an 11.4-point gain over the original no-test baseline without stronger-model or Oracle feedback at evaluation time. Code is publicly available at https://github.com/MSR-Orchard/execcritic.
cs.AI / 59 / 2609.09134
Co-Evolving Harnesses and Models: On-Policy Correction Helps Weaker Models Catch Up Where Imitation Fails
Abstract
Agent harnesses (the system prompt, tool set, execution hooks, and context-management scaffolding around a model) are a critical determinant of agentic task success. Automated harness evolution can enable smaller models to perform well on domain-specific tasks at a fraction of frontier-model cost. Since both the harness and model weights shape behavior, we ask how harness evolution and lightweight fine-tuning should be combined. Across seven enterprise agent tasks, we first evolve a harness with the weaker model, then find that a stronger expert often uses it more effectively, suggesting expert supervision could close the remaining gap. However, training the weaker model on the expert's complete trajectories under the evolved harness backfires: performance regresses on all seven tasks by 4 to 30 points across Qwen3-Coder and Gemma 4, even though the same procedure helps under the unevolved harness. Our analysis shows that imitation transfers knowledge and increases scaffold usage, but disrupts model-harness fit: the weaker model adopts the expert's planning strategy without the competence to execute it and no longer matches the harness evolved around its native planning style. We therefore develop an on-policy expert-correction pipeline, automated by a meta-level MLE agent, that localizes the failing turn in the weaker model's own rollout and asks the expert to rewrite only that turn. This preserves the model's planning style and combines the gains of harness evolution and model adaptation. Our results identify and resolve a source of contention between harness and weight updates, yielding a compatibility-preserving recipe for economical co-evolution on domain-specific enterprise tasks.
cs.AI / 60 / 2609.09137
A Data-Driven Framework for Identifying and Prioritizing RPA Opportunities in Healthcare Processes
Abstract
Robotic Process Automation (RPA) is widely used to reduce administrative burden in United States hospitals, yet an estimated 30-50% of RPA initiatives underperform because processes are selected informally, without a repeatable method to catalogue candidates, prioritize them, match each to an automation tier -- a Python bot, an open-source orchestrator such as n8n, or an enterprise platform such as UiPath -- and forecast financial return before committing resources. We propose a four-module, data-driven framework unifying these decisions: a Process Taxonomy of twenty recurring hospital processes across five value streams; a Prioritization module deriving an Automation Suitability Index from an Analytic Hierarchy Process matrix with an explicit consistency check; a Tool-Tier Selection module recommending the least-cost technology sufficient for a process complexity, integration, and compliance profile; and a Return-on-Investment module quantifying labor savings, error-cost avoidance, payback, and net present value. Applied to a synthetic portfolio spanning all twenty processes, plus a reference data-flow architecture linking it to hospital EHR/payer/ERP systems: 12 of 20 clear the prioritization threshold; the ranking is robust to +/-20% weight perturbation (Spearman correlation 0.83, top-5 set preserved 97.7%, 2,000 Monte Carlo trials); an Automation Risk Index flags four qualifying processes as Critical risk; a budget-constrained portfolio optimization shows diminishing marginal NPV as spend scales from $400K to $1.03M; and a second Monte Carlo analysis shows portfolio NPV stays positive at its 5th percentile. The framework is a conceptual synthesis of the literature rather than an instrument calibrated on primary hospital data; we discuss HIPAA governance and a research agenda for empirical validation. A supplementary Python implementation accompanies the paper.
cs.AI / 61 / 2609.07937
TDDN: Text-aligned Diffused DINO Network for Puzzle Understanding
Abstract
Structured visual reasoning, such as image puzzles, demands fine-grained visual perception, an ability current Vision Language Models (VLMs) lack. VLMs built on CLIP-based ViT backbones trade fine-grained detail for high-level semantics, and we show this loss propagates downstream. To recover it, we fuse DINOv3 and CleanDIFT representations into a perception encoder (DiffusedDINO) and align it with RoBERTa-L, yielding a text-aligned model TDDN that preserves this perceptual advantage: with frozen backbones and only $\sim$590K alignment pairs, TDDN matches CLIP on image-text retrieval, surpassing it on three of four settings. It does so while more than tripling CLIP's dense-prediction accuracy (ADE20K 5.20 $\to$ 18.11 mIoU, COCO-Stuff 7.35 $\to$ 24.44), despite CLIP's massive training corpus. TDDN leads on segmentation benchmarks among general-purpose contrastive encoders, including SigLIP$\,$2. We further introduce Puzzle Perception, a segmentation and visual question answering dataset that probes fine-grained spatial understanding, on which TDDN doubles CLIP's segmentation accuracy (11.04 $\to$ 22.51 mIoU).
cs.AI / 62 / 2609.08038
SAFER-Activities: A Dataset for Smart Assessment of Fall Events and Routine Activities
Abstract
Smart healthcare monitoring systems require precise action recognition to ensure well-being and timely intervention in critical situations such as falls, particularly for mobility-challenged individuals. Existing datasets are often clip-based, lacking the frame-level detail needed to recognize actions online, as they unfold. To address this, we introduce SAFER-Activities, a dataset for fall detection and physical activity monitoring, with a dedicated subset for wheelchair use scenarios. It comprises over 66 hours of video data captured by multiple cameras, with 85,310 action instances and frame-level annotations for 30 action classes. We benchmark action recognition on SAFER-Activities with 2D and 3D skeleton models, RGB models with frozen backbones, and multimodal fusion strategies, and evaluate on in-lab, out-of-distribution, and cross-dataset test sets. Skeleton-based models generalize best under domain shift; fusing frozen RGB features with the skeleton stream improves in-domain recognition over the baseline CNN1D, most clearly on the wheelchair subset, but degrades out of distribution. Cross-dataset and qualitative evaluations confirm that models trained on SAFER-Activities transfer well to unseen environments and external fall data. To support research on robust fall detection and activity monitoring, we release the dataset and code at https://safer-activities.github.io/.
cs.AI / 63 / 2609.08182
WSPolypNet: Weakly Supervised Polyp Localization in Colonoscopy Videos
Abstract
Because dense frame-level annotation of colonoscopy videos is costly, we propose WSPolypNet, a weakly supervised framework for polyp localization using only video-level labels. WSPolypNet employs a 3D convolutional neural network trained with video-level supervision to generate class activation maps (CAMs), which identify candidate polyp regions without requiring frame-level spatial annotations. The CAM-derived localization cues are further enhanced using a multi-view strategy and provided to MedSAM2 as point prompts. MedSAM2 then propagates segmentation masks across the video, refining the coarse localization cues according to polyp boundaries. WSPolypNet achieved CorLoc scores of 47.80%, 43.68%, and 35.01% at IoU thresholds of 0.3, 0.5, and 0.7, respectively, compared with 36.87%, 33.72%, and 27.94% in the single-view setting. For small polyps, the multi-view strategy improved CorLoc@0.5 from 16.01% to 30.97%. The framework also achieved a recall of 94.51%. These results demonstrate the potential of weakly supervised spatiotemporal learning to substantially reduce spatial annotation requirements for polyp localization in colonoscopy videos.
cs.AI / 64 / 2609.08242
CS-CLIP: Compositional Scene Graph-guided CLIP for Robust Compositional Reasoning
Abstract
Vision-language models (VLMs) demonstrate strong performance across compositional reasoning benchmarks, which require reasoning over semantic perturbations of objects, attributes, relations, and their interactions. However, our controlled analysis reveals that existing compositionality-aware VLMs exhibit element-specific biases, often underperforming vanilla CLIP on certain compositional elements. To address this, we propose Compositional Scene Graph-guided CLIP (CS-CLIP), which uses scene graphs to identify compositional elements and construct structured negatives via selective masking. We further retain negatives that are most contradictory to the original caption, forcing the model to rely on compositional structure rather than surface cues. CS-CLIP achieves state-of-the-art compositional reasoning with robust performance across compositional elements. It also preserves general vision-language capabilities such as cross-modal retrieval and downstream visual reasoning, while requiring fewer training samples than prior methods.
cs.AI / 65 / 2609.08305
FPicker: Topology-Guided Evolution for Filament Tracing in Low-SNR Microscopy
Abstract
Automating filament tracing in Cryo-Electron Microscopy (Cryo-EM) is essential for 3D helical reconstruction but challenged by intersecting topologies and extremely low Signal-to-Noise Ratios ($\text{SNR} = σ_s^2/σ_n^2$ < 0.1 or -10 dB). Existing paradigms fail: pixel-wise segmenters suffer from severe topological fracturing, box-based detectors face ghost center drift, sequential trackers derail due to error accumulation, and traditional active contours collapse under artificial closed-curve constraints. To resolve these bottlenecks, we present FPicker, the first topology-guided framework reconciling these incompatibilities. It unifies perception via a center-endpoint representation and an open-curve evolution module to explicitly model non-cyclic connectivity. On simulated benchmarks, FPicker outperforms top baselines by over $40\%$ relative gain in mean spatio-angular precision (mSAP) and reduces topological gap rates by over $60\%$ under extreme noise ($-20\text{ dB}$). By learning intrinsic physical geometry rather than local texture, FPicker demonstrates strong potential as a resilient geometric backbone. Its zero-shot performance on the real-world EMPIAR dataset exhibits robust topological resistance, achieving a state-of-the-art 82.9\% mSAP upon fine-tuning. Our results also suggest modeling physical priors is a highly robust path toward bridging the sim-to-real gap in signal-starved scientific imaging. The code is publicly available at: https://github.com/tomzhaosky/FPicker.
cs.AI / 66 / 2609.08346
Segment Any Motion with Radar: Robust Multimodal Moving-Object Segmentation and Tracking
Abstract
Moving-object perception must decide which image regions correspond to real motion and keep every instance identified over time. Methods that read motion from appearance, optical flow, or estimated trajectories lose that evidence under poor illumination, adverse weather, reflections, and occlusion. Radar is a natural remedy because it measures radial velocity directly instead of inferring it from photometric correspondence. However, existing benchmarks do not jointly provide radar measurements, dense moving-instance masks, and temporally consistent identities for surveillance. We therefore introduce RGBTR-Motion, a synchronized and calibrated fixed-camera benchmark that pairs RGB, thermal, and radar streams with dense instance masks and temporally consistent identities across diverse surveillance scenes. We also develop SAM-Radar, an RGB, thermal, and radar-based segmentation and tracking framework built on SAM 3. SAM-Radar's radar-aware detector fuses calibrated RGBT features with radar returns that are grounded at their projected image locations, and motion supervision, implemented as foreground classification of those projected returns, teaches the detector to reject clutter without any text prompt. The tracker associates accepted radar returns with individual trajectories and uses them as physical evidence that a visually degraded target remains present. This allows it to bridge short periods of low visibility or occlusion and reconnect a reappearing target to its existing identity instead of starting a new track. SAM-Radar attains 0.7027 IoU and 0.8090 F1-50, and raises MOTA, HOTA, and IDF1 by 0.2977, 0.1603, and 0.2857 over the strongest competing values.
cs.AI / 67 / 2609.08402
Towards Embodied Air-Ground Cooperative Object Search: Benchmark, Dataset and Agentic Method
Abstract
Air-Ground Object Search (AGOS) in urban environments is a challenging embodied task, which requires an Unmanned Aerial Vehicle (UAV) and an Unmanned Ground Vehicle (UGV) to jointly search for and verify a specified target vehicle from multi-view visual references. To study this underexplored problem, we introduce AGOS-Bench, the first dedicated benchmark for evaluating whether general-purpose Vision-Language Models (VLMs) can integrate aerial discoveries and ground-level verification through UAV-UGV cooperation. We further provide AGOS-Dataset as the companion resource of exemplary trajectories constructed by an automatic pipeline. It consists of 7.7k episodes for searching objects of diverse categories and attributes, spanning three difficulty levels. To address the AGOS task, we propose AGOS-Agent, a training-free and tool-augmented approach. The agentic method relieves VLMs from complex and dynamic coordination via a deliberate search-handoff-verify cooperation protocol, only demanding VLMs for scene understanding and decision-making. Extensive experiments on nine VLMs show that AGOS-Agent improves overall success rate for eight of the nine evaluated backbones while reducing decision steps for all nine. On the hard split, the SR and SPL of Gemini-3.6-Flash increase from 8.6% to 55.7% and from 7.6% to 44.0%, respectively.
cs.AI / 68 / 2609.08442
AirAnchor: Bridging Local and Global Spatial Information for Zero-Shot Aerial Vision-and-Language Navigation
Abstract
Aerial Vision-and-Language Navigation requires drones to follow natural-language instructions and navigate through complex urban environments. Accurate navigation relies on both local and global spatial information, which support immediate action grounding and long-horizon path planning, respectively. However, existing zero-shot methods typically operate at a single spatial scale, relying either on local representations constructed online from current observations or on global memories built offline from historical experience. To address this limitation, we propose AirAnchor, a new paradigm that bridges local and global spatial information through spatial anchors and integrates both into a shared navigation framework, enabling comprehensive spatial grounding for decision-making. AirAnchor consists of three core components: (1) Query-Driven Spatial Anchor Grounding, which identifies decision-relevant anchors from visual observations and organizes them into local spatial representations; (2) Persistent Object Spatial Memory, which incrementally maintains an object knowledge base as persistent global spatial memory and retrieves landmark-related spatial priors; and (3) a Spatially-Informed Navigation Agent, which explicitly integrates both local and global spatial information into an agentic framework for decision-making. Extensive experiments on AerialVLN demonstrate that AirAnchor substantially outperforms existing zero-shot baselines, validating the effectiveness and efficiency of the proposed paradigm.
cs.AI / 69 / 2609.08627
SynthRCT: Scalable Conditional Deformation Synthesis for Synthetic Repeat CT Generation
Abstract
In proton therapy, plans are typically optimized on a single planning CT, making robustness evaluation essential under anatomical changes. However, current scenarios often rely on simplified perturbations that poorly capture complex, patient-specific variability. We propose SynthRCT, a scalable conditional generative framework for 3D anatomical deformation synthesis. Based on a conditional variational autoencoder, SynthRCT learns a latent deformation space and decodes sampled latent codes into local stationary velocity fields conditioned on an input anatomy. Local fields are assembled into coherent full-volume transformations, enabling memory-scalable generation for large field-of-view CT data. We validate the approach on respiratory 4DCT data with multiple breathing-phase anatomies per subject. SynthRCT enables patient-specific sampling of plausible anatomical transformations beyond predefined robustness scenarios. Code available at: https://github.com/TomasGuija/SynthRCT.
cs.AI / 70 / 2609.08636
From Where to How: Continuous 4D Interaction Forecasting from Egocentric Video
Abstract
Egocentric 4D interaction forecasting aims to anticipate both where future interactions will occur in 3D and how the human body will move to realize them, providing an important capability for assistive robotics and human-computer interaction. Existing methods struggle to translate semantic understanding into precise continuous 3D localization and to balance motion diversity with structural consistency in pose forecasting. More fundamentally, these tasks are often modeled separately, leaving the continuous geometric and temporal correspondence between interaction locations and body motion insufficiently captured. To address these challenges, we introduce Coherent4D, a large-scale egocentric dataset for continuous 4D interaction forecasting, comprising approximately 233K samples across three domains. Each sample pairs a sequence of future 3D interaction locations with corresponding full-body poses, aligned in time and expressed in a shared coordinate system. We also provide evaluation metrics in continuous space. Building on this formulation, we propose HIGFlow, a Hand Interaction Guided Residual Flow framework that models forecasting as a cascaded where-to-how process. HIGFlow first forecasts continuous future interaction locations by combining semantic grounding with short-horizon visual dynamics, and then uses the predicted location sequence to condition a deterministic motion anchor and residual Flow Matching for diverse yet structurally consistent full-body motion forecasting. Extensive experiments across all three domains demonstrate consistent improvements over representative baselines on both location and pose forecasting, while ablations validate the contributions of the proposed components. The project page is available at https://corrineqiu.github.io/from-where-to-how/.
cs.AI / 71 / 2609.08659
TriCCOT: Tri-part Convolutional Conformal Transformer for Onboard Space Object Detection
Abstract
Onboard object detection in Earth observation is constrained by limited computational resources and the absence of fully corrected imagery. While convolutional detectors are hardware-efficient, they often struggle to extract robust representations from raw and noisy data. Conversely, transformer-based models provide stronger global reasoning capabilities but remain difficult to deploy on FPGA accelerators due to quadratic attention complexity and non-compatible operations. We introduce TriCCOT, a tri-part architecture for robust and deployable onboard object detection. TriCCOT combines a convolutional region proposal network, a conformal prediction stage, and Aper-GATES, our hardware-friendly attention-based classifier. The region proposal network generates candidate bounding boxes, which are subsequently enlarged via conformal prediction, providing a distribution-free probabilistic coverage guarantee. The resulting crops are processed by Aper-GATES, which reformulates self-attention through convolutional projections, global channel statistics, and hardware-friendly gating operations, avoiding standard transformer operations that are poorly suited to CNN-oriented accelerators. Experiments on the DIOR and VDVRaw datasets demonstrate competitive detection performance and improved robustness to spatial blur and signal-dependent noise when compared to FPGA-compatible architectures. Finally, we report full deployment on a Xilinx Versal VCK190 FPGA without modifying the underlying DPU architecture, enabling unified CNN-Transformer inference for spaceborne embedded applications.
cs.AI / 72 / 2609.08686
CausalChapter: Improving Long-Video Chaptering with Interventional Dependency Modeling
Abstract
Long-form instructional videos require automatic chaptering to support browsing, navigation, and knowledge access. Recent long-context language models can perform chaptering from textualized video inputs, but they remain costly and brittle for content-dense lecture videos with long transcripts, smooth topic transitions, and detailed chapter outputs. A scalable segment-then-caption paradigm reduces this cost, but introduces two new challenges: boundary error propagation and fragmented cross-chapter context. We propose \textbf{CausalChapter}, an intervention-inspired framework for long-video chaptering that estimates prediction-level influence through lightweight masking and removal interventions. For boundary localization, our Local Dependency Shift module detects drops in predictive dependency between adjacent temporal windows; for chapter description generation, our Cross-Segment Support Selection module reranks historical contexts according to their support for the current prediction. Experiments on long-video chaptering benchmarks show that CausalChapter improves boundary localization, chapter description quality, and cross-chapter coherence.
cs.AI / 73 / 2609.08755
Kairos: A Dataset for Fine-Grained Video-Language Modeling over Space, Time, and Dynamics
Abstract
Many emerging video language modeling tasks require systems to move beyond clip-level abstraction and model visual content as it unfolds over extended time horizons. However, most existing video datasets rely on coarse or sparsely aligned supervision, which compresses temporal variation and limits the ability of models to learn reusable representations of continuous visual dynamics. We introduce Kairos, a video dataset for video-language modeling with time-resolved annotations. Kairos consists of long-duration videos, ranging from ten minutes to half an hour, annotated with fine-grained temporal alignment. The annotations capture ongoing actions, entity appearances and attributes, interactions, and evolving contextual cues along the video timeline. This time-resolved structure supports fine-grained evaluation, long-range modeling and reasoning, instruction data construction, representation learning, and video generation. Kairos provides a general-purpose foundation for modeling visual experiences over time.
cs.AI / 74 / 2609.08796
Hi-FLoop: Hierarchical State-Feedback Loops for Multi-Timescale World Modeling
Abstract
Multi-agent traffic simulation seeks diverse, coordinated, and physically realistic futures from maps and observed history. Long-horizon closed-loop generation must reconcile multiple decision time scales while its context evolves with generated states. Existing methods often unfold long futures from the initial scene and resolve intent, interaction, and motion monolithically, weakening cross-scale consistency and adaptation. We present HI-FLOOP, a branch-consistent multi-timescale state-feedback framework. Eight scene-level Worlds represent joint hypotheses, and all agents share the selected World identity throughout an 8-second rollout. Within the branch, an 8-second Goal anchors intent, a 2-second Preview coordinates interactions, and 1-second Control produces physical motion. Every 0.5-second commit feeds back only its executed prefix as new facts, while unexecuted hypotheses never enter factual memory. Joint Preview Interaction (JPI) induces a sparse directed future graph from Preview and uses conflict probabilities and signed arrival-time differences to gate interaction refinement. For generated-state recovery, a prefix-frozen A-to-B cascade lets frozen Model A generate 0-1 seconds, then transfers typed physical state, admissible context, and the branch index, but no latent state, to an independent Model B for re-encoding and 1-2-second recovery. On the full H-D public-validation split of 955 scenarios, one complete S1 run yields an Overall score of 0.689987 with the official evaluator. Under agent-centric oracle evaluation, HI-FLOOP achieves oracle-minADE@8 of 1.196636 m over the 8-second horizon and 0.526 m over the 6-second horizon.
cs.AI / 75 / 2609.09082
GoDeep: Annotation-Free Open-Vocabulary 3D Scene Understanding via Language-Space Lifting
Abstract
Open vocabulary 3D semantic segmentation methods typically lift CLIP features into 3D. This embeds points in a joint vision-language space known to behave like a bag-of-words on compositional tasks. Furthermore, even annotation free variants often require a large 3D training corpus and a dedicated 3D encoder per domain. Instead we use a vision-language model purely as a translator. It produces structured, entity-level descriptions of each posed image. These descriptions are grounded, projected, and aggregated directly in a general-purpose, language-only embedding space, with no 3D training corpus or encoder required. On ScanNet++, our pipeline is competitive with strong annotation free baselines trained on ScanNet. On a 5-building cultural heritage benchmark, raw scores initially favor a CLIP-based variant, but a single systematic vocabulary correction reverses this ranking. An effect confirmed by a second, independent correction on a different class, indicating that language-space embeddings track physical content more faithfully. This fidelity extends to genuinely out-of-vocabulary (OOV) objects on ScanNet++ proving that language-space embeddings separate presence from absence objects far more sharply than CLIP-based embeddings do. GoDeep also localize these OOV objects within the scene, all without any 2D-3D annotation. Because every representation remains discrete text, predictions are also explainable at the point level. Finally, exploiting both a heuristic weighting, that favors precise over merely frequent observations and GoDeep's explainability property, we propose an aggregation strategy, as a proof of concept, that favors finer elements localization.
cs.AI / 76 / 2609.09124
Canonical Color as a Lens into Concept Decodability in Vision Encoders and VLMs
Abstract
Visual encoders construct a representation of the image input for Vision-Language models. How much conceptual, as opposed to immediately visible, information does this representation contain? We use canonical color as a controlled test case to ask whether vision encoders make canonical-color information linearly accessible, even when color is removed from the input image. We construct a dataset of objects with canonical colors, and probe vision encoders for both color and object identity using color and grayscale images. We find that canonical color remains decodable from grayscale images, and is tied to predicted object identity, indicating a conceptual link. Extending this analysis to full VLMs, we find that VLM post-training can have a surprisingly large effect on color decodability in the vision encoder. Overall, canonical color provides a usefully controllable lens for tracing object-level conceptual semantic information in vision encoders and VLMs.
cs.AI / 77 / 2609.08789
Silent Revision: Measuring Undisclosed Change in the Safety Frameworks of Frontier AI Developers
Abstract
Frontier AI developers publish safety frameworks that commit them to evidencing whether their models are dangerous. The European Union and California now treat these documents as instruments of accountability, and both already impose duties on their revision. Neither requires the revision to be legible, in the sense that a reader could learn from the developer's own account what changed. We introduce the silent revision rate, the share of material changes to a framework's commitments that the developer's published account does not identify, and we release the versioned, hash-pinned corpus needed to compute it. The corpus contains every public version of the safety frameworks of the twelve developers that have published one, together with each provider's changelog, redline or announcement. We trace 710 commitment instances across twelve consecutive version pairs, code them against a frozen codebook, and adjudicate 244 individually. Three findings follow. First, 67% of material changes (95% CI 62 to 72) are silent under a strict standard and 53% under a lenient one, falling to 49% at section granularity. Second, silence appears to track the form of the account, since narrative announcements run at 74% against 63% for itemised changelogs, whereas account length in words barely matters; on the test that respects nesting the difference is suggestive. Third, 77% of traced changes weaken or remove a commitment, and in seven of eight pairs weakenings are more often silent than strengthenings. The statutory remedy therefore exists and specifies the wrong artefact. A justification explains why a framework changed, an enumeration states what changed, and only the latter makes revision auditable. We argue that publication duties should carry an enumeration duty, which one provider already meets, voluntarily and incompletely.
cs.AI / 78 / 2609.08869
OntoKG-EQ: A provenance-grounded, competency-question-governed knowledge graph for auditable analyst querying
Abstract
Analysts in emerging equity markets keep answering the same questions. Did fundamentals match the market's response? How does the local currency co-move with returns? Which firms outperform sector and benchmark, and which disclosures coincide with abnormal trading? These answers come from ad-hoc spreadsheets that are hard to reproduce, audit, or trust. We present OntoKG-EQ, a knowledge-based system that makes such queries reproducible, evidence-linked, temporally explicit, valid, and inspectable. It couples a bounded, competency-question-governed core ontology with a provenance-aware knowledge graph in which every class, property, shape, and metric is justified by one of five frozen questions. The system materialises market data into the graph, computes the metrics, validates its structure against declarative shape constraints, answers each competency question with a graph query, derives typed findings, and generates an explanation tracing each result to its observations, evidence, sources, and provenance. We evaluate on curated datasets from three emerging markets (Pakistan, Malaysia, Indonesia). Once each market's data is mapped into the common schema, the ontology, shapes, queries, and rules are reused unchanged. A relational-database baseline shows the graph changes no analytics. Its value is governance, provenance, and self-explaining structure. Because answers are rendered deterministically from the validated graph, their consistency with it is guaranteed by construction. Used as a reference, the system measures how consistently eight open language models transcribe the same evidence (provenance coverage 0.00 to 1.00). A study with a 17-participant convenience panel finds the evidence bundle significantly increased perceived trust and completeness. Code and data are openly released.
cs.AI / 79 / 2609.08310
Exploring Bottom-Up Clustering for Creating Semantic IDs
Abstract
The success of generative retrieval has largely been attributed to the use of Semantic IDs, which improve over arbitrary item-level identifiers such as hashes by capturing the semantics of items. The main challenges faced when constructing Semantic IDs, however, is in mapping each identifier to a unique product and capturing information valuable to downstream tasks. Past works have appended additional codewords to de-duplicate item identifiers and utilized residual quantization to create hierarchical clusters. In this work, we present an algorithm for generating Semantic IDs that ensure the identifiers are both unique and preserve the structure of the original embedding. Key to our work is the use of bottom-up clustering to preserve local structure in the embedding space, improving the clustering quality of the resulting Semantic IDs and their utility for downstream generative retrieval.
cs.AI / 80 / 2609.08443
SequenceO1: End-to-End Ultra-Long (100K) Sequence Modeling in Recommendation with Low-Rank Caching
Abstract
Modeling long-term user behavior is central to sequential recommendation and billion-scale industrial recommender systems, yet production ranking models operate under strict latency, memory, communication, and training-throughput constraints. At the 100K scale, the challenge extends beyond attention complexity: raw sequence features must be stored, transferred, and repeatedly processed during training and online serving. Existing approaches based on history truncation, multi-stage behavior retrieval, compressed lifelong histories, or train-short/infer-long extrapolation either weaken end-to-end optimization or retain substantial length-dependent cost. We present SequenceO1, an end-to-end framework for ultra-long user behavior sequence modeling, deployed at full traffic on Douyin with histories of up to 100K interactions. SequenceO1 follows a compress-then-reason design. Its Sketch Attention (SA) uses learnable prototypes and prototype-wise normalization to compress the raw history into a fixed-size, target-agnostic user representation. Target-conditioned Stacked Target-to-History Cross Attention (STCA) then models complementary time scales: a recent 10K suffix for short-term interests and the compact sketch for long-term preferences. To make training and inference practical, SequenceO1 combines low-rank user representation caching, multi-request user-level batching, pipeline lift, and a fused FlashSA kernel to amortize feature storage, communication, and computation across targets, training instances, and consecutive requests. Production experiments show consistent offline and online gains, while the compact cached sketch retains most of the benefit of directly scaling end-to-end sequence ranking to 100K. These results provide a practical model-system approach to efficient attention, sequence compression, and scalable long-sequence and long-context recommendation systems.
cs.AI / 81 / 2609.08123
DISEIL: Demonstration Distillation for Sample-Efficient Imitation Learning
Abstract
A robot that can be taught a new task from a handful of demonstrations has to work out for itself what it still cannot do, and then ask for exactly that. Interactive imitation learning takes a step in that direction by letting a policy practice on its own and calling an expert when it goes wrong. Existing methods decide when to interrupt the learner. A further 2 decisions are left to whichever episode happened to trigger the interruption: which failure to correct, and where the demonstration should start. This paper is a first attempt at making both of them deliberately. DISEIL (Demonstration dIstillation for Sample-Efficient Imitation Learning) marks each failed episode at the step where the policy first becomes unreliable, represents that moment with a geometric descriptor, and groups the failures into recurring failure modes. A vision-language model and a language model read the selected mode and write a request for the next demonstration, and a store of task constraints checks that the request can be carried out before any expert time is spent. No model produces a robot action. Across 5 simulated tasks under state and image observations, changing only what the expert is asked for gives the highest mean held-out success rate in all 10 settings, with a tie in 1, and the margin is widest at the smallest budget we tested. The scope is narrow: a single round of practice at a time, in simulation, with experts that are mostly scripted. The longer-term aim is a learner that also tracks what its demonstration set already covers, and that asks a human teacher for the missing behavior in proportion to the effort each request costs them.
cs.AI / 82 / 2609.08164
Dual-Layer Semantic-Spatial Belief Mapping for Aerial Object Goal Navigation
Abstract
Aerial Object Goal Navigation (ObjectNav) requires an unmanned aerial vehicle (UAV) to locate a described target in an unknown outdoor environment using onboard visual observations. Vision-language models (VLMs) can interpret open-ended target descriptions and visual observations, but their frame-level outputs are often noisy, sparse, and spatially transient. We propose AeroBelief, a dual-layer semantic-spatial belief mapping framework that transforms transient VLM observations into persistent spatial guidance. It separates broad contextual plausibility from target-specific evidence: an intuition layer accumulates scene-level semantic cues for exploration, while an evidence layer preserves qualified target-specific observations for approach and confirmation. Evidence-gated fusion combines the two layers into spatial belief hotspots. We further introduce object-conditioned visual reasoning with conservative evidence qualification to improve observation reliability before spatial accumulation. In parallel, egocentric regional guidance converts quadtree coverage into UAV-centered, yaw-aligned directional proposals and stabilizes them through temporal commitment. Its regional scoring is independent of semantic belief values, maintaining exploration pressure and reducing repeated low-gain search. Experiments on the UAV-ON benchmark show that AeroBelief achieves the best reported overall SR, OSR, and SPL among the compared methods, reaching 21.61%, 35.57%, and 10.62, respectively. These results support the effectiveness of persistent semantic-spatial belief, conservative evidence qualification, and temporally stable regional guidance for aerial ObjectNav.
cs.AI / 83 / 2609.08224
3DWay: Generalizing Robot Manipulation via 3D Consistent Waypoints
Abstract
Intermediate representations are key to bridging the modality gap between generalizable manipulation policies and large-scale pretrained vision-language models (VLMs). Among these, trajectory-based representations compactly represent motion-relevant cues, yet most existing approaches predict trajectories in 2D image space, resulting in intrinsic 3D ambiguity. Moreover, using 2D trajectories with depth still leaves the free-space waypoints ambiguous, limiting reliable 3D reasoning. To address this, we propose predicting 3D consistent waypoints (3DWay) from multi-view images. By reformulating 3D waypoints prediction as generating multi-view consistent 2D waypoints followed by geometric triangulation, we enable explicit 3D motion specification while preserving the strong priors of pretrained VLMs. The predicted waypoints can guide existing VLA models for better generalization or be directly executed on simple tasks. Extensive experiments show that 3DWay substantially improves 3D spatial grounding and vision-language reasoning, demonstrating strong potential for generalizable robot manipulation. Codes will be released at https://github.com/ziqin-h/3DWay.
cs.AI / 84 / 2609.08250
CALIPER: Clean Scenes Cannot Rank Physical Inference in Pretrained Visual Representations
Abstract
How far a pushed object slides depends on its mass and friction, which no single image reveals. Pretrained visual encoders are increasingly used as the perception front end of world models for manipulation, and their physical competence is assessed with perturbation benchmarks and linear probes, almost always in a clean, fixed-camera scene. We show that these assessments cannot distinguish an encoder that infers physics from one that does not. CALIPER (calibrate, then predict) is a direct test: an object of unknown mass and friction is struck twice at known speeds, a third strike is shown only up to the moment of contact, and a linear readout on frozen features must predict how far the object slides. Swapping in another object's calibration clips checks that the evidence is actually used. Across 2,000 simulated episodes and eight representations, from V-JEPA 2 to a randomly initialised ViT and raw pixels, calibration adds +0.50 R^2 and the swap removes it. Yet in the clean scene every representation lands within 0.02 R^2 of the ceiling set by true simulator state, because a fixed camera exposes the object's displacement directly in pixel coordinates. Resampling camera, lighting, and clutter for every clip spreads the same representations across 0.50 R^2; when the readout chooses a push speed for a goal distance, V-JEPA 2 misses by 4 mm and the random ViT by 20 mm, no better than ignoring the object. Linear probes track none of this: a change in frame aggregation moves a probe more than pretraining does, and erasing the probed mass direction from the same representation costs nothing in one scene and 0.35 R^2 in the other. Whether a benchmark can rank models is an empirical property, and we give three checks that establish it.
cs.AI / 85 / 2609.08338
A Multi-Modal Perception Pipeline for Object Detection and Tracking in Autonomous Racing
Abstract
Object detection and tracking are fundamental components of perception systems for autonomous driving. Achieving robust performance under adverse conditions such as limited visibility, sensor noise, and failures remains an open challenge, particularly in autonomous racing, where vehicles operate at very high speeds, experience strong vibrations, and interact under small safety margins. This paper presents a multi-modal late-fusion perception pipeline for object detection and tracking in the autonomous racing domain. The proposed system extends previous work by exploiting all onboard sensors through a late-fusion approach and a dedicated multi-object tracking framework. Independent detections from cameras, LiDARs, and RADARs are combined to provide timely and robust state estimates of surrounding vehicles. The tracking method explicitly compensates for detection delays and embeds in its model prior knowledge of vehicle dynamics and track layout. Experimental evaluation on real-world data across diverse critical scenarios, representative of challenging edge cases also in urban driving, confirms the effectiveness of the proposed pipeline and its suitability to support safe and adaptive planning decisions.
cs.AI / 86 / 2609.08339
RoboCousin: Build Your Own Simulation Playground for Robust Bimanual Robotic Manipulation
Abstract
Bimanual manipulation policies require large and diverse training datasets, yet collecting demonstrations on physical robots is expensive and difficult to scale. Simulation can generate data efficiently, but existing pipelines typically operate within closed asset libraries and predefined scenes: adding a newly observed object or environment still requires substantial effort to reconstruct geometry, specify physical and semantic properties, annotate interactions, and integrate the result into executable tasks. We present RoboCousin, an extensible simulation-based data-generation platform that turns user-provided observations into reusable assets, scenes, and expert trajectories for bimanual manipulation. Built on RoboTwin~2.0, RoboCousin converts object images into simulation-ready assets with visual and collision geometry, semantic and physical metadata, and automatically generated grasp-contact candidates. It further constructs digital cousins that vary compatible objects, backgrounds, layouts, and language instructions while preserving task-relevant affordances and spatial relations. The same asset system supports tabletop and room-level scene construction, with collision-aware base control for interaction beyond a fixed workspace. We release RoboCousin-OBD, containing more than 3,000 annotated object instances and 50 background environments, and use RoboCousin to generate over one million expert trajectories across 50 tasks. Simulation and real-robot experiments show that the automatically generated interaction annotations are comparable to curated annotations, generated assets provide effective sim-to-real supervision, and tabletop cousins can improve transfer beyond training on a single reconstructed scene. RoboCousin therefore provides a practical path for expanding both the scale and coverage of synthetic bimanual manipulation data.
cs.AI / 87 / 2609.08638
CASD: Chunk-Aligned Semantic Distillation for Multi-StageRobot Manipulation
Abstract
An action chunk can span several stages of a manipulation task, yet a label for its first step describes only the current stage. We introduce Chunk-Aligned Semantic Distillation (CASD), which derives semantic targets for entire action chunks. An offline vision--language model segments demonstrations into described stages. Their occupancy within each action chunk determines a weighted semantic target, including transitions between stages. A CASD generator learns to predict this target from the current observation, robot state, and task instruction. We then freeze the generator and train a policy conditioned on its predictions. The semantic branch runs once per policy query, without online VLM calls or reasoning-trace decoding. Teacher matching on annotated LIBERO training episodes is above chance for both single-stage and boundary-crossing chunks. We evaluate three Fast-WAM variants and a DreamZero integration across four benchmarks, including distribution shifts on LIBERO-Plus. Compared with published references, IDM+CASD reaches 98.9\% versus 98.0\% average success on LIBERO, while Uncond falls below its reference. Joint+CASD reaches 93.0\% versus 90.6\% on RoboTwin 2.0, and DreamZero+CASD reaches a 47.9\% four-category MolmoSpaces manipulation average versus 40.7\%. Performance varies across backbone integrations.
cs.AI / 88 / 2609.08673
BIFTA: Brain-Inspired Few-Shot Tactile Adaptation for Unknown Sensors
Abstract
Advances in tactile sensing have made contact-rich perception possible, accelerating progress in robotic manipulation, material understanding, and embodied interaction. However, because optical design, elastomer mechanics, and imaging geometry differ substantially across tactile sensors, models trained on known sensor types can suffer an abrupt performance collapse on unknown sensors. To address this problem, we propose the Brain-Inspired Few-Shot Tactile Adaptation (BIFTA) framework; it draws on the brain's rapid sensory adaptation mechanism to adapt a frozen encoder to an unknown tactile sensor from a small labeled support set. BIFTA preserves pretrained representations through dual-view statistical memory, constructs support-conditioned spectral graphs to repair sensor-dependent feature neighborhoods, and applies uncertainty-gated recurrent propagation to strengthen reliable cross-query evidence. Extensive benchmarks across three tactile datasets show that BIFTA substantially improves adaptation to unknown sensors: with only 10\% labeled target data on SITR, it raises mean Sparsh accuracy from 6.86\% for the frozen source classifier to 87.09\%, exceeding the strongest implemented prior comparison by 47.22 percentage points, and these gains generalize across datasets, pretrained backbones, and tactile tasks. These results validate BIFTA for data-efficient adaptation to unknown tactile sensors and offer a promising route toward tactile models that transfer across heterogeneous hardware.
cs.AI / 89 / 2609.09119
DeCAL: Towards Physically-Grounded Dexterous Vision-Language-Action Models via Contact-Aware Latent Co-Imagination
Abstract
Dexterous manipulation involves contact-rich and fine-grained interactions with the physical world, posing significant challenges for existing vision-language-action (VLA) models due to severe visual occlusions and complex contact dynamics. While recent works have incorporated tactile sensing into robotic manipulation, most approaches still rely on homogeneous multimodal fusion, lacking adaptive tactile integration and explicit modeling of physical dynamics. In this work, we present DeCAL, a physically-grounded dexterous vision-language-action model that unifies understanding, imagination and action generation for contact-rich dexterous manipulation. Built upon a Mixture-of-Transformers (MoT) architecture, DeCAL leverages specialized experts for each capability while enabling efficient information flow among them. To effectively leverage tactile information, we introduce Adaptive Visuo-Tactile Fusion that dynamically regulates tactile interactions via a contact-aware gating strategy. Furthermore, we propose Visuo-Tactile Latent Co-Imagination to jointly model visual and tactile dynamics, equipping the policy with implicit physical world knowledge. Experimental results show that DeCAL consistently achieves state-of-the-art performance across all tasks, attaining a 71% average success rate and an 83.4% progress success rate, while also demonstrating strong generalization to unseen scenarios. The website is available at https://aureleopku.github.io/DeCAL.
cs.AI / 90 / 2609.09158
TANGO: Humanoid Navigation in Cluttered Environments with a Whole-Body Vision-Language-Action Model
Abstract
We study the problem of navigating cluttered indoor environments with a humanoid robot. Unlike conventional methods that model navigation as a 2D path planning problem, humanoid traversal in cluttered environments requires continuous geometry-aware whole-body adaptation, including coordinated arm placement, torso adjustment, and gait modulation for collision-free movement through complex 3D spaces. We introduce TANGO, the first whole-body vision-language navigation framework for language-conditioned humanoid traversal in cluttered environments. Given a natural-language instruction and egocentric RGB observations, TANGO directly predicts 29-DoF joint-space actions for downstream whole-body control. We train TANGO entirely in simulation by synthesizing diverse collision-free traversal behaviors via global path planning, kinematic whole-body motion generation, obstacle-aware motion editing, and RL-based tracking. This pipeline provides dynamically feasible action supervision for learning language-conditioned whole-body policies. In extensive simulation experiments, TANGO demonstrates state-of-the-art performance in vision-language navigation, while outperforming strong modular baselines in navigating challenging scenes requiring obstacle negotiation. Lastly, we deploy TANGO zero-shot on a Unitree G1 humanoid robot, and observe robust language-guided traversal in cluttered real-world scenes without training on any real-world navigation data.
cs.AI / 91 / 2609.08390
Noise Adaptive Streaming Audio-Visual Speech Token Enhancement for Robust Full-Duplex Spoken Dialogue Models
Abstract
Full-duplex spoken dialogue systems enable simultaneous listening and speaking, but their audio-only perception often fails under background noise and overlapping speech, leading to incoherent responses. Recent audio-visual dialogue approaches show that incorporating visual cues such as lip movements improve robustness under audio corruption. However, existing approaches often adapt the large speech dialogue model itself to process visual input, requiring costly multimodal training. We propose AV-STE, a modular streaming audio-visual front-end that restores corrupted semantic speech tokens from noisy audio and lip video before they reach the speech LLM. The downstream dialogue model remains entirely frozen, preserving its pretrained conversational capabilities. When integrated with frozen Moshi, AV-STE improves average GPT-4o-judged response coherence from 1.42 to 1.91 under same-dataset speaker interference while largely preserving turn-taking behavior. Gains also transfer to out-of-domain Seamless Interaction.
cs.AI / 92 / 2609.08672
X2Streaming-ASR: wait when uncertain, emit when ready for streaming ASR
Abstract
Streaming automatic speech recognition (ASR) for real-time voice agents and full-duplex dialogue must provide accurate partial transcripts with low commit latency. Existing systems commonly use a fixed chunk size, look-ahead, or target delay, or encourage emissions near estimated acoustic boundaries. These approaches do not directly optimize how much additional context to use at each output position under a single-pass, hard-commit constraint. We propose X2Streaming-ASR, which decomposes streaming recognition into when to commit and what to commit. Its three-stage training procedure first establishes streaming recognition ability, then warm-starts the commit policy with automatically probed trajectories, and finally refines the policy using character-level, segment-assigned group-relative rewards for recognition accuracy and latency. Across AISHELL-1/2/3 and WenetSpeech, X2Streaming-ASR achieves a mean character-level commit latency of 27-84 ms relative to forced-aligned character endpoints, compared with 409-585 ms for the evaluated streaming baselines. It achieves the best streaming CER among the evaluated systems on AISHELL-1 and AISHELL-3 with substantially lower latency.
cs.AI / 93 / 2609.08977
Omni Interaction Agent Technical Report
Abstract
In this work, we present Gander, an end-to-end model that unifies omni perception, realtime interaction, and agentic capabilities within a single framework. In contrast to turn-based conventional paradigms, Gander continuously receives streaming inputs across multiple modalities, including video, speech, and text, enabling natural full-duplex interaction in both everyday conversations and complex workflow-oriented agent scenarios. Users can interrupt the model at any time, while the model can also proactively provide intermediate feedback or ask follow up questions. To natively support these capabilities, Gander adopts two key architectural designs: 1) It employs a Cerebellum-Brain collaborative framework, in which the Cerebellum is responsible for realtime interaction and omni conversational capabilities, while the Brain handles complex reasoning and higher-level agentic tasks. The two components interact continuously through tool calling and the agent orchestration runtime. 2) The Cerebellum is built upon a streaming Thinker-Talker architecture, user inputs and model outputs are further flattened into an ordered token stream at the chunk level, providing a unified representation for low latency, continuous interaction. We conduct comprehensive evaluations of Gander across four dimensions: conversational ability, omni understanding, interactive capability, and agentic intelligence. Internal human evaluations demonstrate that Gander maintains the natural and expressive spoken dialogue capabilities of SOTA open source models while achieving competitive performance in omni interaction. Gander also demonstrates robustness in challenging real-world scenarios, including background noise interference, multi-party interactions, and backchannel communication. We release Gander together with its models, code, and data to facilitate further research and development in the community.
cs.AI / 94 / 2609.08606
Neptune: An AI model for Global Ocean Subseasonal Prediction
Abstract
Subseasonal-to-seasonal (S2S) forecasting is societally critical, supporting decision-making in sectors ranging from water and agricultural management to disaster risk reduction, energy planning, and insurance. Achieving reliable predictions at these timescales requires representing the ocean and its dynamics, but traditional physics-based Ocean General Circulation Models (OGCMs), are computationally expensive and difficult to develop and improve because of the code complexity. In this work, we propose Neptune, an end-to-end data-driven framework for global ocean and sea-ice components emulation tailored for S2S timescales, up to 60 days. Neptune combines Convolutional Neural Networks (CNNs) and Spherical Fourier Neural Operators (SFNOs) to effectively capture local features and global cross-scale interactions, thereby obtaining a coherent representation of the ocean state. Forced by prescribed daily atmospheric fields, Neptune emulates ocean state variables, from temperature and salinity, to zonal and meridional currents, from sea surface height to sea ice thickness and concentration, with daily outputs at the ocean surface and through the water column. Specifically, we propose two variants of Neptune, Neptune-1 and Neptune-025, capable of emulating the ocean state at 1° and 0.25° resolution, respectively. Evaluated against a suite of metrics, including statistics (RMSE, CRPS and ACC), physical coherency (Ocean Heat Content, Eddy Kinetic Energy and Ice Brier Score) and climate indices (ENSO and Z20 metric, IOD), Neptune successfully reproduces the spatio-temporal evolution of the oceanic fields up to 60 days, and is stable over long timescales. Neptune provides compelling evidence that end-to-end data-driven ocean emulators can become a powerful component of next-generation S2S forecasting systems, emulating ocean state at high spatio-temporal resolution.
cs.AI / 95 / 2609.08166
AI for AI: Optimizing Additional Infrastructure Build-out to Power Artificial Intelligence Data Centers
Abstract
The twenty-first century's transformative technology, artificial intelligence, is increasingly constrained by the twentieth century's transformative technology, the electricity grid. Rapid growth in electricity demand from data centers is leading to higher electricity prices, without a compensating supply-side response. We develop a framework linking data-center load growth, available generation capacity, and market-clearing prices to understand this phenomenon. We first analyze a deterministic model to show how differing estimates of demand and supply growth rates affect prices. We then model the expansion of new data centers and their associated electricity demand, together with build-outs of new electricity supply, as stochastic processes,resulting in probabilistic distributions of supply, demand, and prices rather than a single forecast. Finally, we formulate generation expansion as a stochastic control problem in which a revenue-maximizing investor dynamically chooses the intensity of supply-side investments. The analysis highlights a central challenge of the data-center build-out: even when rapid demand growth increases the need for new generation, the uncertainties related to load forecasts, development execution risks, and value cannibalization from overbuilding capacity may weaken incentives to invest at the pace required to keep electricity prices stable.
机器学习 (cs.LG)
101
cs.LG / 1 / 2609.09011
Closed-Form of the Local Galactic Potential and Stellar Distribution Function from Gaia DR3
Abstract
The local dark matter density determines the strength of the signal expected in direct-detection experiments, yet published estimates from stellar motions disagree by more than their errors, and the most recent machine-learning analysis of Gaia data finds a local density consistent with zero. According to Jeans' theorem, a distribution function built from integrals of motion satisfies the collisionless Boltzmann equation (CBE) trivially for any choice of potential, so a search that simultaneously fits the distribution function and the potential to the CBE identifies neither. Our pipeline instead estimates the distribution function in isolation, linearizing the equation in terms of accelerations and allowing for direct measurement of the local force field, and then fits closed forms to that field via symbolic regression. Throughout, we find that the usable information lies not in the CBE residual but in the stellar number counts, the observable most distorted by survey selection. Along the vertical profile, our recovered potential agrees with the classical self-gravitating isothermal disc.
cs.LG / 2 / 2609.08219
Speed Limit for Information Acquisition in Stochastic Learning Dynamics
Abstract
Neural networks acquire internal representations through learning. In this work, we formulate stochastic gradient descent (SGD) as a Markovian stochastic process and derive a Fisher-information flow speed limit that bounds the rate at which trainable parameters can acquire information about latent variables in the data-generating process. The resulting inequality decomposes the information flow into drift and noise contributions, thereby quantifying the roles of deterministic learning forces and SGD-induced fluctuations from an information-theoretic perspective. We verify the bound in analytically tractable basis-function linear regression, where the information budget predicted by the bound reproduces the ordering and characteristic time scales with which different latent variables are encoded in the learned parameters. These results establish Fisher-information speed limits as a quantitative framework for diagnosing when and how different aspects of the data-generating mechanism are acquired during stochastic learning.
cs.LG / 3 / 2609.07974
A Sub-4 Approximation for Fair $k$-Means
Abstract
Fairness in clustering has attracted sustained research interest, motivated by the need to ensure equitable representation of protected groups in machine learning applications. We study fair $k$-means clustering in Euclidean space, where the proportion of each protected group in every cluster must lie within specified lower and upper bounds. These constraints make it challenging to determine both cluster centers and point assignments. We propose an approximation algorithm that combines a linear programming relaxation with geometric transformations of the input to construct candidate center sets. Given a $ρ$-approximate algorithm for weighted $k$-means and any $ε>0$, our algorithm returns a fractional solution whose cost is at most $1+(3-1/Γ)ρ+O(ε)$ times the optimal integral fair cost, where $Γ\approx6.357$ is an upper bound on the integrality gap of the standard Euclidean $k$-means LP. With a PTAS as the subroutine, the approximation ratio becomes $3.8427+O(ε)$, improving the previous factor of $5+O(ε)$ to below $4$. The solution satisfies all fairness constraints exactly and can be rounded to an integral assignment with a bounded additive violation of fairness and no increase in cost. The same approximation guarantee extends to the $k$-sparse Wasserstein barycenter problem.
cs.LG / 4 / 2609.07915
JEDI: JEPA-to-Edge Distillation for Efficient Cropland Segmentation from Satellite Imagery
Abstract
Large vision models provide useful representations for remote-sensing segmentation but are often too expensive for deployment at the satellite or field edge. Existing feature-level distillation methods also tend to assume similar teacher and student architectures and often stop feature alignment when task training begins. We introduce JEDI (JEPA-to-Edge Distillation), a two-stage framework that transfers representations from a large I-JEPA Vision Transformer teacher to a compact SegFormer student. First, JEDI aligns the student's terminal representation with the teacher's token space using cross-architecture projection and spatial alignment. It then jointly optimizes supervised segmentation, temperature-scaled response distillation, and persistent feature alignment throughout task adaptation. On CalCROP21, JEDI-B0 achieves 68.0 mean Intersection-over-Union (mIoU) with 4.04M parameters, improving over the standalone student by 16.0 points and coming within 2.0 points of the 70.0 mIoU achieved by the 639M-parameter teacher. We evaluate SegFormer B0, B1, and B2 students with 4.04M, 14.33M, and 28M parameters, respectively. Across all three variants, JEDI consistently outperforms response-, structure-, channel-, and relational-distillation baselines under the same teacher-student setting. These results show that persistent representation alignment is especially valuable under aggressive compression, substantially reducing model size and computation while preserving segmentation performance.
cs.LG / 5 / 2609.08041
MamMA: A Mamba-Based Pedestrian Trajectory Prediction Algorithm Considering Occupancy Map and Pedestrian Awareness States
Abstract
Many pedestrian trajectory prediction algorithms have been proposed to improve the safety of navigation for mobile robots working in human-robot coexistence environments. Some pedestrian trajectory prediction algorithms extract information about obstacles near pedestrians from top-down view images to improve the accuracy of trajectory prediction. However, mobile robots typically create local occupancy maps using LiDAR, rather than top-down view images. Meanwhile, the vision sensors on board robots provide egocentric view images, which contain fine-grained behavioral information about the pedestrians near the robot. To better use the information collected by LiDAR and on-board vision sensors, we propose MamMA, a Mamba-based pedestrian trajectory prediction algorithm considering occupancy maps and pedestrian awareness states. MamMA divides the occupancy map by patches and extracts obstacle features from each patch to create map features. Pedestrian awareness states are divided and considered, as some studies show that awareness states affect the perception and speed of pedestrians. Furthermore, a Mamba-based model is proposed to predict the future trajectories of pedestrians based on different types of features. Experiments on the STCrowd, SiT, JRDB, ETH, and UCY datasets show that MamMA achieves better average displacement error and final displacement error than the state-of-the-art algorithms.
cs.LG / 6 / 2609.08043
A Quantitative Evaluation Framework for Temporal Explainability in Echocardiographic Video Segmentation
Abstract
Deep learning has achieved state-of-the-art performance in echocardiographic video segmentation, with an increasing number of models incorporating temporal information. However, quantitative evaluation of temporal explainability remains largely unexplored. We propose a quantitative framework for evaluating Grad-CAM explanations using four complementary metrics measuring temporal consistency, saliency motion, anatomical overlap, and temporal overlap. Using EchoNet-Dynamic, we compare a baseline 2D U-Net with ConvLSTM U-Net models trained across multiple temporal strides. While segmentation performance remained comparable across all models, intermediate ConvLSTM explanations exhibited substantially lower saliency consistency and greater centroid motion than final prediction explanations. Temporal Bottleneck explanations were significantly more stable than Encoder Bottleneck explanations across all strides, while final ConvLSTM Decoder3 explanations were broadly comparable to those of the 2D U-Net. Importantly, conventional frame-wise explanation metrics cannot determine whether variation in intermediate explanations reflects meaningful temporal feature evolution or explanation instability. These findings establish a preliminary quantitative framework for temporal explainability and motivate temporal-aware XAI methods that explicitly account for evolving representations in medical video models.
cs.LG / 7 / 2609.08230
ActionSplice: In-Flight Action Editing for Interactive World Models
Abstract
Chunk-autoregressive video world models typically condition each generated chunk on one action. An action received during sampling must therefore wait for the next chunk, condition future solver evaluations on a state produced under the previous action, or trigger rollback that repeats completed evaluations. We introduce ActionSplice, an inference framework that formulates this problem as Counterfactual State Transport (CST). A lightweight corrector transports the interrupted backbone-native representation toward the matched state induced by the revised action at the same solver step. The world model and sampler remain frozen, and sampling resumes without replaying completed evaluations. The retargeting variant $\mathrm{CST}*{R}$ updates the entire active chunk, while the temporal-splicing variant $\mathrm{CST}*{T}$ preserves a temporal prefix and updates only the suffix. Across minWM-Wan Action2V and HY-WM1.5, $\mathrm{CST}*{R}$ reduces rollback-relative LPIPS by 61.5% and 75.9% relative to direct condition swapping. $\mathrm{CST}*{T}$ reduces suffix LPIPS by 56.1% and 77.5%, respectively, while providing $2.73\times$ and $1.69\times$ pixel-ready speedups over waiting. Under the HY-WorldPlay protocol, $\mathrm{CST}_{R}$ obtains a PSNR of 25.66 dB, an SSIM of 0.6902, and an LPIPS of 0.1337 against the original rollout.
cs.LG / 8 / 2609.08345
CoVeR: Coverage-Based Token Pruning for Multi-View 3D Reasoning in VLMs
Abstract
Representing a 3D scene as multi-view images allows 2D VLMs to reason in 3D by reusing priors from pre-training, sidestepping the scarcity of annotated 3D data. However, it produces thousands of redundant visual tokens whose cost grows with every view. Existing visual token pruners fall into two families, each limited in the 3D multi-view setting. Learned importance methods rank tokens by attention or encoder features; because redundancy here is fundamentally spatial, they keep near-duplicate tokens from a few prominent regions and leave most of the scene unrepresented. Voxelization methods improve spatial coverage but cannot enforce an exact token budget and saturate as multi-view observations overlap in 3D, capping retention well below the target. We show that spatial coverage is associated with 3D reasoning performance and introduce CoVeR, a deterministic, training-free selector that uses only token coordinates, with no learned signals. CoVeR selects tokens that collectively cover every region of the scene, and solves the limitations of both families: it enforces an exact per-scene budget, breaks the voxelization saturation plateau, and avoids the near-duplicate selections of learned importance. Extensive experiments show CoVeR outperforms prior SOTAs on all three 3D reasoning benchmarks and generalizes as a plug-and-play module tested across four VLMs. Notably, with only $\approx$8% of visual tokens, it preserves 93.5% of full-token performance, surpassing SOTA by 3.9 percentage points on average across benchmarks.
cs.LG / 9 / 2609.08367
To Adapt or Not to Adapt? Selective Adaptation for Vision-Language Models
Abstract
Test-time adaptation (TTA) has emerged as a prominent strategy for adapting vision-language models to distribution shifts during inference. We conduct a per-sample analysis of model predictions before and after adaptation, and observe two failure modes in existing TTA methods that echo previous work. Adaptations are frequently negligible, yielding no change in the model's predictions, and more severely, they can be detrimental by flipping previously correct predictions to incorrect ones. This naturally raises a question: Can we identify and skip such negligible or harmful adaptations? In this work, we introduce a new problem of selective adaptation, which aims to determine whether a given test sample should undergo adaptation or be skipped. To this end, we propose Cross-Augmentation Similarity (CAS), a simple baseline that performs adaptation only when predictions across augmented views exhibit low similarity. Notably, CAS not only preserves but in some cases improves overall accuracy, even when skipping nearly 85% of the adaptation process. We hope other researchers will explore this new direction and surpass the performance of our baseline. Our code is available at https://github.com/sirujiang/selective-adaptation.
cs.LG / 10 / 2609.08505
Temporal State Transport in Video Generation: Diagnosing and Correcting Spectral Imbalance
Abstract
Reliable video generation requires more than high-quality frames to form a coherent story: a model must maintain a persistent state, transporting visual attributes such as identity, scene layout, motion, and fine details across time. Existing training-free methods mainly strengthen cross-frame attention or analyze local attention entropy, but these views do not reveal whether temporal interactions stay in a healthy transport regime. In this work, we study video generation through the perspective of Temporal State Transport. We introduce Spectral Tension, a signed diagnostic that compares local attention diffuseness with global spectral diversity, and use it to identify two opposite temporal failures: fragmented transport and over-mixing hotspots. Based on this diagnosis, we propose Spectral Transport Homeostasis, a training-free regulator that softly corrects pathological temporal states while largely preserving balanced ones. Experiments on pretrained video generation models show that the original model often occupies imbalanced temporal regimes, whereas our method selectively applies larger corrections to the worst temporal hotspots and improves temporal consistency and visual quality without finetuning. Code: https://github.com/lytang63/temporal-state-transport
cs.LG / 11 / 2609.08657
Charts Are Beyond Pixels: Probing for Layer-Wise Chart Understanding and Editing
Abstract
Charts are structured visual compositions whose elements have distinct functional roles, semantic correspondences, and visibility relations. This structural view motivates evaluating whether models can understand and manipulate charts at the layer level. Existing chart benchmarks, however, primarily assess the correctness or fidelity of final outputs and do not directly evaluate these layer-wise behaviors. We present LayerWiseBench, a benchmark organized around three core concepts, layer attribution, layer binding, and visibility ordering, that structure its chart-understanding and chart-editing evaluations. Generated from executable chart programs, LayerWiseBench pairs each rendered chart with spatially aligned per-layer RGBA assets and construction-derived labels for functional roles, semantic bindings, and visibility relations. From this layer-wise representation, we derive controlled understanding questions, editing targets, reference images, and evaluation regions. It contains 2,800 source charts across 14 chart paradigms, from which we derive 7,329 layer-wise understanding questions and 53,791 instruction-guided editing variants. Among the evaluated VLMs, Qwen3.5-27B, which achieves the highest QA macro-average, obtains 93.04% accuracy on layer attribution and 97.46% on layer binding, but only 61.46% on visibility ordering. Across the four evaluated image editors, overall mIoU ranges from 1.49% to 4.93%, and visibility-constrained edits have the lowest mIoU for every editor, ranging from 0.37% to 2.00%. Taken together, these results identify tasks involving front-to-back relations between overlapping components as a recurring challenge across understanding and editing, motivating more explicit modeling of component identity and visibility relations.
cs.LG / 12 / 2609.08873
High-Magnetization Sampling at Low Temperatures: Ising Models and Bayesian Sparse Linear Regression
Abstract
Sparsity is a powerful structural resource in optimization and statistics. We develop frameworks for leveraging sparsity in sampling problems over the Hamming slice $\mathcal{X}_k^d:=\{\mathbf{x}\in\{\pm 1\}^d:|\{i:\mathbf{x}_i=1\}|=k\}$, in high-dimensional regimes where $k\ll d$ (i.e., where $\mathcal{X}_k^d$ is \emph{highly magnetized}). We use our frameworks to design improved samplers for canonical problems in the study of \emph{Ising models} and \emph{Bayesian sparse linear regression}. Our first main result considers the \emph{Sherrington--Kirkpatrick} (SK) model restricted to fixed-magnetization slices $\mathcal{X}_k^d$. We give a polynomial-time sampler for fixed-magnetization SK models at any inverse temperature $β>0$, under arbitrary external fields, provided that $k\le c_βd$ for an appropriate constant $c_β$. By combining this result with an annealing strategy for estimating normalizing constants, we obtain polynomial-time samplers for the SK model at arbitrarily low temperatures under a sufficiently strong external field of strength $h$. In the large-$β$ limit, our framework permits sampling at field strengths within constant factors of the \emph{Almeida--Thouless line} delineating the replica-symmetric and replica-symmetry-breaking regions ([dAT78]), improving polynomially over the field strength $h(β)$ required by the recent work of [BAR26]. Our second main result concerns the measurement complexity of polynomial-time Bayesian sparse linear regression. Recent work by [KSTZ25] shows how to sample from the canonical \emph{Gaussian spike-and-slab posterior} with expected sparsity $k$, at any signal-to-noise ratio, given $n\gtrsim k^3\log^3 d$ Gaussian measurements. We improve this requirement to $n\gtrsim k^{3/2}\log^2 d+k\log^3 d$, using a common sparsity-aware framework underlying both our results.
cs.LG / 13 / 2609.08312
Non-Coherent Over-the-Air Federated Learning: Protocol, Convergence, and Device Scheduling
Abstract
To mitigate the scalability bottleneck in the radio access network (RAN) in federated edge learning (FEEL), over-the-air federated learning (AirFL) exploits waveform superposition over multiple-access channels (MACs) for analog model aggregation. However, coherent AirFL typically relies on stringent PHY-layer conditions such as accurate channel state information (CSI), tight time/frequency synchronization, and frequent transceiver calibration for signal alignment. However, these requirements, if not impossible to be met, incur substantial communication and computation overhead. In this paper, we propose a non-coherent AirFL (NCAirFL) protocol over a broadband single-antenna MAC, leveraging binary dithering, unbiased non-coherent detection, and long-term error feedback to waive the need for instantaneous CSI. For NCAirFL with general smooth non-convex objectives and a constant learning rate, we establish a convergence bound achieving the convergence rate in the same order of $\mathcal{O}(1/\sqrt{T})$ as communication-ideal FedAvg, where $T$ is the total number of communication rounds. To further improve communication efficiency under data and wireless resource heterogeneity, we also derive a lower bound on the expected single-round objective decrease in the global loss conditioned on device scheduling, building upon which a surrogate objective function is obtained for jointly optimal device selection and power control. Experimental results on MNIST and CIFAR-10 corroborate that NCAirFL achieves learning performance close to FedAvg in practical settings, with the proposed device scheduling policy substantially accelerating convergence.
cs.LG / 14 / 2609.08759
A Note on Scaling in Randomly Rotated Quantization and Its Connection to the CDEF +1 Pythagorean Relation
Abstract
Quantization schemes based on randomized rotations have recently received renewed attention, including the roles of MMSE and unbiased reconstruction scalings. In this note, we point out the connection to classical results in statistical signal processing and communication theory. Specifically, the two reconstruction scales used in the EDEN line of work admit a natural interpretation as finite-dimensional, realization-dependent counterparts of the Wiener and unbiased coefficients in the classical CDEF formulation. At finite blocklength, the CDEF +1 relation holds pointwise for each rotation realization as an exact geometric (Pythagorean) identity, but does not hold after averaging the distortions over the rotation. The classical SNR relation $\sf{SNR}_{\rm MMSE}=\sf{SNR}_{\rm MMSE,U}+1$ is recovered as $d\to\infty$: once the overall scale is handled separately, the empirical coordinate statistics of a randomly rotated vector approach their i.i.d. Gaussian counterparts, and the rotation-dependent quantities concentrate. Importantly, EDEN goes beyond this classical correspondence: for every finite $d$, its Haar-rotation formulation guarantees exact conditional unbiasedness, a stronger property than the second-order notion of unbiasedness in CDEF. We further comment on two distinct roles random rotations play in quantization: one is approximate Gaussianization of the coordinates; the other is decorrelation of reconstruction errors across quantization branches.
cs.LG / 15 / 2609.07897
The Accuracy Paradox: Empirical Diagnostic of Default Decision Thresholds in Multi-Label Enzyme Commission Prediction [With Code]
Abstract
Automated prediction of Enzyme Commission (EC) numbers plays a central role in functional annotation and computational drug discovery. However, standard multi-label machine learning pipelines frequently rely on default decision thresholds (t=0.50), assuming balanced prior distributions across target heads. In this study, we present a systematic empirical diagnostic of uncalibrated fixed decision boundaries operating under severe class imbalance across N = 14,096 annotated compounds categorized into six primary EC classes (EC1-EC6). Our results highlight a pronounced Accuracy Paradox: while the multi-label system achieves a deceivingly high mean accuracy of 77.16%, the macro F1-score (0.3976) and macro recall (0.3872) reveal severe predictive breakdown. Majority target classes suffer from hyper-sensitivity and over-prediction, whereas minority classes exhibit sharp recall decay, culminating in a total decision boundary collapse for EC6 (Recall = 0.00%) despite underlying discriminative power (ROC-AUC = 0.5857). Feature correlation analysis further reveals high linear redundancy among topological indices relative to fingerprint density metrics. Ultimately, this diagnostic study demonstrates that standard point predictions mask critical errors in bioinformatics workflows. We establish target-specific threshold optimization and post-hoc conformal calibration as essential, open-source post-processing safeguards for reliable applied machine learning and deep learning architectures.
cs.LG / 16 / 2609.07917
AVCG: A Generalized Variational Framework for Counterfactual Generation under Hypothesis Distributions
Abstract
Counterfactual explanations formalize "what-if" scenarios by identifying modifications to an input instance that obtain a desired alternative prediction. Traditionally, whether generated via instance-specific optimization or amortized single pass models, these approaches rely on a single, deterministic point-estimate predictor. However, this ignores predictive uncertainty and hypothesis variability, leading to brittle explanations that frequently become invalid if the underlying model is retrained or updated. To address this fragility, we propose the Amortized Variational Counterfactual Generator (AVCG), a generalized optimization framework that formulates counterfactual generation as optimization over an arbitrary distribution of plausible predictive hypotheses rather than a single deterministic predictor. This formulation naturally accommodates Bayesian posteriors, Rashomon-restricted hypothesis spaces, and other uncertainty representations within a unified optimization framework. Evaluation across multiple benchmark datasets demonstrates that the AVCG framework produces counterfactual explanations that remain highly valid under predictive uncertainty and model changes, while maintaining competitive plausibility and single-pass runtime performance.
cs.LG / 17 / 2609.07922
Prevalence calibration as shortcut mitigation
Abstract
Shortcut learning denotes the widespread situation in which a classifier exploits spurious correlations rather than diagnostic features. Existing mitigation strategies mostly aim to learn shortcut-invariant representations; their empirical success is limited and they cannot be applied to classifiers using frozen foundation model encoders. We propose to reframe shortcut learning as fundamentally a calibration problem: unconstrained learning implicitly calibrates each shortcut group to its training set disease prevalence, rendering the resulting classifier necessarily over-confident in one group and under-confident in the other. Building on this insight, we prevalence-equalize calibration between shortcut groups through two encoder-agnostic methods, an in-processing regularizer and a post-hoc prevalence-equalized recalibration step. Across chest-drain-pneumothorax benchmarks on CheXpert and SIIM-ACR, spanning fine-tuned CNNs and frozen foundation-model backbones, both methods substantially outperform all baselines. Post-hoc recalibration of a standard ERM-trained DenseNet raises misaligned-group AUROC from 0.23 to 0.73, indicating that shortcut reliance degrades the classification head rather than the underlying representation. Besides two new state-of-the-art shortcut mitigation approaches, our findings more fundamentally connect shortcut learning to calibration theory and algorithmic fairness.
cs.LG / 18 / 2609.07956
Streaming Hierarchical Inference with Tabular Foundation Models
Abstract
Tabular Foundation Models (TFMs) have recently demonstrated strong predictive performance through in-context learning, but their deployment in high-throughput data streams remains challenging due to communication overhead and latency. We propose \textit{HINT}, a hierarchical inference framework that combines edge-based retrieval with cloud-based TFM inference. A graph-based approximate nearest neighbor memory maintained over a sliding window provides local predictions and uncertainty estimates, allowing confident samples to be processed locally while uncertain instances are selectively offloaded, together with their retrieved context, to a cloud-hosted TFM. The framework exposes an offloading threshold and a neighborhood retrieval policy that can be varied to balance predictive performance and communication cost. Experiments show \textit{HINT} consistently identifies favorable trade-offs.
cs.LG / 19 / 2609.07961
$α$-Graph: Attention-Infused Normalizing Flow Approach to Tractable Graph Modeling
Abstract
Graph modeling, a crucial task for representing complex relationships in graph-structured data, has achieved significant success in recent years. However, current graph modeling methods rely on traditional Graph Neural Networks and pre-training approaches to implicitly learn the underlying relational structure of graph data. Thus, these prior methods cannot capture the complex graph structure and correlations among inputs. In this paper, we introduce a novel Attention-based Normalizing Flow-based Approach\footnote{Our implementation and models will be released publicly for research reproducibility.} (ANFA or $α$) that provides an explicit, interpretable, and tractable Graph Modeling ($α$-Graph). In particular, we propose a new Unconditional Graph Normalizing Flow with an Invertible Attention Mechanism to capture the complex relational structure of graph data. To further enhance the expressiveness of the model, we introduce Conditional Graph Normalizing Flow with Learnable Queries that enables efficient modeling of correlations in graph-structured data. We show that our Conditional Graph Normalizing Flows behave similarly to Unconditional Graph Normalizing Flows, enhancing expressiveness while maintaining training stability and efficiency. Our experimental results on three benchmarks will illustrate the effectiveness and the state-of-the-art (SoTA) performance of the proposed $α$-Graph method.
cs.LG / 20 / 2609.07975
Heat Field Signatures: From Point Clouds to Smooth Geometry
Abstract
Bringing multiscale geometric analysis directly to irregular point clouds remains difficult: quantities such as local dimension, anisotropy, density variation, and geometric transitions are typically estimated through explicit neighborhood, manifold, or graph constructions, or left for neural networks to infer from coordinates. We introduce Heat Field Signatures (HFS), which lift a point cloud to a multiscale family of smooth ambient heat fields, providing a direct interface from discrete samples to geometric analysis. From this field, HFS computes closed-form global and local signatures directly from pairwise distances, capturing heat concentration, intrinsic dimension, anisotropy, and scale transitions. We further introduce the Heat Dimension Spectrum (HDS), a compact summary of multiscale geometric composition. HFS can be used as a closed-form descriptor, a lightweight learned representation, or a geometric feature channel for neural point-cloud models. Across synthetic and real-world benchmarks spanning subcellular, neuronal, tree, and protein data, HFS outperforms strong point-cloud and multiparameter-persistence baselines while substantially reducing end-to-end cost. On SCOP protein-fold classification, HFS improves over the strongest deep baseline by nearly $24$ percentage points using coordinates alone, while standalone HFS representations are exactly rotation-invariant by construction. More broadly, HFS turns a classical heat field into a practical interface for multiscale geometric analysis in modern point-cloud learning.
cs.LG / 21 / 2609.07982
Semi-Supervised Learning under Spatially Biased Sampling
Abstract
Standard semi-supervised learning (SSL) typically relies on labelled and unlabelled data sharing a common marginal distribution. This assumption is often violated by biased spatial sampling mechanism, when labels are collected under spatially biased or preferential site selection. We treat this marginal mismatch, spatial autocorrelation, and spatial non-stationarity as three distinct mechanisms, varied independently via a labelled-sampling concentration parameter, a spatial length scale, and a non-stationarity strength parameter, and ask how mismatch degrades SSL, whether the cluster and manifold assumptions survive it, and how the resulting failure can be diagnosed. Using a controlled synthetic framework alongside PovertyMap-WILDS, California housing, socio-economic and US air quality monitoring datasets, we systematically vary the degree of mismatch while accounting for spatial autocorrelation and non-stationarity. Through a series of analyses including a segmented-regression changepoint, we show that in the synthetic generator, SSL performance does not degrade gradually but instead exhibits a threshold-like breakdown between approximately 0.71 and 0.77 once distribution mismatch becomes sufficiently severe. We further demonstrate that spatial non-stationarity contributes to performance loss independently of marginal mismatch and that models become increasingly overconfident outside the regions where labels are available. To support practical deployment, we evaluate several distribution-divergence measures as indicators of reliability and introduce a kernel-weighted local divergence metric that provides a more stable estimate of spatial mismatch than a naïve localised approach. These findings provide empirical evidence and diagnostic tools for better documenting the risk of incorporating unlabelled spatial data into semi-supervised learning workflows.
cs.LG / 22 / 2609.07983
Solving the Elastic Wave Equation with Physics-Informed Neural Networks: A Robust and Critical Assessment
Abstract
Physics-Informed Neural Networks (PINNs) have recently emerged as a promising approach for solving Partial Differential Equations (PDEs), offering a meshfree alternative that integrates physical principles into the learning process. This presents a new paradigm compared to traditional discretization methods and purely data-driven machine learning techniques. While promising, PINNs are not a panacea; they inherit challenges such as spectral bias and unstable convergence. Moreover, their potential in seismology remains largely unexplored. In this work, we provide a robust and critical assessment of PINNs for solving the elastic wave equation in seismology. We investigate the performance of PINNs on problems with varying degrees of complexity across various seismic sources and parameter models, from constant to highly heterogeneous settings. A pivotal aspect of our work involves investigating whether embedding physical principles directly into the network architecture enhances convergence and accuracy. We test an extensive range of neural architecture designs, from unrestricted, uninformed PINNs to highly specialized ones. We find that integrating an understanding of wave physics into the network design significantly improves accuracy. For instance, introducing a custom wavelet or plane wave layer, coupled with encoder and decoder layers, consistently yields a relative $L_2$ error approximately half that of the standard PINN, as evidenced across numerous experiments. We further demonstrate that this novel architecture enhances accuracy when applied to the acoustic wave equation, underlying the versatility of our network. Another key contribution of our research is the successful conditioning of PINNs on seismic source locations. This signifies a considerable advancement towards rapid seismic hazard detection and seismic analysis.
cs.LG / 23 / 2609.07997
Sharp Structure-Agnostic Minimax Risk for Partial Linear Models
Abstract
We characterize the sharp structure-agnostic minimax risk for coefficient estimation in the partial linear model when the outcome and treatment nuisances are learned by two distinct black-box learners, which resolves the open problem in double machine learning posed by Gu (2025). For each nuisance \(q\in\{μ,π\}\), we characterize the available learner by an approximation-error budget \(a_q\) and a stochastic-error budget \(s_q\), with the latter controlled through localized Rademacher complexity. Writing \(\mathcal E_n\) for the minimax mean-squared error, we show that \[\mathcal E_n\asymp1\wedge\left\{\frac1n+\left(a_μa_π+\min\left\{a_πs_μ+s_π^2,\,a_μs_π+s_μ^2\right\}\right)^2\right\}.\] The main new ingredient is a novel lower bound for the general two-learner problem. Our proof constructs four finite-mixture testing experiments using orthogonal code functions. Across these experiments, the hidden perturbations are placed outside both learner classes, outside only the treatment learner class, outside only the outcome learner class, or inside both learner classes. These four configurations capture, respectively, the interaction between the two approximation errors, the two asymmetric interactions between one learner's approximation error and the other learner's learning error, and the joint estimation difficulty of learning both nuisances. Combining the four resulting lower bounds yields the displayed rate, which matches the latest upper bound in Gu (2026). Our result shows that standard double machine learning can overstate the intrinsic difficulty of target estimation and provides a target-specific principle for learner selection: approximation error and stochastic complexity must be jointly balanced across the two nuisance learners rather than optimized separately.
cs.LG / 24 / 2609.08034
Two-Scale Localized PCA-Net: Coarse-Global and Local-Residual Representations for Artifact-Reduced PDE Operator Learning
Abstract
Localized dimensionality reduction improves the scalability of operator learning for high-dimensional partial differential equations (PDEs), but independently decoded local patches can introduce block offsets, interface mismatches, and spurious high-wavenumber content. We introduce Two-Scale Localized PCA-Net, which decomposes the solution into a coarse-global component and local residual corrections. A compact global PCA basis captures domain-scale structure, while nonoverlapping local PCA bases represent the remaining fine-scale residual. A block-balanced latent objective couples the two representations, and optional interface-aware fine-tuning further promotes continuity through reconstruction and trace losses. On Poisson benchmarks, the two-scale representation substantially reduces reconstruction error and visible block artifacts relative to plain and overlap-based localized PCA-Net while approximately halving PCA fitting cost relative to overlap. On heterogeneous Darcy flow, it strongly reduces interface and discrete-residual errors, with more modest reconstruction gains. Ablations show that the primary improvement arises from the two-scale output representation, while interface-aware fine-tuning provides complementary continuity refinement. Overall, separating globally coherent structure from localized residual detail provides an efficient representation for artifact-reduced PDE operator learning.
cs.LG / 25 / 2609.08078
A Machine Learning Framework for Predicting Restaurant Food Waste to Support Sustainable Food Management
Abstract
Food waste in the restaurant sector poses a substantial challenge to environmental sustainability and economic efficiency. This paper presents an exploratory machine learning framework for estimating daily restaurant food waste quantities from operational and contextual features. A structured dataset was constructed by integrating restaurant demand records, meteorological data and temporal event indicators, yielding 77,980 records across 27 features. Because large-scale ground-truth food waste measurements are not publicly available, the target variable was derived from operationally justified assumptions, with the complete construction formula and controlled stochastic variability disclosed for full reproducibility. Four supervised regression models, namely Linear Regression, Decision Tree, Random Forest and Gradient Boosting, were evaluated under a chronological 70-30 train-test split that respects the temporal ordering of restaurant operations, augmented by 5-fold time-series cross-validation. All reported metrics are explicitly scoped to performance against the constructed target and do not imply validation against measured food waste. Ensemble methods consistently outperformed linear baselines. Random Forest attained an MAE of 6.19 kg, RMSE of 8.36 kg and $R^2$ of 0.817 on the realistic feature subset following systematic exclusion of algebraically leakage-prone variables. Feature importance analysis identified menu diversity, operational area and temporal activity patterns as the primary predictive drivers. The full dataset, target construction formula, codebase and experimental configurations are publicly released to support reproducibility and future extension to empirically measured waste data.
cs.LG / 26 / 2609.08080
Proactive Context-Forecasted Safety Constraints for Nonstationary Reinforcement Learning
Abstract
Ensuring safety in reinforcement learning under nonstationarity requires anticipating changes in risk before they lead to unsafe behavior. Existing approaches typically rely on safety constraints defined at design time or updated reactively during execution, assuming that such constraints remain valid over time. However, in nonstationary environments with evolving contexts and changing driving layouts, these assumptions may fail. We propose a framework for proactive safety constraint generation based on context forecasting. The approach infers latent environmental context from observations, predicts its future evolution, and constructs safety constraints adapted to anticipated conditions. This enables the agent to proactively avoid unsafe regions instead of reacting only after safety violations occur. We evaluate the method in driving environments with structured context variation. The experiments include a sweep over nonstationarity intensities and additional held-out driving layouts, including highway, intersection, and racetrack scenarios. Results show that proactive constraint generation substantially reduces collisions under both seen and out-of-training nonstationarity intensities and generally remains effective across held-out driving layouts while maintaining usable task performance. These findings suggest that context-based constraint generation is a promising approach for safe reinforcement learning under nonstationarity.
cs.LG / 27 / 2609.08102
Learning Metamaterial Eigenmodes with Wavelet-Encoded Fourier Neural Operators
Abstract
Machine learning surrogates based on neural operators have shown broad applicability in solving forward PDE problems. However, eigenvalue problems, in which an eigenparameter and one of several valid eigenmodes must be simultaneously solved, remain difficult because standard operator learning formulations assume a unique input-output map. This work demonstrates that Fourier Neural Operators (FNOs), combined with wavelet-based encodings of PDE inputs, can learn and predict multiple eigenmodes of the elastic wave equation, corresponding to deformation modes of acoustic waves propagating through arbitrary metamaterial geometries. We provide a mechanistic explanation and experimental evidence for why wavelet encodings are well matched to the dual spatial-spectral structure of the FNO, enabling deterministic mode selection on both continuous-valued and binary-valued geometries within a single model, and for why prediction accuracy varies with geometric discontinuities. For metamaterial design, the resulting surrogate accelerates the simulation stage of the design cycle by three orders of magnitude relative to finite element analysis on a consumer-grade CPU, while preserving high fidelity. These results also carry broader implications for designing input encodings in other multi-mode PDE solvers based on spectral neural operators.
cs.LG / 28 / 2609.08106
Nyström Attention Matches Full Attention for Cross-Sectional Stock Prediction
Abstract
MASTER's inter-stock multi-head attention -- the module responsible for modeling cross-sectional stock relationships -- accounts for 42.5% of model parameters and 25% of predictive value. We systematically decompose this module and uncover a surprising structure: the learned attention is near-uniform (perplexity 278/300), yet forcing exact uniformity eliminates all cross-sectional discrimination. Spectral analysis resolves this paradox: the deviation from uniformity is low-rank (effective rank ~65, top-10 modes capture 96.5% of energy), explaining why sparse approximations consistently fail while Nystrom low-rank attention (m=32 landmarks) matches full O(N^2) attention at O(mN) cost -- certified equivalent via TOST at both N=300 (5 seeds, Rank IC p=0.003) and N=800 (10 seeds, Rank IC p=0.034). Additional findings include: (i) attention anti-correlates with return similarity (Spearman rho = -0.614; on the industry-labeled subset, -0.645 unconditionally and -0.627 after controlling for industry, beta, and volatility), suggesting complementarity-seeking rather than correlation mining; (ii) all graph-based alternatives degrade performance, with hard masking worse than complete module removal; and (iii) at N ~ 3,500 with adapted architectures, no cross-stock module (GCN, Nystrom, or MASTER-style pipeline) significantly outperforms a per-stock LSTM baseline (n=4 seeds), indicating that the benefits observed at smaller scales do not trivially transfer. These results establish that the inter-stock attention's value resides in a compressible, dynamic, near-global redistribution that rewards low-rank approximation but resists sparsification.
cs.LG / 29 / 2609.08133
Sparse Data Augmentation for Optimization with Provable Guarantees
Abstract
In nonconvex optimization problems arising in geometric machine learning, data augmentation is commonly used to promote invariance by averaging empirical losses over transformations of the data. Computing the fully augmented objective, however, requires access to every element of the transformation group $G$, which may be prohibitively expensive when $G$ is large or accessible only through sampling. We study whether full augmentation can instead be approximated using a small, fixed sample of transformations acquired before optimization and reused thereafter. Under suitable regularity conditions, we show that, with probability at least $1-δ$, gradient descent (GD) on the resulting sparsely augmented objective returns an $\varepsilon$-stationary point of the fully augmented objective using $\mathcal{O}\bigl((\log |G|+\log(1/δ))/\varepsilon^2\bigr)$ group-transformation-oracle queries. By comparison, standard group stochastic gradient descent (group-SGD), which samples a fresh transformation at every iteration, uses $\mathcal{O}(1/\varepsilon^4)$ transformation queries. Therefore, gradient descent with fixed sparse augmentation requires fewer transformation queries than both GD applied to the fully augmented objective and group-SGD. Our proof techniques, which may be of independent interest, establish a uniform approximation of the full group-averaged gradient field by a random group average using spectral properties of group-induced operators and tools from representation theory.
cs.LG / 30 / 2609.08135
KBBQ: A Predictive Noise Law and the Limits of Spectrum Flattening in FP4 Quantization
Abstract
We develop a second-order theory of quantization noise in matrix multiplication in which the quantization format is characterized by the variance it assigns to each element. The constant variance profile of integer quantization recovers existing integer-noise theory, while the multiplicative profile of floating-point rounding reduces the data dependence to a scalar, the participation factor $κ$, yielding a closed-form signal-to-noise-ratio law. The resulting functional also admits a closed-form upper bound $κ^{*}$ that no function-preserving linear transform can exceed and that is attained by a recent state-of-the-art method. Building on this analysis, we introduce KBBQ (\textbf{K}appa-\textbf{B}raked \textbf{B}lockwise \textbf{Q}uantization), which parameterizes the extent to which a transform approaches this ceiling. At W4A4, across four base models and two FP4 formats, KBBQ outperforms the prior state of the art without additional deployment-time computation.
cs.LG / 31 / 2609.08136
GPU-Enabled Large-Scale Optimization Using Randomized Linear Algebra
Abstract
This paper introduces rlaopt, a PyTorch-based package for large-scale optimization and scientific computing using randomized numerical linear algebra (RandNLA). Despite substantial progress in RandNLA-based algorithms, few implementations combine GPU acceleration with a simple interface for specifying optimization problems. rlaopt addresses this gap by providing GPU-enabled solvers for positive-definite linear systems and convex empirical risk minimization with constraints and regularizers. These solvers use RandNLA to accelerate conjugate gradient (NystromPCG), operator splitting (NysADMM), and stochastic gradient methods (SAPPHIRE). Moreover, rlaopt includes a modeling language that lets users specify problems using natural mathematical syntax. rlaopt automatically checks compatibility with the selected solver and performs the required problem decomposition. The solvers also support differentiation through their iterations, enabling applications such as hyperparameter tuning. Experiments on ridge regression, bounded multinomial logistic regression, and bounded elastic net identify when randomized preconditioning improves performance and demonstrate substantial speedups from GPU execution. The package is open-source under an Apache license, with source code at https://github.com/udellgroup/rlaopt and version 0.1.0 available on PyPI.
cs.LG / 32 / 2609.08152
Topology-induced Operators Reveal Complementary Graph Representations without Training
Abstract
Graph representation learning has largely focused on designing increasingly sophisticated models to transform graph topology into vector representations, or embeddings. However, the extent to which embedding quality depends on model learning, rather than on the underlying topological transformations, remains unclear. Here, we show that informative embeddings can be derived without complicated model design and gradient-based training. Propagating random features through implicit hierarchical structures induced by random walks and anonymous walks yields embeddings that capture node proximity and structural role, respectively. These two training-free embeddings preserve complementary aspects of graph organization and perform competitively with classic and recent methods across various node-, edge-, and graph-level tasks. They often require substantially less computation, resulting in a favorable quality-efficiency trade-off. Combining the two types of embeddings further improves inference quality of some tasks compared with using either embedding type alone. Our results suggest that informative graph embeddings can arise from carefully chosen topological transformations before any learning operation is applied.
cs.LG / 33 / 2609.08200
SIM: Subspace Interaction-based Method for Token-Level Text Anomaly Detection
Abstract
Token-level text anomaly detection, as an emerging trend of text anomaly detection, moves beyond coarse-grained document-level detection by localizing anomalous tokens within text. By providing fine-grained abnormality prediction, token-level text anomaly detection plays a critical role in various real-world applications, such as spam filtering and fake news detection. However, existing methods still rely on the global distance calculation for scoring, during which the local anomaly signals are severely diluted by numerous redundant normal feature dimensions. Moreover, pre-trained language models used in these methods inevitably smooth out surface anomalies, further limiting their effectiveness in token-level anomaly detection. To address these limitations, we propose a Subspace Interaction-based Method (SIM for short) for token-level text anomaly detection. To prevent local signal dilution, SIM adopts a subspace interaction-based anomaly detector, which decouples high-dimensional token embeddings into multiple low-dimensional ones, amplifying localized anomaly signals hidden within specific dimensions. To counteract the over-smoothing effect, we design a hard pseudo-anomaly generation module to construct pseudo-anomalous tokens, simulating the subtle anomalies obscured by semantic smoothing. Also, a probabilistic boundary loss is developed to standardize anomaly scores into statistical distances, effectively enforcing anomalous instances to deviate significantly from the normal distribution center. Extensive experiments on multiple benchmark datasets verify the effectiveness of SIM and demonstrate its remarkable efficiency, robustness, and interpretability. The source code is available at: https://github.com/yankehan/SIM-TAD.
cs.LG / 34 / 2609.08244
CUNO: Curriculum and Preference Optimization for Stable Graph Unlearning under Mass Deletion
Abstract
Graph unlearning removes the influence of designated training data from a trained graph model without retraining from scratch. However, existing methods suffer a sharp drop in model utility under large deletion ratios (mass deletion), a phenomenon we refer to as catastrophic unlearning. We find that a key cause is the uniform treatment of all deleted samples, which is particularly damaging in graph learning: structural dependencies cause different nodes to play vastly different roles in the learned model, yet existing methods apply the same forgetting operation to the entire forget set. Based on this insight, we propose CUNO, a curriculum-based graph unlearning framework that removes the forget set progressively, ordering samples by their estimated unlearning difficulty across multiple stages. CUNO further employs a distribution-level negative preference optimization (NPO) objective at each curriculum stage that steers the model away from its original behavior on the current forget subset while preserving retained performance. Our theoretical analysis shows that the curriculum design is most beneficial when the forget set spans a wide range of unlearning difficulty, a condition naturally satisfied under mass deletion. Comprehensive experiments confirm that CUNO consistently mitigates catastrophic unlearning: at 20% deletion, it retains 74% of the original utility compared to 26-53% for existing methods, and maintains more than half the original utility even at 50% deletion. Our code is publicly available at https://anonymous.4open.science/r/cuno-D4FF.
cs.LG / 35 / 2609.08268
Synergistic Fusion of Topological Structure and Temporal Semantics of Mobility for Urban Region Embedding
Abstract
Urban region embeddings have shown promising results in diverse urban sensing tasks such as crime, income, and service-call prediction. Recent methods improve representation quality by integrating mobility data with auxiliary modalities, using cross-view attention or contrastive objectives to align heterogeneous features into a unified region representation. However, leveraging the temporal dynamics of human mobility remains under-explored. Regional inflow and outflow fluctuate throughout the day, and inter-region connections emerge, persist, and dissolve over time. Moreover, prevailing fusion strategies combine views additively and miss the joint signal that emerges only when views co-occur. To address these gaps, we propose Mobility Stream-Structure Synergy (MoSS), which derives complementary views from mobility data: a Sequence view that preserves each region's hourly inflow/outflow profile, and a Structure view based on zigzag persistence diagrams that capture how regional connectivity emerges, persists, and dissolves over time. A synergy module then extracts emergent representations from the co-occurrence of these views through multi-degree interactions, explicitly capturing higher-order signal across views. Extensive experiments on New York City and Chicago show that MoSS achieves state-of-the-art performance across three downstream tasks using mobility data alone, outperforming baselines that rely on auxiliary modalities.
cs.LG / 36 / 2609.08276
Online Signature Verification Using Augmented Path Signature and T-Mamba
Abstract
Handwritten signature verification is vital for personal authentication across commercial and financial applications. Although deep learning methods are widely adopted for online signature verification (OSV), they often struggle with capturing highly discriminative features and modelling long-range dependencies. To address these issues, we propose a novel framework that integrates the augmented path signature (APS) descriptor with the T-Mamba model. The APS descriptor first applies time and basepoint augmentations, then computes sliding-window path signatures. The path signature is a non-parametric feature map from rough path theory that effectively captures geometric structures and nonlinear inter-channel interactions. Inspired by the efficacy of state space models (SSMs) in sequence modelling, our T-Mamba model employs a hybrid design combining two temporal convolutional network (TCN) blocks with a time-scanning Mamba. This design enables the model to learn both local temporal patterns and global long-range dependencies, substantially improving verification accuracy. Our framework achieves state-of-the-art EERs on three public benchmark datasets (MCYT-100, SVC-2004 Task 2, DeepSignDB), validating its effectiveness and robustness, especially when the training data is limited. Our code is publicly available at https://github.com/DLRL04/OSV-using-APS-and-T-Mamba.
cs.LG / 37 / 2609.08277
Adaptively Incorporating Directional Hints into Zeroth-Order Optimization
Abstract
We study zeroth-order optimization of non-convex functions with the aid of directional hints, which are cheap but potentially inaccurate approximations of the true gradient direction, given by linear subspaces at each iteration. To leverage these hints adaptively while maintaining robustness to their quality, we introduce Control-Variate Zeroth-Order Descent (CV-ZOD), a new framework that refines the classical zeroth-order gradient estimator with a control variate that can be set based on the directional hints. We first show that the oracle algorithm that optimally sets the reference vector and step size at each iteration achieves a convergence rate that interpolates between the first-order $O(1/T)$ rate and the zeroth-order $O(d/T)$ rate, depending on the quality of the hints along the trajectory. We then develop a practical variant of CV-ZOD that achieves the same oracle guarantee up to logarithmic factors, without any prior knowledge of the hint quality. We validate the method empirically on simulation-based scientific optimization tasks, demonstrating sustained progress on non-convex landscapes where zeroth-order descent is slower and existing guided methods stall as guidance deteriorates.
cs.LG / 38 / 2609.08286
HypLTSF: A Hyperbolic Geometric View of Multi-Scale Hierarchies for Long-Term Time Series Forecasting
Abstract
Multi-scale modeling has become an effective approach for long-term time series forecasting, capturing temporal patterns that range from fine-grained local dynamics to coarse global trends. Representations across these temporal scales are inherently hierarchical, with coarser scales abstracting and aggregating information from finer ones. While existing approaches readily exchange information across these scales, the hierarchy itself is typically left as an emergent byproduct of such interactions rather than captured as a geometric structure in its own right. In this paper, we introduce HypLTSF, a framework that endows the multi-scale hierarchy with a concrete geometric form by embedding scale-wise representations into the Poincaré ball, whose exponentially expanding volume naturally accommodates hierarchical structures. To align this geometry with the temporal hierarchy, HypLTSF imposes two constraints: (1) a radial constraint that orders embeddings by their level of abstraction, and (2) an angular constraint that groups fine-scale patterns sharing a common coarser-scale ancestor. Extensive experiments on long-term time series forecasting benchmarks show that HypLTSF achieves state-of-the-art performance, suggesting that explicitly modeling the multi-scale hierarchy as a geometric structure is effective for forecasting.
cs.LG / 39 / 2609.08330
EMBLEM: Enhancing Multi-script Table Detection through Masking
Abstract
Table detection is a core task in document analysis, supporting downstream applications such as information retrieval, document reconstruction, and visual question answering. While existing deep learning models perform well on English and Chinese documents, they struggle with multilingual, multi-script documents due to script diversity and the limited availability of labeled data. To address this challenge, we introduce MANDALA (Multi-script Annotated Documents for Table Detection), a manually curated dataset of 2,323 table-containing pages spanning 18 languages and 15 scripts across diverse domains. We also propose EMBLEM, a masking-based paradigm for Multi-script Table Detection (MTD). EMBLEM generates masked images that conceal script- and font-specific details, enabling models pre-trained on abundant English documents to focus on script-agnostic page layout. Experiments across three table detection architectures show that EMBLEM consistently outperforms strong baselines on MANDALA while remaining competitive on five standard English-dominant benchmarks. Using only English masked images for fine-tuning, with no multi-script training data, EMBLEM achieves an absolute F1-score gain of 20.8% on MANDALA. We release MANDALA along with the accompanying code and models at https://github.com/IITB-LEAP-OCR/EMBLEM.git.
cs.LG / 40 / 2609.08337
Distillation as Probability Transport: Routed On-Policy Distillation
Abstract
On-policy distillation (OPD) transfers teacher knowledge on student-generated trajectories, but efficient sampled objectives reduce the teacher distribution to scalar credit on individual tokens. Such credit indicates whether a token should gain or lose probability, yet leaves the corresponding redistribution unspecified. We recast OPD as teacher-guided probability transport and propose RouteOPD (Routed On-Policy Distillation), which decomposes local teacher--student disagreement into student-excess sources and teacher-deficit destinations and couples them into explicit transport pairs. RouteOPD optimizes pairwise log-odds toward jointly realizable targets obtained from a bounded teacher potential, while adapting the transport budget to the concentration of teacher demand. This formulation directs updates toward teacher-preferred destinations and controls their magnitude within a single transport operator. Experiments across four teacher--student settings and four mathematical-reasoning benchmarks demonstrate that RouteOPD consistently outperforms sampled reverse-KL OPD, with improvements accompanied by higher routing fidelity and lower background leakage. These results demonstrate the effectiveness of explicitly modeling probability transport in on-policy distillation.
cs.LG / 41 / 2609.08354
Geometry-Aware Bayesian Parameter-Efficient Fine-Tuning on the Stiefel Manifold via Stein Variational Gradient Descent
Abstract
Several geometry-aware approaches to low-rank adaptation have emerged for parameter-efficient fine-tuning of large pre-trained models. These methods aim to take full advantage of the geometric structure of low-rank manifolds for improving the efficiency in subspace utilization and reducing redundancy by enforcing orthogonality constraints during optimization. The strong empirical results of these techniques have motivated further study into whether predictions from such geometry-based adaptation methods could be overconfident. In this paper, we build on the singular value decomposition factorization of adapters to develop a framework based on Stein variational gradient descent (SVGD). In this formulation, the low-rank matrices are transported along the Stiefel manifold to match the targeted distributions while retaining their crucial geometric structure. Since this geometry-aware SVGD approach provides multiple solutions during inference, it supports uncertainty quantification and produces better-calibrated adapters on the Stiefel manifold. Extensive experiments show that our method delivers strong model calibration and attains higher prediction accuracy than SVGD and related uncertainty estimation methods that are formulated in Euclidean space.
cs.LG / 42 / 2609.08375
IPM-FM: A Foundation Model with Consensus Feature Selection for Industrial Process Monitoring
Abstract
Industrial process monitoring is fundamental to the safety and economic performance of modern process plants. Current practice remains a one-task-one-model paradigm that is label-inefficient and prone to degradation under operating drift. Foundation models have reshaped language, vision, and generic time-series forecasting, but it has not been adapted to industrial process monitoring. This setting poses domain-specific challenges, including safety-critical decisions and asymmetric sampling between process variables and laboratory measurements. We propose the industrial process monitoring foundation model (IPM-FM). It first learns general-purpose representations from unlabeled industrial process data through self-supervised pretraining, then adapts to specific monitoring tasks using a small amount of task-labeled data, and finally produces calibrated predictions through an uncertainty-aware prediction head. IPM-FM integrates a self-supervised Informer backbone with a multi-criteria consensus feature selector, a recursive lag-feature regression head, and a calibrated Monte Carlo dropout uncertainty module. On a seven-year hydrotreater dataset for diesel flash-point soft sensing, IPM-FM attains an RMSE of 2.99, $R^2$ of 0.50, and 97\% coverage of its 95\% predictive interval, outperforming the strongest classical and from-scratch sequence baselines by 8.3\% and 14.6\% in RMSE respectively, supporting the viability of a unified pretraining--adaptation framework for industrial process monitoring.
cs.LG / 43 / 2609.08379
Geographically Regularized AUC-Maximizing Personalized Federated Learning
Abstract
Accurate diagnostic and risk-prediction models are important for supporting clinical decision-making during infectious disease outbreaks. However, privacy and governance requirements may restrict patient-level data sharing across healthcare institutions, and data distributions often vary. Moreover, AUC is widely used to evaluate discriminative performance, motivating its direct optimization in model development. We propose geographically regularized AUC-maximizing personalized federated learning (GrAUC-PFL), which directly optimizes a smooth pairwise AUC surrogate to learn personalized models while keeping patient-level data local and accounting for institutional heterogeneity. Graph-based regularization encourages geographically neighboring institutions to have similar coefficient vectors while retaining a personalized models. Simulations and a real-data application suggest improved discriminative performance, particularly when geographically neighboring institutions have similar data-generating characteristics.
cs.LG / 44 / 2609.08381
Equivariance Breaks the Learning Rate
Abstract
Equivariant networks are commonly trained with Adam, yet recent work reports that matrix-structured optimizers such as Muon can perform better on these architectures without explaining why. We identify one source of this difference inside equivariant linear layers. Each irrep block learns a channel-mixing matrix $W_l$ shared across its $2l+1$ components, giving the expanded map $W_l \otimes I_{2l+1}$. For a single application of the layer, the gradient of $W_l$ sums $2l+1$ outer product contributions and has rank at most $2l+1$. Adam rescales stored weights individually without using the irrep boundaries, so one learning rate can produce different spectral step sizes across blocks within a layer. We address this mismatch by normalizing each block update separately, without introducing a new hyperparameter. This changes only the scale of the update, leaving Adam's moment estimates and its direction within each block unchanged. We evaluate the mechanism in a controlled $\mathrm{SO}(3)$-equivariant model with a matched dense control and in an e3nn interatomic potential model trained on rMD17 and MD22. The toy setup isolates a mismatch that grows with width while the dense control shows no corresponding growth. In the interatomic potential model, block normalization and tuning Adam's momentum coefficients independently improve performance, but neither alone matches Muon. Combined, they make Adam competitive with Muon on all datasets, indicating that blockwise step control and momentum accumulation account for much of Muon's advantage.
cs.LG / 45 / 2609.08399
MLIP Detective: Active Failure Mode Discovery Beyond Benchmark Scores for Machine-Learning Interatomic Potentials
Abstract
Universal machine-learning interatomic potentials (u-MLIPs) aim to generalize across diverse configurations. Benchmarks enable reproducible evaluation but may not expose failures outside their predefined scope. Here, we show that physics-informed search can complement benchmark-based evaluation by uncovering hidden failure modes. We introduce MLIP Detective, an agentic framework for active failure mode discovery. Starting from benchmark evidence, MLIP Detective generates falsifiable, physics-informed failure hypotheses, screens them with inexpensive simulations, and escalates only the most suspicious cases to human experts together with proposed verification protocols. Without issue-specific prompting, MLIP Detective identified and characterized a systematic anomaly in MACE-MPA-0: the model predicted some relaxed adsorbate-surface systems involving O- or F-containing adsorbates to be higher in energy than their corresponding separated fragments. Using cross-model comparisons, MLIP Detective further inferred a likely training-data origin for the anomaly, consistent with recent reports.
cs.LG / 46 / 2609.08412
Stochastically Perturbed Weights: Ensembles from Deterministic Machine-Learning Weather Models
Abstract
Machine-learning weather models (MLWMs) now match or outperform operational numerical weather prediction (NWP) at global medium-range forecasting, at far lower inference cost. Many deployed MLWMs are deterministic, producing a single forecast with no estimate of its own uncertainty, whereas a growing family of trained-probabilistic models generate calibrated ensembles directly, at the price of a dedicated training run. We ask instead how much uncertainty can be extracted from a deterministic checkpoint that already exists, without retraining it. Where physical ensembles represent model uncertainty by stochastically perturbing parametrisation tendencies, we perturb the network's raw weight tensors at inference time, a scheme we call stochastically perturbed weights (SPW). We also ask whether it works, where and on which scales to inject the noise, and where it fails. A three-phase ablation across four deterministic backbones, Aurora, GraphCast, SFNO, and AIFS, selects one production baseline per model, benchmarked against the trained-probabilistic AIFS-ENS, FourCastNet 3 and Atlas as well as the operational ECMWF ensemble (IFS-ENS) over 112 initialisation times. At a 240 h (10-day) lead time the SPW ensembles reach continuous ranked probability skill scores (CRPSS) between 0.04 and 0.13 below the best trained-probabilistic baseline, at zero marginal training cost. No injection site works across models: the productive tensor group is architecture-specific, so SPW is at present a tuning procedure rather than a plug-and-play recipe. Its main failure mode is a coherent whole-field offset that overdisperses the domain mean, and restricting the noise to coarse scales or perturbing the initial conditions each repair part of it.
cs.LG / 47 / 2609.08445
Topological Fraud Detection in Latent Transaction Spaces
Abstract
Working entirely on topologically anonymized embeddings, we perform fraud detection using iterative rounds of unsupervised filtering followed by supervised sniping. The result is an ultra-low latency privacy--preserving triage that allows institutions to flag suspicious activity without compromising Personally Identifiable Information.
cs.LG / 48 / 2609.08554
Not All Variables Agree: Reliability-Aware Variable-Wise Gradient Surgery for Multivariate Time-Series Forecasting
Abstract
In data-driven training, multivariate time-series forecasting is usually optimized with a scalar loss averaged over samples, variables, and horizons. This averaging is convenient, but the optimizer sees only the aggregated gradient, which does not reveal whether the variable-wise contributions align or oppose one another. To quantify how often this disagreement arises, we measure the variable-wise gradients directly and find that 30.6% of their pairwise cosine similarities are negative on average across seven datasets. However, conflict and harm are not the same thing. Under shared training 35 of the 64 variables do worse than a full-input single-target oracle, and the harmed fraction is not reliably predicted by how often gradients conflict. We propose Per-Variable Surgery (PV-Surgery), an optimizer-side training strategy for backbones with cache-compatible layers. One backward pass builds variable-wise gradient proxies from output-side signals and keeps the pointwise forecasting loss. Reliability-aware selection targets layers whose proxy sums closely approximate their shared-gradient slices. Conditional pooling forms anchor and conflict pools without dropping variables. Common-direction surgery aligns variable or pooled gradients with their normalized mean and restores input norms to avoid reweighting. In experiments across five backbones, seven datasets, and four horizons, PV-Surgery lowers MSE by 3.61% and MAE by 2.93% on average. For multivariate forecasting, this indicates that the variable-wise structure hidden by mean-loss training is a usable optimization signal.
cs.LG / 49 / 2609.08561
Certified Topological Interaction in Neural Representations: Class Disentanglement Is Mostly Pairwise
Abstract
Class disentanglement (the separation of a representation's class-conditional point clouds along depth and over training) is usually read off descriptive curves. We measure it as certified topological interaction between labeled point clouds, using the recently introduced Intersection Euler Characteristic Profile: the Euler characteristic of the overlap of the clouds' ball unions as a function of scale, computed by one Alpha-complex sweep with no boundary-matrix reduction. Every number carries a test: exact permutation tests in both directions, a guarded separation certificate, and a paired test for the comparative claims applications make. Across 111 trained networks and 52,650 certified measurements, disentanglement is depth-graded and concentrated in the first epochs, and interaction quotients rank class pairs by confusability (Spearman rho=0.83), on par with cheap separability statistics. In a 96-model factorial population, augmentation is the one training choice that separates classes relative to chance; weight decay compresses the overlap without separating, and depth and width do nothing. The structural finding is one only a k-fold statistic can pose: the joint entanglement of a class triple sits below that of its strongest pair in 97% of triple-layer cells and 99.5% of deep cells, far below a measured null floor, in vision encoders and frozen language models alike. This pairwise dominance is a regularity, not a law: expected from the nesting of overlaps but not forced by geometry, present at initialization and in raw pixels, and manufactured in the last stage alone when a network memorizes random labels. The unnormalized profile mass predicts test accuracy (R^2=0.94), the quotient does not, and neither beats a linear probe. One lesson is reported in full: the paired test must use a scale-free statistic, or it certifies feature-norm dynamics as disentanglement.
cs.LG / 50 / 2609.08581
AlphaRJM: Reward-Jump Memory for Stochastic Return-Guided Alpha Discovery
Abstract
Formulaic alpha discovery is a pool-dependent symbolic search problem in which informative feedback is observed primarily when a complete expression is evaluated. This delayed feedback creates two coupled difficulties: the retained alpha pool does not preserve the full history of realized evaluation feedback, and the value of an intermediate construction action is uncertain because its consequence depends on the formula eventually completed. We introduce AlphaRJM, which addresses these difficulties through Reward-Jump Memory, an event-driven latent state that remains fixed during token construction and updates only at terminal evaluation events using the realized pool reward and evaluation outcome, and an action-conditioned SDE return critic that represents future discounted discovery returns with stochastic particles. The particles guide action selection through their mean and uncertainty and are learned using a distributional Bellman objective combining energy-distance matching, mean calibration, and jump regularization. Empirically, AlphaRJM delivers strong and stable gains across multiple equity universes, forecasting horizons, and random seeds, while ablations confirm the complementary roles of persistent evaluation history, stochastic return modeling, and distributional supervision.
cs.LG / 51 / 2609.08582
Leveraging Cardiac Imaging to Improve ECG-Based Detection of Chagas Disease in Resource-Constrained Settings
Abstract
Chagas disease is a major cause of cardiomyopathy in Latin America. Cardiac magnetic resonance (CMR) imaging can characterize its structural abnormalities, but scanners and expert readers remain scarce in endemic regions. Electrocardiography (ECG) is inexpensive and widely available, yet structural disease must be inferred indirectly from electrical signals. We propose to transfer CMR-derived structural knowledge to ECG through contrastive pre-training. Using 63,193 paired ECG-CMR examinations from the UK Biobank, we align an ECG encoder with a clinically grounded CMR embedding space using an asymmetric InfoNCE objective. Despite seeing no Chagas cases during pre-training, the resulting representation improves ECG-based Chagas detection. Across CODE-15% and SaMi-Trop, a frozen linear probe achieves an AUROC of 0.851 and sensitivity at the top 5% of predicted risk (Top5%-TPR) of 0.427 in five-fold cross-validation, compared with 0.827 and 0.377 for an unaligned ECG-FM baseline. On the PhysioNet/CinC 2025 Challenge test set, our model obtains the highest AUROC on SaMi-Trop-3 and the best ELSA-Brasil challenge score among the three top-performing methods, indicating that imaging-supervised ECG representations can generalize to populations and settings beyond the pre-training distribution.
cs.LG / 52 / 2609.08594
Multi-Level-Set-Based Physics-Driven Neural Network to Solve 3-D Inverse Scattering Problems
Abstract
This paper proposes a level-set-based physics-driven neural network solver (LSPDNN) for 3-D electromagnetic inverse scattering. To mitigate boundary blurring and reconstruction artifacts in voxel-wise contrast reconstruction, the proposed solver exploits the piecewise homogeneity of practical scatterers by representing unknown targets with multiple coordinate-dependent neural level-set components. Specifically, a soft-union multi-material model is proposed to separately describe the object support and material distribution. The global support is formed by the union of multiple level-set components, while the local contrast is determined by normalized component weights and learnable complex permittivity candidates. In addition, a model-consistent total variation (TV) regularization is imposed on the material-region indicators, rather than directly on the reconstructed contrast, to suppress fragmented material assignments without excessively smoothing material interfaces. An adaptive loss balancing strategy is further introduced to reduce the dependence on manually selected regularization weights. For each measurement instance, the neural level-set parameters and material candidates are optimized by minimizing a physics-consistent objective function. Numerical and experimental results demonstrate that LSPDNN can reconstruct scatterers with clear boundaries, more uniform material regions, and substantially reduced background artifacts. The results highlight the advantage of the neural level-set parameterization in challenging 3-D inverse scattering cases involving irregular shapes, closely spaced objects, multiple materials, and measurement noise.
cs.LG / 53 / 2609.08615
Why shared attention vectors fail: a case for outcome-indexed tuning
Abstract
Dimensional attention in learning is often implemented as a globally shared attention vector, where each stimulus dimension corresponds to a single scalar. These scalars are learned by models through gradient-descent on error, where predictive features acquire more salience. We show that under multi-outcome learning, where models predict more than one outcome, this shared vector becomes unstable; it collapses to its bounds and prevents the models from learning meaningful attentional tunings for learning and generalization. We address this by introducing an outcome-indexed attentional matrix that converts globally shared attentional tuning into an outcome-indexed representation. We present an analysis of the unstable shared vectors and derive the conditions under which it holds. Empirically, three synthetic experiments benchmark the proposed attention matrices and show that they converge to meaningful representations, something shared attention vectors fail to do. These results suggest that outcome-indexed attentional matrices are a general fix for gradient-based attentional processes, which improves models of learning under multi-outcome conditions.
cs.LG / 54 / 2609.08618
Target-Independent Micro-Interventions for Predicting Training Response Across Language-Model Families
Abstract
Benchmark scores describe what a checkpoint can do now, but they do not determine how it will respond to the next training episode. We measure this missing state by branching four short, standardized, target-independent micro-interventions from the same checkpoint and recording their effects in a common capability space. Together with current capability, these responses form L-State; its pulse block supports a flexible direct readout and a structure-preserving operator readout. Under smooth local dynamics, the operator construction admits an end-to-end cross-family bound with explicit source- and target-family coordinate heterogeneity. In three-family leave-one-family-out development, both pulse readouts reduce source-standardized MSE by 39.4% relative to capability alone, while separating the best response and direction estimates. On sealed GLM-4-9B, the direct and operator readouts reduce MSE by 71.8% and 78.3%, respectively, and the operator readout raises sign balanced accuracy from 0.366 to 0.754. On sealed Granite-3.1-8B, the direct readout reaches RMSE 0.544 and a development-fitted action-wise selector reaches 0.554, compared with 1.172 for capability alone. A five-family audit finds that the operator coordinate varies by action and family, and that modeling these deviations improves retrospective held-trajectory prediction. Target-independent interventions therefore expose training-response information that current capability misses, with direct and structured readouts covering complementary transfer regimes.
cs.LG / 55 / 2609.08622
Leveraging contextual events on structure-aware next activity prediction
Abstract
Predictive process monitoring aims at forecasting various aspects of running processes. Among the different tasks, next activity prediction represents the most extensively investigated. However, only a limited number of existing approaches explicitly encode contextual information, i.e., the environmental conditions in which the process is executed, typically modeled through event log attributes or aggregated measures. In this paper, an approach based on the concept of Instance Graphs is introduced. To incorporate contextual process instances, several encoding strategies are proposed and evaluated by measuring their impact on prediction performance. For each encoding strategy, a set of prefix-Instance Graphs is generated and subsequently provided as input to a Graph Neural Network for the classification task. The proposed approach is evaluated on multiple real-world event logs, and the experimental results demonstrate that incorporating contextual process instances benefits prediction performance.
cs.LG / 56 / 2609.08642
SUN: Reaching for Novelty in Reinforcement Learning
Abstract
Exploration in reinforcement learning (RL) remains a fundamental challenge. Recent goal-conditioned RL strategies (which select goals to encourage broader state coverage) have shown promising results, but none scores a goal by novelty and reachability jointly: the two signals are traded off by hand, applied in sequence, or one is neglected outright. In this paper, we introduce a reachability-aware goal-selection framework that explicitly integrates these two aspects, and that can be seamlessly incorporated into any off-policy RL algorithm. To this aim, we propose SUccessor-to-Novelty (SUN), an indicator derived from successor value functions to identify goals that are both novel and reachable. We prove that SUN recovers count-based bonuses in the limit, bounds short-horizon hitting probabilities, and provably rejects unreachable goals. We further present an adaptive goal-selection strategy that leverages these properties, and an accurate yet lightweight pseudocount to avoid the overhead of classic methods. We back up all our claims with thorough benchmarks: SUN consistently outperforms state-of-the-art methods in standard and novel environments with unreachable or hard-to-reach states, irreversible transitions, obstacles, mazes, and unbounded spaces.
cs.LG / 57 / 2609.08650
Difficulty-Adaptive Tree-Structured Policy Optimization for Expanding Reasoning Coverage in RLVR
Abstract
Reinforcement Learning with Verifiable Rewards (RLVR) has been central to the recent success of Large Reasoning Models. However, while RLVR significantly improves single-sample accuracy, it often fails to expand the model's intrinsic reasoning coverage (pass@k) due to limited exploration during training. To address this, we optimize the structural design of train-time rollouts to enhance pass@k. Our analysis identifies three key design principles: (1) difficulty-adaptive rollout can play an important role in expanding pass@k, beyond serving as an efficiency heuristic; (2) tree-based rollout outperforms parallel sampling in discovering correct answers; and (3) sentence-entropy-guided forking overcomes the localization phenomenon of token-level branching to maximize semantic diversity. Building on these insights, we propose DATPO (Difficulty-Adaptive Sentence-entropy-guided Tree-structured Policy Optimization). DATPO integrates difficulty-adaptive tree search with a sibling-diversity advantage term, explicitly promoting semantic diversity to expand reasoning coverage during training. Experiments on mathematical reasoning benchmarks demonstrate that DATPO outperforms baselines especially in pass@k, which directly translates to superior test-time scaling performance.
cs.LG / 58 / 2609.08663
MoEMB: Scaling Universal Multimodal Embeddings with Efficient Mixture-of-Experts Models
Abstract
Universal multimodal embedding (UME) increasingly demands encoder's capacity for handling a broad range of tasks and modalities with increased complexity. Prior scaling methods either increase the representation size, retrieval effort, or scales the encoder into a heavy multimodal LLM. Recent works, such as Think-Then-Embed (TTE), explore scaling via reasoning tokens. However, embedding models are hard to scale up: increasing parameters directly tradeoffs for the large training batch size that contrastive learning needs, and retrieval has to be served under tight latency. Moreover, UME tasks are diverse in complexity, where scaling up embedders can bring significant redundant computation. In this work, we propose MOEMB, which instead scales UME along the expert axis through mixture-of-experts (MoE), growing encoder capacity while preserving single-vector, non-autoregressive encoding. Through a systematic study of the design space and training recipes for MoE-based UME, MoEMB sets a new state of the art on both MMEB-V2 and MRMR among models trained on public MMEB-family data: with only 3B active parameters, MoEMB surpasses TTE-based methods with >4x active parameters, using significantly less computes. To further improve the scalability and efficiency, we conduct the first comprehensive study of adaptive computation for MoE-based embedding, spanning diverse strategies across training-based and inference-only methods. Together, these results support expert scaling as an effective and efficient direction for UME, with adaptive computation further improving efficiency for MLLM-based embedding models towards large-scale retrieval and recommendation systems.
cs.LG / 59 / 2609.08683
Neither Adversarial Training Nor Purification: Emergent Adversarial Robustness from Oscillatory Predictive Learning
Abstract
Adversarial robustness in computer vision is still largely achieved through adversarial training or test-time adversarial purification, both of which introduce significant computational overhead by generating adversarial examples during training or performing iterative denoising at test time. We study whether empirical robustness can instead emerge from architectural and representation-learning inductive biases. We introduce Oscillatory Predictive Learning (OPL), a two-stage framework that combines Artificial Kuramoto Oscillatory Neurons (AKOrN) with predictive self-supervised pretraining using X-PhiNet. Because our default checkpoint uses randomized initial oscillator states, we compare it with other randomized adversarial defense methods that provide precise, reproducible, and strong attack protocols. Experiments on CIFAR-10 and CIFAR-100, with additional corruption evaluation on CIFAR-10-C, demonstrate that our method achieves competitive results under the AutoAttack-rand evaluation protocol. On CIFAR-10 and CIFAR-100, OPL attains 76.63$\pm$0.76$\%$ and 50.44$\%$ robust accuracy, respectively, under $\ell_\infty$, $ε=8/255$, AutoAttack-rand with EoT $K=20$.
cs.LG / 60 / 2609.08685
HOPE: Heterophily-Aware Open-Set Node Classification with Pseudo-Extrapolation
Abstract
Standard open-set node classification methods rely on the homophily assumption, where connected nodes share labels. However, real-world graphs are often heterophilic, exposing the limitations of current methods and posing new challenges to open-set node classification. On the one hand, cross-class connectivity causes representations from different known or unknown classes to become intertwined after aggregation, undermining their discriminative capacity. On the other hand, structural mixture invalidates threshold-based open-set methods and cross-class feature interpolation, leading to unreliable unknown-class rejection. To address these challenges, we propose HOPE, a Heterophily-aware Open-set node classification method with Pseudo-Extrapolation. To adapt open-set graph neural networks (GNNs) to heterophilic scenarios, HOPE uses a structure-augmented feature initialization layer to capture multi-hop structural patterns. Meanwhile, we design a trustworthy neighborhood aggregation mechanism for standard GNNs to dynamically filter noisy cross-class neighbors. To enhance unknown-class rejection, we introduce a heterophily-guided pseudo-extrapolation strategy. It dynamically maintains known-class centers and extrapolates along cross-class neighborhood displacement directions, synthesizing pseudo-unknown proxies near structurally ambiguous regions. Finally, we optimize the network with joint classification and logit margin regularization, routing synthetic proxies into a dedicated rejection slot without imposing geometric margin constraints in the representation space. Extensive experiments on multiple datasets show that HOPE consistently outperforms state-of-the-art models, validating its effectiveness, robustness, and efficiency.
cs.LG / 61 / 2609.08690
Hyperparameter Scaling Laws Across MoE Sparsity
Abstract
Mixture-of-Experts (MoE) models expand model capacity without a proportional increase in training compute, but increasing sparsity makes reliable hyperparameter transfer challenging. In this work, we show that conventional hyperparameter scaling laws are insufficient for ultra-sparse MoEs: the optimal learning rate and batch size vary with activation ratio, and these shifts cannot be explained by either total or activated parameter count alone. To characterize this dependence, we conduct 1,800 pre-training runs spanning six activated-parameter scales and models with up to 6B total non-embedding parameters, processing approximately 20 trillion tokens at a cost of 200,000 equivalent H800 GPU-hours. Our results reconcile conflicting findings in prior work by revealing two scaling regimes. At fixed sparsity, the optimal batch size follows a power-law relationship with training tokens $D$, whereas the optimal learning rate scales with training compute $C$ and remains robust to the allocation between model size and data. Across sparsity levels, the activation ratio $A$ enters both relationships as an additional multiplicative power-law factor. These observations lead to unified hyperparameter scaling laws that transfer across MoE sparsity levels. Large-scale evaluation shows that the scaling form outperforms alternative functional forms. On a held-out ultra-sparse MoE with 12B total parameters and only 1/64 of its experts activated, the predicted hyperparameters remain close to the observed optima, supporting joint extrapolation across model scale and sparsity. Further experiments demonstrate transfer across expert granularities and isolate the effect of activation ratio from that of total expert count.
cs.LG / 62 / 2609.08725
BAFF: Bid-Aware Filter Family for Mitigating Training Data Interference in RTB A/B Tests
Abstract
In online A/B tests for real-time bidding (RTB), control and treatment models are typically trained on a shared serving log that includes data generated by the counterpart model. This shared-log training biases each model's training data through two channels: the counterpart model may have selected a different ad from the ad-candidate pool (ad-ranking disagreement) and may have bid a different price (bid-pricing disagreement), potentially distorting the A/B test outcome. Log-splitting eliminates the bias but sacrifices training data; log-sharing retains all data but leaves the bias unaddressed. We formalize the Bid-Aware Filter Family (BAFF), a class of (k,l)-parameterized hard filters that controls tolerance to each channel independently, providing a structured search space between these two extremes. We further propose a three-stage online measurement protocol that enables evaluating data-sharing strategies by their deviation from an interference-free reference model in production. In offline simulation, a (k,l) sweep surfaces operating points with smaller deviation from the interference-free reference model than both log-sharing and log-splitting. In a live RTB deployment on a demand-side platform (DSP), filter-based variants preserve the reference model's business metrics (e.g., CPC, CTR) more closely than both baselines. The best operating point is setting-dependent, underscoring the practical value of the search space itself.
cs.LG / 63 / 2609.08740
PAC-Bayesian Bounds for Learning Partially Observed Stochastic Linear Time-Invariant State-Space Systems with Inputs and Sub-Gaussian Noise
Abstract
In this paper we derive a Probably Approximately Correct (PAC)-Bayesian error bound for partially observed linear time-invariant (LTI) stochastic dynamical systems in state-space form with inputs and sub-Gaussian noise. Such bounds are widespread in machine learning, and they are useful for characterizing the predictive power of models learned from finitely many data points. The bound derived in this paper relates the expectation of prediction errors with the prediction error generated by the model on the data used for learning. In addition, we show that it can also be used to derive bounds for the parameter estimation error. In turn, this allows us to provide finite-sample error bounds for the prediction error and parameter estimation error for a wide class of system identification algorithms. Furthermore, as LTI systems are a sub-class of recurrent neural networks (RNNs), these error bounds could be a first step towards PAC-Bayesian bounds for RNNs.
cs.LG / 64 / 2609.08788
Adaptive Anisotropic Attention for Axis-Structured Signals
Abstract
Dense self-attention treats all token pairs as equally plausible before learning, an interaction-isotropic prior that can be mismatched to structured signals. For structured, low signal-to-noise ratio (SNR) signals such as EEG, dependencies are organized along the electrode and time axes, and this uniform prior exposes each token to many irrelevant interactions. We introduce Adaptive Anisotropic Attention (AAA), which splits attention into two paths: a temporal path, where each token attends to the tokens of its own electrode across time, and a spatial path, where it attends to the tokens of the other electrodes at the same time step. A small gate predicts, for every token, a convex combination of the two path outputs: two non-negative weights that sum to one. On six EEG downstream tasks, the resulting model, AXON (AXis-factorized Operator Network), improves mean balanced accuracy over a dense baseline under both linear probing and full fine-tuning. We show that both paths (temporal and spatial) are necessary and that the weighted sum beats a hard choice of one path; most of the benefit comes from the gate learning a different temporal/spatial balance at each layer of the network. Controlled audio spectrogram experiments show that axis factorization transfers beyond EEG. These results suggest that aligning attention with the natural axes of structured signals provides a useful inductive bias.
cs.LG / 65 / 2609.08798
Eliciting Weak-to-Strong Generalization with On-Policy Reverse Distillation
Abstract
Weak-to-strong generalization asks whether stronger models can learn from weaker supervisors and surpass them. This question is particularly important for successive model generations and multi-domain consolidation, where repeating frontier-scale post-training from scratch can be prohibitively expensive. Yet conventional distillation treats the weak teacher as an optimization target, potentially imposing its capacity ceiling on the student. We introduce On-Policy Reverse Distillation (OPRD), which evaluates the teacher's policy shift relative to its reference policy on student rollouts and amplifies the component of the student's verifier-driven policy gradient along that direction. By rescaling only verifier-supported updates, OPRD preserves the stationary points of policy optimization while accelerating learning beyond the teacher. In both successive model transfer and multi-teacher distillation, OPRD achieves higher performance with fewer student updates than existing RL and distillation approaches. Response-style analysis shows that OPRD students remain closer to models trained with verifier-based RL alone than to their weak teachers, suggesting that teacher guidance accelerates rather than redirects the student's own optimization. Results in conventional strong-to-weak distillation further demonstrate that OPRD effectively combines verifier-driven policy optimization with teacher guidance regardless of capacity ordering.
cs.LG / 66 / 2609.08851
Length Generalization for Transformers via Compression
Abstract
Recent advancements in transformer length generalization theory enable us to reliably predict when a transformer can learn to solve a task. In particular, the C-RASP hypothesis (a formalized version of the so-called RASP-l conjecture) posits that transformers length-generalize on a task if and only if a solution is expressible in the C-RASP language. While this hypothesis has strong empirical validation, theoretical problems arise from the fact that no computable length generalization bounds exist for C-RASP, alongside the discovery of seemingly contradictory experiments. To address these problems, we refine the C-RASP hypothesis utilizing the recently-proposed fragments C-RASP+ and C-RASP1. These fragments have computable length generalization bounds, though in the worst case requiring an extremely large (double exponential) sample size. It is an open question whether these sample size bounds are tight. In this paper, we resolve this open question by providing an exponentially tighter bound. In doing so, we show a polynomial length generalization bound for transformers if we adopt compressed strings, via a novel connection to power words. As an application, we show how this yields a fine-grained analysis of the C-RASP conjecture that resolves contradicting experimental evidence against it.
cs.LG / 67 / 2609.08855
Earth System World Model for What-If Simulations: A Case Study for Terrestrial Ecosystems
Abstract
Machine learning emulators have become essential for accelerating expensive Earth-system simulations, but most existing approaches remain passive forecasters: they reproduce simulator trajectories under prescribed forcings without an explicit interaction mechanism for user-specified interventions. This limits their use in interactive scientific workflows and Earth-system digital twins, where users often need to explore how a system would respond if selected state components were changed. We propose an action-conditioned world-modeling framework for Earth-system emulation that reformulates simulator trajectories as supervision for controllable state-transition learning. The key idea is transition-action pretraining: naturally observed state changes are treated as label-free action supervision, allowing the model to learn both prescribed dynamics and action-conditioned responses without manually annotated interventions. We further introduce masked response learning to infer unobserved variables under partial state edits and learn coupled system dependencies. We test this framework on ecosystem dynamics across six global regions and multiple stand ages. Experiments show that the model preserves competitive long-horizon emulation accuracy while enabling controllable structural interventions and coherent responses in coupled ecosystem-cycle variables. These results suggest a practical route from passive Earth-system emulators toward interactive, intervention-aware scientific surrogates.
cs.LG / 68 / 2609.08901
The BatchNorm Illusion: Diagnosing Normalization Artifacts in Machine Unlearning Evaluation
Abstract
Approximate machine unlearning aims to remove the influence of specific training data from a trained model without retraining from scratch. We identify a previously undocumented confound in how unlearning is evaluated on BatchNorm-based architectures: a single forward pass over retain data, an operation that modifies no weight, can deterministically rewrite the model's normalization state and reverse the apparent surface-metric forgetting. We formalize this operation as a weight-preserving fixed-point operator and prove that any pre-versus-post gap it induces is provably attributable to BN running statistics rather than to any modification the unlearning method made to the weights. This attribution claim cleanly separates measurement failure (BN artifact) from encoder failure (residual weight-encoded information, recently documented in concurrent work), and the same operator framework yields a unique decomposition of linear-probe elevation into BN-measurement-bias and encoder-geometry components. Empirically, the artifact reverses headline forget accuracy by up to 78 pp across nine evaluated methods on standard benchmarks; an attacker with as few as 10 unlabeled images recovers most of the masked accuracy; and a strict GroupNorm control reduces the artifact to zero across all methods. The tested membership-inference attacks change little under recalibration, locating the observed evaluation failure in forget accuracy and linear probing.
cs.LG / 69 / 2609.08970
GraphFAS: A Distributed System for Automated Graph Feature Generation and Selection in Industrial Transaction Networks
Abstract
Industrial fraud detection often relies on costly expert-crafted features that overlook graph-structured relational signals, while GNNs often do not meet the interpretability and deployment requirements of financial risk control. We propose GraphFAS (Graph Feature Automated Selection), a distributed feature selection procedure based on Boruta that bridges this gap through: (1) a non-parametric graph feature generation module that constructs explicit, interpretable structural features via multi-hop subgraph extraction and multi-scale aggregation without learned parameters; and (2) an automated distributed feature selection algorithm extending Boruta with median-based aggregation across partitions to robustly identify informative features at scale with minimal domain expertise. Compared with end-to-end GNN pipelines, GraphFAS decouples feature aggregation from model training, enabling direct integration with tabular models and direct compatibility with TreeSHAPbased explanations. Deployed in Alipay, GraphFAS delivers orderof-magnitude improvements in engineering efficiency while showing strong performance against expert-driven and graph-learning baselines on large-scale graphs.
cs.LG / 70 / 2609.08992
Physics-Informed Deep Learning for False Ventricular Tachycardia Alarm Reduction in the ICU
Abstract
False ventricular tachycardia (VT) alarms are a leading contributor to alarm fatigue in intensive care units. We propose a deep learning framework combining a 1D SE-ResNet with ICU-realistic data augmentations and a physics-informed auxiliary reconstruction task based on the three-element Windkessel hemodynamic model, implemented as a differentiable forward simulation. By requiring the network's latent representation to produce physiologically plausible arterial pressure waveforms, artifact-driven ECG patterns are penalized while true VT remains coherent across modalities. Evaluated on the VTaC benchmark under a strict real-time protocol (10-second pre-alarm window), our method achieves a 5-point Challenge Score improvement over prior state-of-the-art. Ablation studies confirm that the physics-informed objective is the primary performance driver, providing gains in accuracy, 2x label efficiency, and more localized and clinically meaningful ECG segments.
cs.LG / 71 / 2609.09062
Multi-Task Learning for Sparsely-Labeled Time Series: A Case Study on Cold-Hardiness Modeling
Abstract
We present a real-world case study of multi-task learning (MTL) for temporal process modeling from limited data with temporally sparse labels. Specifically, we investigate multi-task learning for the important agricultural problem of predicting grape cold hardiness, which is the temperature at which lethal freezing occurs. Cold hardiness changes in response to weather and is difficult to measure directly in the field. Thus, growers rely on predictions to decide when to apply costly frost mitigation measures. We apply recurrent neural networks (RNNs) for daily cold-hardiness prediction from time series weather data. A major challenge is that the cold hardiness response varies across plant cultivars and ground-truth data for each cultivar is temporally sparse and limited. To address this challenge, we investigate multi-task learning (MTL) approaches for combining data, where different tasks correspond to different cultivars. We develop a variety of MTL architectures and evaluate them in both MTL and transfer learning settings. Our results show significant differences between architectures and that certain architectures are able to consistently outperform single-task learning and state-of-the-art scientific models. Additionally, we show similar results for the qualitatively different, but related, task of budbreak prediction. Further, improved accuracy for budbreak and cold hardiness is achieved by a single MTL model that simultaneously learns both tasks.
cs.LG / 72 / 2609.09075
ThinkPrior: Zero-Rollout Difficulty Priors for Cold-Start Prompt Selection in RLVR
Abstract
In reinforcement learning with verifiable rewards (RLVR) trained with group relative policy optimization (GRPO), the KL-free reward-advantage term studied here depends on within-group reward variation. If all rollouts in a group are correct or all are wrong, their group-relative advantages are identically zero; these zero-advantage silent groups provide no reward-advantage gradient, yet uniform sampling spends 39% of a run's rollouts on them. History-based prompt selection must first spend target-policy rollouts to estimate difficulty, creating a cold start with rollout waste; ThinkPrior instead uses an external anchor in one offline pass to construct a zero-rollout difficulty prior before the first target-policy rollout. The verifier-scored anchor pass rate supplies an external-anchor initialization for a Beta posterior; ThinkPrior selects by expected learnability and then updates from training outcomes, changing neither the loss nor the optimizer. On Qwen2.5-Math-7B across sixteen seeds, ThinkPrior more than halves early silent groups and cuts wasted rollouts through step 30 by nearly a fifth, while we detect no difference in final accuracy. On this 250-prompt pool the fixed-budget result is a reallocation rather than a net saving. The measured ThinkPrior+DAPO composition reduces generated rollouts by 10.6% while both arms retain the same 3840-rollout update budget. The prior requires no target-policy rollout before the first selection, but the posterior thereafter uses target-policy outcomes.
cs.LG / 73 / 2609.09099
Curriculum Learning as Transport: Understanding Curricula with Wasserstein Geodesics
Abstract
Curriculum learning is governed by several coupled design choices---how difficulty is defined, how examples are ordered, how much exposure each level receives, and how quickly training moves across levels---making it hard to isolate what actually helps. We present Wasserstein curriculum paths, a simple transport-based framework that decouples these factors by representing curricula as trajectories of training distributions over discrete difficulty levels. Across a calibrated synthetic suite with 12 tasks and 33 difficulty axes, we use this framework to isolate the effects of ordering, matched exposure, endpoint smoothness, and pacing under fixed training budgets. We find that curriculum effects are strongly context-dependent: no single strategy dominates across tasks, difficulty axes, and budgets, and curricula mainly change where a fixed budget is spent most effectively. Within this framework, easy-to-hard ordering improves hard-level performance relative to exposure-matched static sampling, showing that the benefit is not explained by cumulative exposure alone. We further show that endpoint smoothness and pacing substantially affect where along the difficulty spectrum a curriculum is effective. Finally, we show that the same transport view naturally supports extensions to learned pacing through geometry and to structured difficulty spaces beyond one-dimensional orderings.
cs.LG / 74 / 2609.09116
When Does Scale-Invariant Optimization Become Unstable? An Exact Schedule Law with Weight Decay
Abstract
Normalization renders large parts of neural networks effectively scale invariant, inducing a hidden feedback loop in which learning-rate schedules and weight decay interact through the parameter norm to control the effective step taken by the optimizer. We show that this interaction is governed by an exact discrete-time law: a single scalar quantity captures all schedule and decay forcing, while norm growth induces an opposing geometric self-quenching effect. This yields a sharp boundary that cleanly separates contraction- and expansion-dominated effective learning rate regimes. To understand the underlying mechanism, we provide exact analysis of a fully solved normalized regression model where the dynamics reduce to two dimensions and show that the balance point is intrinsically unstable, implying that constant learning rate with weight decay cannot stably maintain an interior equilibrium and instead produces recurrent behavior driven by discrete-time Jacobian structure. We further extend this perspective across optimizers through unified homogeneous-optimizer framework that reveals a structural dichotomy in self-quenching strength, providing a first-principles explanation for why adaptive methods exhibit systematically weaker stabilization under normalization. Across dynamical systems and neural networks (MLP, CNN, GPT2 / MNIST, CIFAR, wikiText, OpenWebText), the predicted law holds with high precision and enables direct control of training via the identified scalar, with performance peaking sharply at the predicted boundary. Together, these results isolate a single governing quantity for scale-invariant optimization, providing a precise and actionable lens on training dynamics, optimizer behavior, and schedule design in modern deep learning. Code is available in https://github.com/shasanamin/normalized-optimization-dynamics.
cs.LG / 75 / 2609.09130
Nearly Tight Rademacher Bounds for Sparsely Activated Neural Networks
Abstract
An input may activate few hidden units even when different inputs collectively use an entire network. We study the statistical complexity of this input-dependent sparsity in the one-hidden-layer ReLU model of Awasthi et al. (COLT 2024). For width $s$, at most $k$ active units per input, and effective weight and bias bounds $W,B$, every size-$m$ sample in the class's fixed radius-$R$ input domain satisfies $\mathcal{R}(S)\le CWR\min\{k,\sqrt{sk/m}\log^{3/2}(2m)\}+kB/\sqrt m$. A support-preserving cover and a single normalized chaining argument remove the previous explicit dimension factor, up to logarithms. Lower bounds on appropriate i.i.d. marginals match up to those logarithms, showing how changing active units across inputs retains a width dependence. The input domain matters: zero-bias networks sparse on the entire ball have at most $2k$ nonzero units and complexity $O(kWR/\sqrt m)$, whereas bias bounds comparable to $WR$ restore the worst-case rate on that same domain in only logarithmic dimension. A spherical-cap construction proves the latter claim without assuming sparsity merely on the sampling support. For a specified normalized bounded loss and biases comparable to $WR$, we also obtain agnostic minimax excess-risk bounds of order $\min\{1,\sqrt{s/(km)}\}$ up to logarithms.
cs.LG / 76 / 2609.09135
Entropy-Regularized Rank-Masked Policy Optimization for Test-Time Reinforcement Learning in Code Generation
Abstract
Existing methods for test-time reinforcement learning (TTRL) derive rewards from answer-level self-voting on unlabeled test-time tasks with canonical answers, but this breaks down for code generation because programs cannot be compared by surface form and therefore do not directly provide a usable training signal. To make TTRL applicable to code generation, we propose probe-driven TTRL, which constructs output-free probe inputs from the problem statement, executes candidate programs on these probes, and defines a Probe Consensus Reward (PCR) from the resulting behavioral agreement. PCR provides a behavioral training signal for open-vocabulary programs, but it is not a fully reliable verifier and remains susceptible to reward hacking through spurious consensus. We therefore introduce Entropy-Regularized Rank-Masked Policy Optimization (ERPO), which converts low PCR into conservative negative updates through rank masking and controls policy drift with an entropy ceiling. On coding benchmarks, ERPO substantially improves pass@1 and pass@k in both in-domain adaptation and zero-shot transfer.
cs.LG / 77 / 2609.09140
NOAH: Learning the Full Patient Journey. A Longitudinal Multimodal Time-Aware Model for Representation and Forecasting
Abstract
The digitization of healthcare has generated vast, longitudinal, and multimodal patient records over a lifetime, yet fully exploiting these data to represent and predict patient state trajectories remains a critical challenge. Current AI models often struggle to capture the complex, irregular temporal dynamics and inherent stochasticity of real-world multimodal patient data. Existing AI approaches for modeling longitudinal patient records are predominantly discriminative, limited to a few modalities, constrained by closed categorical vocabularies, treating time as a monotonic inductive bias, or they are limited in forecasting future patient states. We introduce NOAH, a time-aware, task-agnostic, generative transformer model representing and forecasting the full multimodal patient journey. NOAH features a novel bidirectional time integration and a variational latent space to capture the continuous evolution of patient states and the stochasticity of clinical trajectories. Built from over 559 million clinical events from 431,000 hospital visits of 299,000 patients across the MIMIC dataset family, NOAH natively processes medical images, time-series and numeric signals, categorical events, as well as structured and unstructured clinical records. NOAH is the first truly holistic generative model in its field, enabling autoregressive forecasting with optional time control, zero-shot classification, and counterfactual intervention simulation. It generates highly informative and predictive patient state representations that demonstrate strong performance in probing for clinical outcomes, 15 ICD chapters, and 29 comorbidities, as well as in time-to-event prediction. Seamlessly handling diverse modalities and complex temporal dynamics, NOAH provides a versatile, task-agnostic, scalable foundation for intelligent predictive systems in personalized clinical care and digital medicine.
cs.LG / 78 / 2609.09157
Learning Length-Extrapolatable Recurrent Models
Abstract
Recurrent models provide a natural path to long-context modeling, yet models trained with backpropagation through time (BPTT) often fail beyond their training horizon. Classical analyses emphasize gradients that vanish or explode along temporal paths. However, dense per-token losses can still train a shared recurrent rule despite severe decay, showing that decay alone does not determine whether learning fails. We instead study state credit: the signal through which future losses reach earlier recurrent states before contributing to parameter updates. Accordingly, we intervene directly on state credit and propose Credit Stabilization through Time (CST). During backward propagation, CST locally rescales the state-credit signal to stabilize its norm without rotating the component being corrected, while leaving the forward computation unchanged. Because controlled synthetic tasks and real data exhibit different credit dynamics, we specialize CST to each regime. In both settings, CST improves performance beyond the training horizon, with gains observed at up to 128x the training length.
cs.LG / 79 / 2609.08961
Fitting and Learning Basis-Restricted Propositional Formulas
Abstract
For a finite set $O$ of Boolean functions, we consider the class of propositional formulas built using the functions in $O$ as connectives. We determine, for each possible choice of $O$, the complexity of various fitting and learning problems. These include: finding a formula that fits a given labeled sample, finding a small one (an Occam algorithm), minimizing the number of misclassified examples when the sample is not realizable (empirical risk minimization), and several forms of PAC learning. Our results apply both to formulas (represented as trees) and to circuits. We also briefly discuss the status of the same questions for other kinds of propositional fragments.
cs.LG / 80 / 2609.07843
The OCUDU dApp Platform: An Open Runtime and E3 Interface for Real-Time AI-RAN
Abstract
Machine learning has shown its largest gains in the band below 10 ms inside a 3GPP new radio (NR) 5G distributed unit (DU): link adaptation, per-slot scheduling, channel estimation, and the receiver itself. No open platform has let independently built software run there. Prior dApp frameworks reached the band only as external observers of an export stream. This paper is a guided introduction to the OCUDU dApp platform, an open runtime and E3 interface under which signed AI-RAN applications execute inside a production DU under three timing contracts: resident on the GPU receive chain (Class A), inside the scheduler's 100 us admitted deadline (Class B), or as never-blocking observers whose results the scheduler consumes (Class C). The conventional path is never displaced, and every authority is typed, validated, and operator-bounded. The paper explains how the runtime, the embedded E3 agent, and the three public repositories fit together; shows a dApp's source, its signed package, and its lifecycle state machine; defines the contracts a module is written against; and shows how one management surface serves a Python script, an operator's console, and an LLM agent. On a GB10 gNB with attached handsets, dApps of all three classes, including an out-of-tree neural equalizer, ran together on a live cell without a single fallback, and equalizer variants were compared over the air by lifecycle operations alone. Every measured checkpoint is reported with its conditions and its gaps. Platform, SDK, and a zero-hardware quickstart are public under BSD-3-Clause-Clear as a preview release of the OCUDU AI-RAN Working Group 2, inviting feedback, new use cases, and independent vetting ahead of upstreaming into the OCUDU mainline.
cs.LG / 81 / 2609.08669
Learning to build covering structures with continuous adjustments
Abstract
Robotic construction offers the potential to use materials more efficiently and create complex geometries, but current methods rely on rigid, high-precision plans that cannot accommodate the tolerances, inaccuracies, and unexpected changes inherent in physical fabrication. In this work, we introduce a reinforcement learning approach that forgoes predefined plans entirely, instead generating construction sequences adaptively as the structure is built. Our method operates on graph-structured state representations and a mixed (parameterized) action space, requiring both discrete block selection and continuous placement parameters. Because the stability simulation of a structure is computationally heavy, we develop an efficient exploration strategy by incorporating unilateral edges into graph neural networks, extending soft actor-critic (SAC) to this hybrid setting. We evaluate our algorithm, HSAC, against the prior method hybrid-PPO (HPPO), demonstrating significantly higher asymptotic performance and good sample efficiency. We also demonstrate HSAC's robustness to hyperparameter choices and its exploration capability, handling up to 10 discrete actions without performance degradation. Finally, we validate our approach on a physical two-robot setup, successfully building a spanning arch with 3D-printed blocks in closed-loop execution, confirming that policies trained in simulation transfer to real hardware.
cs.LG / 82 / 2609.08800
Ostrich: Taking Large Strides Through Stiff Contact in Differentiable Dynamics
Abstract
Three properties determine whether a differentiable simulator can drive gradient-based optimization through contact: simulation accuracy, gradient reliability, and per-iteration cost. Tape-based engines such as MJX and Newton Semi-Implicit require timesteps small enough to keep contacts numerically tractable, and their backpropagation memory grows linearly with the number of timesteps T. Surrogate models bound memory by approximating contact away, but the resulting gradients lose the geometry the optimization depends on. We present Ostrich, a GPU-accelerated rigid-body simulator that resolves hard contacts and friction with non-smooth Newton iteration at large timesteps (h ~ 0.1 s), and differentiates the converged residual via the implicit function theorem, reusing the forward Schur complement to compute the adjoint at O(1) memory per timestep. On real-robot trajectories over a pallet obstacle, Ostrich holds MuJoCo's sim-to-real accuracy up to a 50x larger timestep. Its gradients converge from random initializations where MJX descends slowly and Newton Semi-Implicit stalls; a warm iteration runs 211x faster than MJX's and 4.7x faster than Semi-Implicit's. On the same scene Ostrich differentiates 8,192 parallel worlds on a single 24 GB GPU, sustaining 29x checkpointed MJX's optimization throughput; without checkpointing both baselines exhaust memory at far fewer worlds. We close with a gradient-based trajectory optimization demonstration over triangle-mesh terrain across a 10 s horizon, a setting where prior engines either restrict to primitive geometry or face the convergence and memory limits shown above.
cs.LG / 83 / 2609.07981
Clean Accuracy Does Not Guarantee Provenance Robustness: A Prospective Codec-Stress Evaluation of Audio Attribution
Abstract
Audio provenance attribution - which system produced a synthetic utterance - is reported at near-ceiling accuracy on clean benchmarks, yet audio reaching an analyst has usually been transcoded. We report a prospectively registered measurement of closed-set attribution after single-stage codec transport, with the analysis region fixed from fidelity metadata before any attribution model was trained. On two corpora, in-support losses reach 53.5 [43.5, 63.6] and 70.3 [63.0, 77.5] Macro-F1 points for WavLM-Base+, and 61.0 [56.8, 65.1] and 49.8 [41.6, 57.9] for W2V2-BERT 2.0, under simultaneous component-level bands. Degradation is strongly condition- and representation-dependent: within one in-support grid WavLM losses run from -0.4 to +53.5 points, and the two encoders differ beyond a prespecified +/-5-point margin at six of twelve conditions. A clean-qualified ECAPA-TDNN and a Proxy-Anchor head degrade comparably, so the effect is not confined to one representation family or a weak linear head. The registered matched-fidelity comparison was not estimable on this grid, and waveform and perceptual measures order the conditions differently: MP3 at 8 kbit/s ranks mid-grid on SI-SDR but last on PESQ-WB while causing the largest loss. For the tested tasks, corpora, representations and codec grid, a clean accuracy figure does not by itself characterise deployment robustness.
cs.LG / 84 / 2609.08593
FedGenSC: Federated Generative Semantic Communication with Channel-Aware Adaptation
Abstract
Integrating generative adversarial networks (GANs) into federated semantic communication (SemCom) is a natural progression, as generative priors can recover semantic fidelity under channel distortion that discriminative decoders cannot. However, naive GAN federation introduces three failure modes that prior work has, to the best of our knowledge, neither identified nor resolved: discriminator aggregation instability under non-independent and identically distributed (non-IID) data, semantic drift caused by divergent local embedding spaces, and channel-agnostic generation that cannot adapt to heterogeneous link conditions. We propose federated generative semantic communication (FedGenSC), which mitigates all three by employing a global generator with local-only discriminators, providing cross-client semantic information through a semantic prototype bank, and conditioning generation on the instantaneous signal-to-noise ratio (SNR). Experiments on the Europarl dataset over Rayleigh fading channels (K=10 clients, Dirichlet α=0.5) show that FedGenSC under non-IID data outperforms the FedDeepSC baseline across the tested SNR range, achieving up to a 58.2% relative improvement in bilingual evaluation understudy (BLEU)-1 at 18 dB. Ablation studies confirm the independent contribution of each component.
cs.LG / 85 / 2609.08463
Inclusive electron-nucleus cross section models from domain adaptation
Abstract
We apply transfer learning (TL) to construct data-driven models of inclusive electron-nucleus cross sections. Starting from an ensemble of deep neural networks pretrained on \(^{12}\)C data, we fine-tune the models separately for \(^{3}\)He, \(^{6}\)Li, \(^{16}\)O, \(^{27}\)Al, \(^{40}\)Ca, and \(^{56}\)Fe. The resulting models improve for all targets, marginally so for oxygen, where the carbon baseline is already adequate, although their predictive robustness depends on the amount, coverage, and precision of the available target data. We systematically study how model performance depends on the number of fine-tuned layers, on the fraction and selection of the training data, and on the overlap between the source and target kinematic domains. The layer-wise analysis shows that oxygen requires only shallow adaptation, whereas helium, calcium, and iron require substantially deeper fine-tuning. Lithium represents the least robust case because of its limited dataset, while aluminum demonstrates a strong sensitivity to a small subset of highly constraining measurements. For selected kinematic configurations outside the coverage of the carbon training data, the adapted models remain consistent with the measurements within their estimated uncertainties. Finally, we compare the resulting predictions with those of the phenomenological F1F2 model.
cs.LG / 86 / 2609.08461
Selective boundary condition reduction via learned error gating
Abstract
Parametric PDEs can admit different boundary conditions with different accuracy and computational cost. We introduce a framework for learning when one reduced boundary condition can replace another: paired solutions train a neural network to estimate the resulting domain and boundary errors, and the simpler condition is used only when both predicted errors meet prescribed tolerances. We focus on singular limits in applications, in which a stiff Robin or nonlinear boundary law is replaced by its limiting Dirichlet form. We evaluate the method on a galvanic corrosion problem and other nonlinear stationary and evolution problems.
cs.LG / 87 / 2609.08380
How to Make the Gradient Mapping Small for Constrained Stochastic Min-Max Problems and Beyond
Abstract
We study the stochastic first-order oracle complexity for constrained or regularized convex-concave min-max optimization and stochastic monotone variational inequalities. We focus on the case when suboptimality is measured in terms of the gradient mapping, also known as, forward-backward or natural residual, an optimality notion that generalizes the gradient norm for unconstrained problems. In this setting, under standard unbiased oracle access with now-standard variance assumptions, the best-known complexity for making the norm of the gradient mapping less than $\varepsilon$ is $\widetilde{O}(\varepsilon^{-4})$, compared to the near-optimal $\widetilde{O}(\varepsilon^{-2})$ that is established in the unconstrained case. We bridge this gap to improve the gradient mapping complexity for constrained convex-concave min-max problems to $\widetilde{O}(\varepsilon^{-2})$. We then extend to prove the same complexity for problems without the bounded variance, by using the Blum-Gladyshev assumption.
cs.LG / 88 / 2609.08537
The Exact Time-Uniform Rate Frontier for Stochastic Gradient Descent on Smooth Convex Objectives
Abstract
We study the time-uniform convergence of the raw iterate of standard stochastic gradient descent (SGD) for unconstrained smooth convex objectives. We prove that, under standard noise assumptions, the time-uniform convergence rate gets arbitrarily close to $\sqrt{\log n / n}$ but never reaches it. More specifically, we prove that for every positive, eventually nondecreasing sequence $h$ satisfying $h(n) = o(\sqrt{n})$, a bound of order $h(n)/\sqrt{n}$, holding simultaneously for all $n$ with probability at least $1-α$ and uniformly over the problem class, is achievable if and only if \[ \sum_{j = 1}^{\infty} \frac{1}{h(2^j)^2} < \infty. \] The constructive sufficiency result follows from a dyadic horizon-free schedule together with an additive conditional-restart inequality. The necessity counterpart applies to every deterministic nonnegative schedule and holds even for a one-dimensional analytic smooth convex objective with Gaussian noise.
cs.LG / 89 / 2609.09152
Silver Rate Is (Almost) Optimal for Gradient Descent Acceleration
Abstract
We study how far gradient descent (GD) can be accelerated by predetermined nonnegative stepsizes in smooth convex optimization. Writing $p_{\mathrm{sil}}=\log_2(1+\sqrt{2})$, we prove an $Ω\left(n^{-p_{\mathrm{sil}}-O(\sqrt{\log\log n/\log n})}\right)$ non-anytime lower bound. In the anytime setting, every infinite nonnegative schedule has infinitely many horizons with error $Ω\left(n^{-\frac{2p_{\mathrm{sil}}}{1+p_{\mathrm{sil}}}-O(\sqrt{\log\log n/\log n})}\right)$. Together with the silver-schedule upper bound [Altschuler and Parrilo, 2025] and the anytime upper bound [Zhang et al., 2025], our results determine the optimal polynomial convergence exponents in both settings.
cs.LG / 90 / 2609.08620
Flexible Spectral-Normalized Neural Gaussian Process for Dynamic Aperture Prediction
Abstract
We address the challenge of scalable uncertainty quantification in large-scale scientific applications, where complex state-of-the-art machine learning methods are often computationally infeasible. Our primary contribution is a simple yet effective empirical Bayes method for automatically tuning the hyperparameters of a flexible, heteroscedastic Spectral-normalized Neural Gaussian Process. This approach retains the expressiveness and uncertainty-awareness of semi-Bayesian neural models while significantly reducing the computational burden by integrating hyperparameter learning directly into the training loop. We demonstrate the practical impact of our method on the task of estimating the dynamic aperture in circular particle accelerators, a fundamental problem in high-energy physics colliders and storage rings, using simulation data from the case of the Large Hadron Collider at CERN. Traditional approaches to DA estimation require extensive particle-tracking simulations, which are prohibitively time-consuming and resource-intensive. Our results show that the proposed method achieves competitive predictive performance and well-calibrated uncertainty estimates at much lower computational cost than state-of-the-art approaches. We stress that, beyond this application, the proposed empirical Bayes framework offers a general solution for training heteroscedastic neural models in situations where manual hyperparameter tuning is impractical. Accordingly, we anticipate that this framework can be applied to other domains that encounter comparable computational limitations.
cs.LG / 91 / 2609.08333
Fixed-Dimensional Latent Flow for Generating Variable-Size 3D Molecules
Abstract
In molecular discovery, molecule size is coupled to composition, structure, and other target properties. Yet most 3D generators require molecule size to be specified before generation. Here, we introduce Equivariant-Free Transformer-Autoencoded Latent Flow Matching, a two-stage generative framework that relies entirely on a single fixed-dimensional molecule-level latent representation to generate variable-size molecules. The second-stage flow matching model samples this latent vector, and an autoregressive Transformer decoder then determines molecule size while generating atom types, coordinates, and chemically informative states. Canonical atom ordering and rigid-pose alignment enable standard Transformers without equivariant layers, while joint decoding of molecular geometry and an enriched chemical state enables reliable, deterministic, chemistry-guided graph recovery without requiring a learned dense pairwise bond decoder. The same fixed-dimensional latent supports unconditional and property-conditioned flow matching, while optional property supervision adds an internal ranking readout, with no separate predictor or reference calculations. On PCQM4Mv2, EF-TALFM achieves the highest fraction of molecules that are unique, training-set novel, pass sanitization and PoseBusters sanity checks, 89.4\%, compared with 75.6\% for UAE-3D and 69.8\% for FlowMol. EF-TALFM also achieves higher measured computational throughput for training and sampling. Across ten target HOMO--LUMO gaps, internal ranking doubles the density functional theory (DFT)-verified hit rate within $0.1\,\mathrm{eV}$, while preserving 97\% novelty among unique verified hits. These results demonstrate that fixed-dimensional molecule-level generation followed by symmetry-resolved autoregressive realization provides a practical architecture for open-ended and property-directed 3D molecular design.
cs.LG / 92 / 2609.08947
ONE CYLinder: A Benchmark for Graph-Based Surrogate Modeling of Unsteady Bluff-Body Flows
Abstract
Graph-based surrogate models offer a promising route to accelerate computational fluid dynamics (CFD) simulations on unstructured meshes. However, their development is limited by the scarcity of benchmark datasets spanning multiple flow regimes and standardized protocols for long-horizon autoregressive prediction. We introduce ONECYL (ONE CYLinder), a new benchmark for unsteady flow past a circular cylinder across laminar, transitional, and high-Reynolds-number regimes. The benchmark comprises 450 high-fidelity Variational Multiscale finite-element simulations (270,000 flow snapshots) with randomized cylinder geometries, providing time-resolved velocity and pressure fields together with mesh connectivity, geometric descriptors, Reynolds numbers, and integrated aerodynamic quantities. Beyond the dataset, ONECYL establishes a unified evaluation framework combining full-field rollout errors, virtual probes, and drag and lift predictions to assess numerical accuracy and physical fidelity. To accompany the benchmark, we develop a Graph Transformer as a reference baseline predicting velocity and pressure fields autoregressively on unstructured meshes. Using ONECYL, we investigate geometric representations and physics-based regularization across the three Reynolds-number regimes. The results show that explicitly encoding the cylinder geometry through a level-set representation consistently improves long-horizon prediction accuracy and generalization to unseen geometries, while divergence-based regularization becomes increasingly beneficial as flow complexity increases. The ONECYL benchmark and its Graph Transformer baseline provide a reproducible framework for evaluating graph-based surrogate models and establish a foundation for future research on long-horizon prediction of unsteady bluff-body flows.
cs.LG / 93 / 2609.08165
A Transformer-Based Delta Expression Encoder for Psilocybin Transcriptional Response: Architecture, Representations, and Biological Validation
Abstract
Understanding why individuals respond differently to psilocybin requires modeling the drug's transcriptional perturbation signature at the cell-type level. I present a Transformer-based delta expression encoder that learns to classify differential gene expression status - upregulated, downregulated, or neutral - from single-nucleus RNA-sequencing data, without supervision from pathway annotations or prior biological knowledge. The model is trained on pseudobulk profiles from 623 examples spanning 18 cell types, 2 drug conditions, and 6 timepoints derived from the Liao et al. 2025 dataset, and achieves 69.4% weighted classification accuracy. Three principal findings are reported, alongside one direct test of a published hypothesis that returned a result inconsistent with that hypothesis. First, per-cell-type classification accuracy ranges from 28.3% (L2/3 IT, a primary HTR2A-expressing psilocybin target) to 99.6% (endothelial cells), consistent with known psilocybin response biology. Second, psilocybin-induced transcriptional downregulation is significantly more stereotyped across individuals than upregulation (Mann-Whitney U=18615.0, p<0.0001), a novel finding with a cortical depth gradient across excitatory subtypes. Third, attention-guided gene co-regulation analysis recovers drug-specific modules without pathway supervision. Separately, a direct test of whether baseline HTR2A expression predicts drug-response separability across cell types found a significant negative correlation (Spearman r = -0.7088, p = 0.0021), the opposite of what a simple HTR2A-gating account would predict.
cs.LG / 94 / 2609.08059
MI-PEFT: Mixture-of-Experts Integrated Parameter-Efficient Fine-Tuning Protein Language Models Improves Acidophilic Proteins Classification
Abstract
Acidophilic proteins that remain stable and functional under highly acidic conditions, are important for industrial biocatalysis, acid-related bioprocessing, and the discovery of acid-stable enzymes. However, their identification relies heavily on time-consuming experimental screening methods. With the rapid growth of protein sequence databases, the need for computational identification methods that are both accurate and efficient has become stronger. The emergence of protein language models (PLMs) has significantly improved the sequence representation of downstream biological prediction tasks. This paper proposes MI-PEFT, a mixture-of-experts integrated parameter-efficient fine-tuning framework. Built on the ESM C-600M backbone, the framework incorporates LoRA-based PEFT methods and a DeepSeekMoE-based classification head to resolve the limitations of PEFT and significantly improve computational efficiency. Notably, this task is characterized by a significant class imbalance in the dataset, making high specificity particularly challenging. The experimental results demonstrate that MI-PEFT on PLMs, especially {\text{C}}^{\text{3}}\text{A}, serves as an efficient tool for identifying acidophilic proteins and a constrained pathway that helps resolve class-imbalance by preserving the pretrained representations.
cs.LG / 95 / 2609.08172
Optimal Slice-Adaptive Tuning of Hybrid Slice Sampling
Abstract
Slice sampling is a Markov chain Monte Carlo algorithm that draws its next state uniformly from a "slice"---a super-level set of the target density function---at each iteration, thereby providing automatic local adaptivity to the scale of the target. In practice the exact slice is not known, so general-purpose implementations use an approximate slice that is grown from a starting interval of length $w>0$, with a computational cost that depends on $w$. This work presents an analysis of the average per-iteration number of target density evaluations, as a function of $w$, of hybrid slice sampling with various slice-finding schemes for targets with contiguous slices. The paper uses the results of the analysis to develop automated, slice-adaptive tuning schemes along with suboptimality bounds and asymptotic convergence guarantees. Simulations demonstrate that the tuning schemes reliably yield near-optimal slice-adaptive tuning with essentially no dependence on the initial setting of $w$.
cs.LG / 96 / 2609.08055
Bayesian Matrix-Valued Graphs for Context-Dependent Multivariate Relationships
Abstract
Many scientific graphs attach several variables to each node, so a single scalar edge weight cannot describe direction-dependent interactions. We model each edge by a symmetric positive-definite (SPD) matrix and infer a posterior over matrix-valued graph geometries, which we call the Bayesian matrix-valued graph (BMVG). We ask how these interactions reconfigure across contexts: how large the change is and which multivariate directions strengthen or weaken. The geodesic distance induced by the affine-invariant Riemannian metric (AIRM) quantifies deformation magnitude and generalized eigenvalues resolve its signed directions.Against fused graphical lasso, Bayesian multiple-GGM, and common principal components, BMVG is competitive on global precision recovery while retaining identifiable matrix-valued edge structure and accurately recovering edge-level deformation directions. In controlled known-truth experiments, it resolves structural change with increasing sample size, including orientation changes that leave ordinary eigenvalues unchanged. In one year of Bay Area weather data, the geometry of 12-hour change reconfigures spatial coupling about as much as whole seasons differ. In TCGA-BRCA, estrogen-receptor (ER)-associated reconfiguration concentrates on specific gene-module pairs and persists under graph-scaffold sparsification and removal of subgroup mean differences. These results establish posterior matrix-valued edge geometry as a unified framework for quantifying and interpreting context-dependent multivariate reconfiguration.
cs.LG / 97 / 2609.07868
SGD in Multiclass Logistic Regression: Sequential Learning and Scaling Laws
Abstract
We study the training dynamics of multiclass logistic regression on high-dimensional Gaussian mixture models with a large number of classes and establish precise scaling laws governing the cross-entropy risk under gradient-based optimization. We show that learning proceeds sequentially across classes, from most to least frequent. When the class priors follow a power law distribution, the risk dynamics decompose into three phases: an initial plateau until the first class is learned, a power-law decay regime during which sequential learning occurs, and a final convergence regime. We then analyze how model capacity interacts with optimization under a fixed compute budget. When the effective dimension is restricted via projection onto leading principal components, the risk decomposes into a capacity term (a power law in the retained dimension) and an optimization term (a power law in training time). Optimizing this tradeoff yields a compute-optimal scaling law for logistic regression, with explicit prescriptions for model size and training time as functions of compute. These results extend theoretical scaling laws from linear regression to multiclass classification, while connecting to empirical scaling laws observed in large-scale neural networks.
cs.LG / 98 / 2609.07959
The Role of Uncertainty in Assessing the Fairness of Machine Learning Models
Abstract
Machine learning models are widely used in clinical applications, social media, law enforcement and critical infrastructure. Verifying whether their outputs are biased against disadvantaged groups or individuals is crucial to ensuring they are fair and allowing their use in such settings. A rigorous risk assessment of possible fairness violations requires quantifying the uncertainty associated with selecting and estimating such models. Yet, this is rarely done in the literature, which focuses on identifying a single model with a suitable trade-off between predictive accuracy and fairness. In this paper, we move beyond point estimation and discuss frequentist and Bayesian approaches to uncertainty quantification for fair machine learning, with practical examples and implications for simulated and real data.
cs.LG / 99 / 2609.08234
Distribution-free inference on the number of changepoints
Abstract
Suppose we are given an ordered sequence of independent data whose distribution changes $K$ times at unknown locations, for some unknown $K \geq 0$. In this paper, we study the problem of performing distribution-free inference on $K$. First, we show an impossibility result: any distribution-free upper confidence bound on $K$ must be trivial and uninformative. Then, using conformal $p$-values, and under only the assumption that the data segments induced by the changepoints are exchangeable (within themselves) and mutually independent, we construct a finite-sample valid lower confidence bound on $K$, which we call the Conformal LOwer bound on Changepoint Count (CLOCC). We show that CLOCC is the only feasible way to provide a lower bound on $K$ under the stated assumptions, a property we refer to as its universality. We provide practical guidelines for choosing score functions that yield efficient and tight lower bounds. We evaluate CLOCC in several synthetic and real-data experiments, where it provides informative lower bounds on $K$, demonstrating its practical applicability.
cs.LG / 100 / 2609.08564
Non-Adaptive 1-Bit Mean Estimation: Minimax Rates and the Sample-Interval Tradeoff
Abstract
We study distributed one-dimensional mean estimation under a 1-bit communication constraint. Each agent observes one sample, drawn independently from an unknown distribution, and returns a single bit in response to a query $Q: \mathbb{R}\to\{0,1\}$ chosen by a central learner. The distribution has mean in $[-λ,λ]$ and $k$-th central moment at most $σ^k$, for a fixed $k>1$. The order-optimal two-stage protocol of Lau and Scarlett uses responses from the first batch to choose the second-batch queries, motivating the question of whether this single round of interaction is necessary. We answer this negatively: for every $k>1$, a non-adaptive protocol attains the adaptive 1-bit minimax rate (and concurrent works reached the same conclusion via different strategies). We further determine the minimax sample complexity among non-adaptive 1-bit estimators when every one-set $Q^{-1}(1)$ is restricted to a union of at most $s$ intervals. Relative to unrestricted non-adaptive 1-bit querying, this constraint adds a term of order $(λσ/(s\varepsilon^2))\log(1/δ)$, giving the full tradeoff between sample complexity and interval complexity to within $k$-dependent constant factors. As a corollary, we identify, order-wise, the minimum interval budget needed to retain the unrestricted 1-bit minimax sample rate.
cs.LG / 101 / 2609.08671
Optimal estimation for Functional Linear Regression with Noisy Discretized Data
Abstract
In this paper, we consider the scalar-on-function linear regression model under a realistic sampling scheme in which the functional covariates are observed on a regular grid and contaminated by additive noise. We propose a two-step estimation procedure: first, the underlying curves are reconstructed from the discrete noisy observations using a Fourier-based projection method; second, the slope function is estimated by a penalized least-squares criterion over finite-dimensional trigonometric spaces, with data-driven selection of the model dimension. We establish oracle-type inequalities for the prediction error, both with respect to the reconstructed curves and to the true latent curves. Under regularity assumptions on the slope function and polynomial decay of the eigenvalues of the covariate, we derive convergence rates for the prediction error and show that our estimator attains the minimax rate when the number of grid points is sufficiently large. Finally, the proposed method is illustrated on simulated data and on a real meteorological dataset.
神经与进化计算 (cs.NE)
1
cs.NE / 1 / 2609.08070
A Gradient-based yet Spike-Timing-Dependent Solution to the Feedback Learning Problem in Neural Microcircuits
Abstract
The brain uses discrete spikes for dynamic computation, yet, how neural microcircuits (NMCs) solve temporal credit assignment using local spike timing remains a fundamental open question. Dominant spiking neural network (SNN) approaches circumvent this by approximating backpropagation through surrogate gradients, decoupling learning from biological spike timing. Here, we reformulate temporal credit assignment as a state separation problem: extracting task-required components induced by historical perturbations directly from the current neural state. This enables an online feedback learning framework for NMCs through a gradient tunneling (GT) algorithm and the lead-lag expansion technique that derives credit assignment from local synaptic spike timing, while remaining compatible with ANN-SNN hybrid architectures. Experimentally, GT-trained NMCs excel at long-timescale evidence integration and noise-robust memory retention, and perform comparably to leading SNN online learning methods on real-world benchmarks with far fewer parameters. The proposed framework addresses the two-decade-old NMC feedback learning problem and suggests a computationally plausible explanation for the brain's learning mechanisms.
计算语言学 (cs.CL)
39
cs.CL / 1 / 2609.07876
LLM Layers Immediately Correct Each Other
Abstract
Recent methods in language model interpretability employ techniques such as sparse autoencoders to decompose residual stream contributions into linear, semantically meaningful features. Such methods are commonly interpreted as identifying features that persist in the residual stream and that subsequent layers build upon. We challenge this view by identifying the Transformer Layer Correction Mechanism (TLCM), wherein adjacent transformer layers systematically counteract portions of each other's contributions. TLCM appears in 5 out of 7 major open-source model families and activates across nearly all tokens in diverse texts. We show that TLCM emerges during pretraining, operates most strongly on contextually dependent tokens, and adaptively calibrates its correction strength based on the preceding layer's output. Using the layer Jacobian, we further show that TLCM selectively corrects specific subspaces while reinforcing others, which we interpret through a ``propose-and-reject'' framework in which layers propose candidate features and subsequent layers selectively remove inappropriate ones. This dynamic suggests that the residual stream at any layer contains transient proposals alongside persistent features, helping explain why SAE feature descriptions often have low specificity, why effective model steering requires extreme feature amplification, and why transcoders hold a theoretical advantage over SAEs.
cs.CL / 2 / 2609.07965
Rethinking Sign Language Translation: The Impact of Signer Dependence on Model Evaluation
Abstract
Sign Language Translation has advanced with deep learning, yet evaluations remain largely signer-dependent, with overlapping signers across train/dev/test. This raises concerns about whether models truly generalise or instead rely on signer-specific regularities. We conduct signer-fold cross-validation on GFSLT-VLP, GASLT, and SignCL, three leading, publicly available, gloss-free SLT models, on CSL-Daily and PHOENIX14T. Under signer-independent evaluation, performance drops sharply: on PHOENIX14T, GFSLT-VLP falls from BLEU-4 21.44 to 3.59 and ROUGE-L 42.49 to 11.89; GASLT from 15.74 to 8.26; and SignCL from 22.74 to 3.66. We also observe that in CSL-Daily many target sentences are performed by multiple signers, so common splits can place identical sentences in both training and test, inflating absolute scores by rewarding recall of recurring sentences rather than genuine generalisation. These findings indicate that signer-dependent evaluation can substantially overestimate SLT capability. We recommend: (1) adopting signer-independent protocols to ensure generalisation to unseen signers; (2) restructuring datasets to include explicit signer-independent, sentence-disjoint splits for consistent benchmarking; and (3) reporting both signer-dependent and signer-independent results together with train-test sentence overlap to improve transparency and comparability.
cs.CL / 3 / 2609.07968
Reasoning Beyond Transcription: Audio Language Models on Child Stuttering Speech
Abstract
Child speech differs from adult speech in acoustics, prosody, and linguistic structures. Speech disfluencies (such as repetitions) further challenge automatic understanding. While Audio Language Models (ALMs) show strong semantic reasoning from speech audio, their ability to reason about disfluent child speech in mixed-speaker settings remains unexplored. We investigate this through two tasks: child-focused semantic summarization and speech entailment. Experiments use recordings of children who stutter in mixed speaker interviews without explicit speaker separation. Models are instruction-guided to focus on the child, preserve clinically relevant disfluencies, and avoid adult-speech leakage. Evaluation combines LLM-based judges and reference-based metrics, anchored by transcript-oracle baselines to isolate errors. Results show that while ALMs extract high-level meaning from stuttered speech, reasoning degrades significantly with increased
cs.CL / 4 / 2609.07971
MeRoTune: RoPE-Safe Merging with a Tunable Dial
Abstract
When you merge two fine-tuned models from the same base checkpoint by simply averaging their weights, you implicitly assume their attention subspaces are still aligned. Recent work attempts to fix misalignments by learning an invertible correction matrix, $M$, for each model's query and key projections. This correction cancels out---using $M$ on the query side and $M^{-T}$ on the key side---right before the dot product. However, this cancellation is only exact if nothing sits between the projection and the dot product. In reality, almost all modern open-weight language models put a rotary position embedding (RoPE) exactly there. In this paper, we show that this cancellation is exact under RoPE if and only if $M$ commutes with RoPE's per-position rotation. We derive the specific class of matrices where this holds: a scaled rotation acting independently within each RoPE frequency pair. This forms a strict, low-dimensional subset of the unconstrained matrices that current methods normally train. Building on this, we turn this constrained matrix class into a new merging method. While keeping the base weights entirely frozen, two fine-tunes each learn their own RoPE-compliant correction matrices. We optimize these corrections against a chosen blend ratio so the final result can be adjusted post-hoc like a dial, rather than locked into a single fixed merge. Our default approach trains at one fixed blend ratio, similar to how LoRA sets its scaling hyperparameter in advance. We also experiment with resampling the blend ratio randomly at every training step, and we report the results of both approaches.
cs.CL / 5 / 2609.08029
BanglaMemeX: Advancing Cultural Metaphoric Image Interpretation in Bangla with a Multimodal Explainable Dataset
Abstract
Vision Language Models have achieved strong performance on multimodal benchmarks, yet their ability to reason about culturally grounded and metaphor-rich content remains insufficiently studied. Internet memes present a challenging setting where meaning emerges from implicit interactions between image, overlaid text, sarcasm, and shared socio-cultural knowledge rather than literal visual recognition. This challenge is amplified in low-resource languages such as Bangla, where code-mixing, stylized scripts, and culturally specific symbolism introduce substantial distribution shift. In this work, we introduce BanglaMemeX, a culturally grounded multimodal benchmark comprising 3,000 Bangla memes annotated with multi-dimensional labels (humor, sarcasm, offensiveness, motivational intent, and overall sentiment) and human-written explanations that explicitly describe textual and visual metaphors. We systematically evaluate modern VLMs on both classification and explanation generation, revealing that current models struggle to interpret implicit cultural cues despite reasonable surface-level accuracy. Our results highlight the need for culturally-aware multimodal systems capable of grounded reasoning under linguistic and cultural distribution shift.
cs.CL / 6 / 2609.08067
Popular Knowledge Propagates More Errors in LLM Knowledge Updating
Abstract
Updating a language model's knowledge through fine-tuning is essential for keeping its outputs current, yet can also induce factual forgetting and new hallucinations. Prior work shows that long-tail knowledge is harder to acquire and newly memorized long-tail facts are difficult to retain during later fine-tuning. We study a complementary question: among facts that a model has encoded correctly, which are most vulnerable to collateral corruption during other updates? To investigate this question under a realistic factual distribution, we construct a large-scale graph FACTPROP of verified Wikipedia facts by linking triples that share head or tail entities, thereby preserving connections among factual knowledge. We fine-tune models on factual statements and measure correct-to-incorrect facts after each update. Our results reveal a pattern distinct from prior findings on long-tail vulnerability during acquisition and retention: among facts that models already answer correctly, those associated with highly connected entities are more likely to be corrupted by neighboring updates, and updates to such facts propagate errors more broadly. Structural popularity therefore predicts both vulnerability and downstream damage. Inspired by this finding, we propose Popularity-based Anchoring (PopAnchor), a lightweight rehearsal strategy that preserves a small set of popular facts and reduces forgetting.
cs.CL / 7 / 2609.08088
Vectorizer: Vectorizing NumPy Programs with Shape-Guided Rewrite
Abstract
NumPy is a widely used Python library for numerical scientific computing, known for its declarative APIs and its optimized implementations. However, writing efficient NumPy programs, which often entails using vectorized array operations instead of explicit Python loops, may not be straightforward. This can be difficult for programmers who are accustomed to imperative array traversal, especially when vectorized API invocations require careful reasoning about shapes, broadcasting, and advanced indexing. This paper presents a rewrite-based approach for vectorizing Numpy programs with explicit loops over array data. Our approach vectorizes loops from the inside out, using array shapes and dataflow analysis to guide a source-to-source transformation that replaces loop bodies with vectorized statements. Following a set of rewrite rules that are correct by construction, our approach is consistently fast. We have implemented the approach as a tool called Vectorizer and evaluated it on 150 benchmarks collected from prior work and Stack Overflow. The evaluation shows that Vectorizer vectorizes 142 of the 150 benchmarks directly and 2 more after minor changes to the original benchmarks, with only 0.53 seconds on average to rewrite each one. The resulting programs are, on average, 74.83x faster than the original loop-based implementations.
cs.CL / 8 / 2609.08139
IGT @ FinMMEval 2026 Task 2: Question-Type Prompting with Targeted Extraction for Multilingual Financial QA
Abstract
We present the IGT system for PolyFiQA Task 2 of the FinMMEval Lab at CLEF 2026, a multilingual financial question answering task over English SEC filings and multilingual news articles (English, Chinese, Japanese, Spanish, Greek) for four companies. Our central observation is that the 344 development questions divide into two families requiring fundamentally different approaches: structured numeric types (R&D ratio, cash flow, capital expenditure) are best answered by direct keyword extraction on filing text, while synthesis types (investment strategy, capital allocation, top-three revenue focuses) require rule-based multilingual news passage selection. A dataset analysis reveals that 17-18 of 19 ground-truth reference answers per synthesis type share an exact evidence label prefix, whose unigram tokens contribute directly to ROUGE-1 overlap. The final system achieves development ROUGE-1 approximately 0.395, a 60% relative improvement over a generic RAG baseline (approximately 0.247), and ranks 3rd of 12 teams on the official test set with ROUGE-1 = 0.3071, Precision = 0.2821, and Recall = 0.4044.
cs.CL / 9 / 2609.08147
ConversationalVoice: Full-Duplex Speech Data from Real Conversations through Source-Faithful Reconstruction and Conversation-Grounded Expansion
Abstract
Full-duplex speech models require training data that preserves turn-taking, overlap, interruption, and backchannel behavior, yet these signals are entangled across speakers in noisy real-world recordings. We present Conversational Voice, a pipeline that converts real two-speaker excerpts into three complementary training-data artifacts. (1) Separation recovers speaker-specific tracks with stable speaker assignments, a canonical transcript, and naturally observed interaction timing. (2) Reconstruction generates speech in matched voices from a fixed source transcript, reconstructs the source turn order, pauses, and overlaps, and adds word-level alignment and delivery instructions. (3) Expansion generates new dialogue constrained by the source context, speakers, and observed interaction pattern. Automatic speaker-verification metrics remain strong across stages, with same-speaker similarity of 0.983-0.991 and positive discrimination margins of 0.199-0.209. Predicted speech quality (NISQA MOS) is 3.56 for separation, 4.41 for reconstruction, and 4.61 for expansion. A Gemini-based automatic evaluator assigns expansion mean scores of 4.94/5 for contextual coherence and 4.80/5 for dialogue naturalness. Expansion and reconstruction exhibit broadly similar interaction profiles; expansion's turn, overlap-event, backchannel, and interruption rates are 4.6%, 8.0%, 13.2%, and 16.0% lower, respectively. We evaluate data properties only; downstream gains in full-duplex model training remain for future work.
cs.CL / 10 / 2609.08161
Snugi-AI-v2 @ eRisk 2026 Task 2: Early Depression Detection via a Learned Stopping Policy with Sustained Confidence Gate
Abstract
We describe the Snugi-AI-v2 submission to eRisk 2026 Task 2, the second edition of contextualized early depression detection from Reddit discussions. Our central contribution is a learned MLP stopping policy trained to directly optimize ERDE50, replacing the fixed and tiered threshold strategies used in all prior eRisk Task 2 submissions. Combined with a sustained confidence gate that commits only after N=3 consecutive rounds of high policy confidence, the system reduces false positives caused by transient emotional posts without sacrificing recall. The pipeline encodes each discussion thread with a frozen MentalRoBERTa model, maps the accumulated representation to a depression probability via an MLP classifier, and delegates the timing decision to the learned policy. Our best run achieves F1 = 0.73 (Run 1) and F_latency = 0.70 (Runs 0 and 3), with a median alert round of 8 out of 500, completing the full evaluation in 1 hour 26 minutes, the fastest among all complete-submission teams. We report a systematic ablation across five runs spanning two encoder variants, four stopping strategies, and three gate values, along with negative results from GRPO policy training, BDI-II post filtering, MentalLongformer encoding, and DeBERTa ensembling. Code: https://github.com/chiuyuwen91/erisk-2026
cs.CL / 11 / 2609.08183
NeoHorse-1: Towards Recursive Self-Improvement via Agentic Post-Training with Routing Harness
Abstract
Recursive self-improvement (RSI) requires a concrete mechanism through which an AI system observes its capabilities and converts that evidence into the next round of learning. We present NeoHorse-1, a family of agent-native models developed to explore this path through agentic post-training. Our system combines a heterogeneous model pool with intelligent routing, recording the predicted capability demand, selected service tier, and subsequent interaction for each user turn. These records are converted into training examples that preserve interleaved reasoning, tool calls, and harness context, and are admitted through structural validation, six-dimensional semantic evaluation, and subscene-level labeling. Routing signals organize supervised fine-tuning into a three-stage curriculum and extend to routing-guided on-policy distillation, where a teacher supervises student-generated responses under the same progression. Capability-guided allocation then converts evaluation feedback into the next training mixture, closing an evaluation-selection-update loop in which what the system learns to do shapes what it learns from next. Across eleven benchmarks covering harness-based agents, tool use, coding, and instruction following, post-training raises the macro-average from 58.94 to 64.87 at 4B and from 65.60 to 69.04 at 9B, substantially narrowing the aggregate gap between the post-trained 4B model and the 9B base model. NeoHorse-1 provides an initial prototype of this feedback-driven process and a path toward harness-mediated RSI across successive iterations.
cs.CL / 12 / 2609.08279
What Eviction Destroys: A Restore-Counterfactual Audit of Forgetting in Agent Memory
Abstract
Agent memory systems must discard stored information when their history exceeds a fixed token budget. Existing budget-accuracy frontiers quantify the resulting loss in accuracy, but do not distinguish irreversible losses caused by eviction from recoverable retrieval failures. We introduce the restore counterfactual, a per-question paired intervention that reinstates the question's gold evidence in the read-time context and reruns the same reader. Combining the change in correctness with whether the evidence was retained after eviction classifies each oracle-answerable error as recoverable, irreversible, or residual; in the residual case, the answer remains incorrect after restoration. We evaluate FIFO, random, redundancy-aware, and LLM-importance eviction on LongMemEval-S at three budgets and under two retrieval regimes, using GPT-4o-mini as the primary reader and judge and GPT-5.4-mini as a robustness reader. Under top-k retrieval at an 80k-token budget, the irreversible share among errors corrected by restoration is 0.67-0.73 for FIFO, random, and redundancy-aware eviction, compared with 0.60 for LLM-importance. At 8k tokens, it reaches 1.00 for all four policies. Recoverable errors occur under top-k retrieval at 80k tokens but are absent under forced-gold injection by construction, so budget-accuracy results are not directly comparable unless the retrieval regime is reported. An exploratory matched-accuracy analysis detects no difference in irreversible rate among accuracy-matched policy pairs at a resolution of 1.2-6 percentage points. The same analysis detects the deliberately destructive control. To our knowledge, this is the first per-item, per-question restore-counterfactual audit of eviction for external agent-memory stores on a standard conversational benchmark.
cs.CL / 13 / 2609.08322
Tracing Stereotypes from Representation to Output in Multilingual LLMs
Abstract
Multilingual LLMs show stereotype-related behavior that varies across languages, but behavioral scores do not show where the relevant information is represented or how it affects the output. To investigate these internal mechanisms, we compare linear probing, attribution patching, sparse autoencoders (SAEs) and feature ablation in Llama-3.1-8B, Qwen3-8B, and Gemma-2-9B. Probe performance peaks substantially earlier than attribution in all three models, with a separation of 36-53% of model depth. Retained Llama-Scope features often match the social category on which they were selected and form recurring semantic families, but their lexical alignment and ablation effects vary across SAE suites. Only 6-18% of evaluated residual-stream features have language-agnostic effects under our criterion, and none are category-agnostic. Language-agnostic features have larger mean ablation effects in Llama-Scope, but this pattern does not repeat in the other SAE suites. Decodability, output influence, and cross-lingual ablation effects therefore need to be measured separately.
cs.CL / 14 / 2609.08364
SentryLine: Evidence-Grounded Question Answering over Evolving Documents in Oncology Care
Abstract
Oncology care operates at constant pressure of absorbing rapidly evolving evidence base in biomedicine. The American Society of Clinical Oncology (ASCO) addresses this through living guidelines, but the format introduces a new burden: any recommendation can change at any point, across multiple versioned documents. We present SENTRYLINE, a living guideline-aware clinical question answering system. SENTRYLINE retrieves guideline passages through a vectorless hierarchical RAG pipeline and returns a role-specific answer with inline citations, factual and temporal verification reports, and drift detection notes that surface when a guideline has been updated. We construct ASCOBENCH, a benchmark of 405 three-turn conversations across four question categories with gold answers from expert annotators(clinicians), and use test set to evaluate SENTRYLINE against five baselines under an LLM-as-judge framework. Experiments across three generation backbones show consistent improvements over four retrieval baselines and ASCO's guideline assistant, with particularly strong gains on Reasoning and Role-Specific questions where multi-hop synthesis and register adaptation are required
cs.CL / 15 / 2609.08372
Reading a Legal Question Word by Word: Embedding Trajectories of 2,144 Vietnamese Legal Headlines
Abstract
A dense retriever encodes a question as one vector, but the question arrives one word at a time. We read 2,144 held-out headlines from Thu Vien Phap Luat (Vietnamese legal library) word by word with Nemotron-3-Embed 8B/1B and Qwen3-Embedding 8B/0.6B, encoding 65,444 prefixes against 20,034 articles, plus every prefix of 3,438 sub-questions from 1,112 multi-question headlines and of 168 answers. (i) The gold article becomes rank 1 after a median of 6-7 content words in every encoder, before the interrogative frame is read, and stays there to the end in 78-85% of cases. (ii) In a multi-question headline the lock is inside the first sub-question 94-98% of the time; the second leaves rank unchanged in 89-95%; encoded alone, the second reaches rank 1 in 42-58% vs 91-96% for the first, at the same lock word (95-97% identical). (iii) Numbers, dates and instrument identifiers move the embedding twice as far as content words and four times as far as interrogative words; 72-78% of steps move toward the gold article, and the closing interrogative frame moves against that direction in 95-99% of headlines. (iv) Rank/cosine clustering yields six archetypes (instant, typical, unstable, late, never-locking) that differ by legal area and form (chi-squared p < 1e-8): real-estate and litigation headlines never lock on a number; environmental and accounting headlines do so a third of the time. (v) An answer read word by word retrieves its article after 8-16 words and addresses the sub-questions in order asked in 83-89% of cases. (vi) A word's step keeps a consistent direction across headlines (cosine 0.25-0.33; 0.44-0.60 for numbers); a preceding question rotates that step by about 60 degrees and a greeting by about 30 degrees; steps shrink as i^{-0.8}; and a two-question headline is within 12-17 degrees of a linear mix of its two questions. We call this a context-modulated additive walk.
cs.CL / 16 / 2609.08373
Structural Jailbreaks Generalize but Do Not Compound: A cross-provider and multilingual study of Involuntary In-Context Learning
Abstract
Aligned language models fail under two independent pressures: the structural jailbreak class recently formalized as Involuntary In-Context Learning (IICL), which reframes a harmful request as the final missing cell of a data-labeling task completed by pattern rather than judged as content; and the erosion of safety alignment outside English. A natural hypothesis is that these compound. We test it directly. Using a deterministic IICL operator and a StrongREJECT-style rubric judge, we red-team two Google Gemini models on two benchmarks, a 30 general-harm behaviours from HarmBench and 30 financial-abuse behaviours from FinProof, each under a single-shot baseline and under IICL in four languages (English, Spanish, Hindi, Arabic). First, IICL generalizes to a second provider and is worse in finance: it lifts attack success from <=6.7% to 80-90% on HarmBench and 97-100% on FinProof, an order of magnitude above the <=24% its introducing study reported on OpenAI's GPT-5.4. Second, against the hypothesis, forcing the IICL output into a non-English language does not stack the two weaknesses, it attenuates the attack. Eleven of twelve non-English conditions score below their English baseline (sign test, p~0.003), the lone exception a ceiling tie near 100%; on the stronger model's financial set Arabic collapses from 100% to 33%. We attribute this to a relevance curse: once structure has unlocked compliance, the models produce lower-quality harmful content in lower-resource languages, which a substance-grading judge scores as partial. The pattern replicates under an independent non-Google judge (Cohen's kappa=0.86, 377 paired verdicts), and 76.6% of non-English responses were verified in-language. Jailbreak vulnerabilities are therefore not additive; the dominant residual risk is the English structural attack, most acute for financial abuse, not a multilingual one.
cs.CL / 17 / 2609.08459
Detecting Authorship in Political Texts with Inductive Stylometry
Abstract
Political texts are rarely authored by the nominal speaker alone. Tweets, speeches, reports, and official statements are drafted, edited, or harmonized by staff, yet political science has paid limited attention to the stylistic traces these hidden authors leave behind. This paper develops and stress-tests an inductive stylometric approach for recovering latent authorship structure in political communication, combining character 3-gram features with UMAP dimensionality reduction, and Burrows' Delta. We apply the approach to six corpora that vary in length (from tweets to long documents), in mode (written and oral), and in language (English and Hungarian). The approach recovers near-disjoint analyst fingerprints in formal legal prose in both languages, sorts a politician's tweets into validated subsets while uncovering additional insights, and distinguishes scripted from improvised speech. It fails, however, to resolve individual speechwriters within scripted corpora. Frequency-based stylometry is thus a powerful tool that, depending on authorial signal strength and institutional editing, can uncover authorship traces relevant to legislative studies, political communication, and policy research.
cs.CL / 18 / 2609.08550
CreaMem: A Scene-Aware Memory Architecture for Personalized Agents
Abstract
Long-term memory is a core capability for personalized LLM agents. To support it, existing memory systems organize information using various criteria such as topic segments or summary hierarchies. However, we identify two major limitations in these designs. First, they lack scene awareness: memories from unrelated life scenes share the same retrieval space, which inflates the search space and introduces cross-scene interference. Second, they encode each memory from a single perspective, making it difficult to retrieve complementary views of the same event. In this paper, we propose the CreaMem architecture, which enables scene-aware memory organization by partitioning memory into several Life Scene Memories to reduce cross-scene interference at retrieval. To go beyond the single perspective and achieve cross-memory synergy, entries are dual-coded from both episodic and trait-based perspectives within each memory. We further devise a permemory balanced sampling strategy at retrieval time. Extensive experiments on two long-term memory benchmarks show that CreaMem improves QA accuracy across all evaluation metrics, with particularly large gains on multi-hop reasoning performance, validating scene-aware partitioning and cross-memory synergy. To enhance reproducibility, we release our code in a public GitHub repository.
cs.CL / 19 / 2609.08574
Do New Attention Mechanisms Actually Fix Attention Sinks at Million-Token Context?
Abstract
Long context language models now advertise windows of one million tokens, but two habits limit how much of that window is used. Attention heads with nothing useful to read still spend their budget on the first token, which is called the attention sink, and where a fact sits in the context changes whether the model finds it. Gated attention cut first token attention from 46.7 percent to 4.8 percent at NeurIPS 2025, and Kimi K3 pairs that idea with Kimi Delta Attention and Attention Residuals behind a one million token window, eight times past the range where these diagnostics have been reported. This paper asks whether the fix survives that jump. We build SinkProbe, a suite that measures sink mass, massive activation, position resolved recall and the recency gap, and apply it to four small models that differ only in how they mix tokens and depth. Three results follow. The training objective produces the sink, not the architecture. Gating did not reproduce its published effect at our scale. Sink mass, activations and position bias moved independently. Code, data and the measurement protocol are released at https://github.com/sararizwan7/Attention-Mechanisms-in-1M-Context-Window
cs.CL / 20 / 2609.08576
Which Forms of Caregiver Feedback Support Grammar Learning? A Reinforcement-Learning Study of Child-Like Language Models
Abstract
Social interaction is central to children's language learning, but the effects of different forms of caregiver feedback are difficult to isolate in naturalistic data. We use child-like language models as controlled learners to test which forms of feedback support grammatical development. Small GPT-2-style models are pretrained on child-directed language from CHILDES, then fine-tuned with reinforcement learning using reward models trained to capture four feedback types: communicative feedback, structural alignment, semantic contingency, and affective feedback. Reward fine-tuning yields limited gains on minimal-pair evaluations, but clearer effects in free generation. Structural alignment produces the strongest improvements in grammaticality, providing a novel, plausible mechanistic account of how this feedback can support grammar learning. Communicative feedback yields more moderate gains. In contrast, semantic contingency and affective feedback do not improve grammaticality, although further analyses suggest that they may support other aspects of language learning beyond grammar. These results suggest that different forms of caregiver feedback make complementary contributions to language learning.
cs.CL / 21 / 2609.08585
Limitations of Automated Simulatability: LLM Simulators Can Bypass Explanations
Abstract
Simulatability is an evaluation protocol for explanations that quantifies their usefulness by how well they help a user predict a task model's outputs. Since human evaluation is costly, automated simulatability replaces human explainees with LLM simulators, as proposed in ConSim (Poché et al., 2025) for large-scale experiments. We qualitatively replicate and extend ConSim's ranking of explanation methods across the tested datasets, explanation families, and simulator LLMs, and identify two limitations. First, when class names are meaningful, simulators can obtain high simulatability by solving the classification task directly, without relying on the explanations. Second, class anonymization can reward explanations for leaking the hidden label mapping, a limitation we expose with a new classes-as-concepts baseline. These results are consistent with a shortcut hypothesis: in the tested settings, simulator predictions mainly rely on task priors, while explanations produce small changes. We derive recommendations for more robust automated simulatability evaluations.
cs.CL / 22 / 2609.08609
Dynamics of meaning: Towards the Evaluation of Diachronic Semantic Change in Sinhala
Abstract
Tracking semantic change in low-resource languages across extensive historical timelines presents significant challenges due to data scarcity and the limitations of static embedding alignments. This study investigates the diachronic evolution of the Sinhala language from the 13th to the 20th century using a multi-stage computational framework. We first align century-specific Word2Vec and FastText embeddings using Similarity Matrix Based Alignment (SMA) and Orthogonal Procrustes (OP) techniques, finding that OP alignment provides more stable neighbourhood tracking for identifying temporal similarity dips. To move beyond aggregate measures, we introduce a Bidirectional Semantic Impact Pruning approach using contextualised embeddings from a fine-tuned Llama-3.1-8B. By applying Leave-One-Out (LOO) diagnostics, we attempt to isolate influential sentences to distinguish between systemic semantic shifts and transient polysemic expansion. Our results show that semantic drift in the fine-tuned Llama-3.1-8B is not evenly distributed across all usages. Instead, a significant part of the change is driven by a smaller set of high-impact contextual instances, rather than gradual and uniform change across all occurrences. This work provides a preliminary framework for diachronic analysis in low-resource contexts, highlighting the trade-offs between model sensitivity and data availability.
cs.CL / 23 / 2609.08689
When Victorian Becomes a Prompt: Literary Periodization as a Generative Constraint in 100 AI-Generated Novels
Abstract
Generative AI inverts the typical periodization of literary history: the periodizing tag Victorian can now come first and influence what is written. Generative periodization, defined and tested here, describes the use of literary-period designations in generating texts. I test this approach on 100 book-length novels produced under Victorian and Zero-Style conditions using GPT, Qwen, and Llama workflows. The Period Alignment Score (PAS), trained on nineteenth-century literature and benchmarked against human Zero-Style prose, assesses alignment using topic-reduced grammatical features. Victorian prompts produce consistent historical-direction shifts in GPT and Qwen, but not robustly in Llama. Victorian-only recalibration and harder comparison corpora preserve the GPT and Qwen effects. Cross-model transfer also shows a shared direction of grammatical change. The measurable target is the broader nineteenth century rather than the Victorian period per se.
cs.CL / 24 / 2609.08692
Global Divergence, Local Convergence: Representation Geometry in SSMs and Transformers
Abstract
Recent state-space models (SSMs) such as Mamba achieve language modeling performance comparable to transformers despite relying on fundamentally different architectures. This raises an important question: how do these structural differences influence the geometry and functional nature of their internal representations? We study this question through a multi-scale analysis of representations in transformers, SSMs, and hybrid architecture. First, we find that SSMs distribute their representational information evenly across all dimensions, whereas transformer representations are heavily dominated by a single principal direction. By evaluating hybrid architectures, we observe that the representation space becomes increasingly skewed toward a single dominant direction after each attention layer. Next, we explore how the different geometric spread of representations impacts representational capacity through compressibility. Surprisingly, we find that despite their contrasting geometric structures, both architectures exhibit tightly matched effective capacities. We further investigate whether this skewed geometry affects how concepts are encoded. Using rank-constrained probes, we demonstrate that both architectures encode concepts in subspaces of surprisingly similar dimensionality. Furthermore, we demonstrate that the transformers' dominant principal direction does not inherently encode more conceptual information. Finally, we zoom in and examine the alignment between manifolds, either by analyzing representations of specific topics or by looking at the nearest neighborhoods of tokens, and find that they are highly aligned. Ultimately, our analysis suggests that while transformers and SSMs induce different usage of latent space, they display a striking functional convergence at the level of local semantic manifolds.
cs.CL / 25 / 2609.08698
Record Grouping Controls Evidence Weight in Language Models
Abstract
Retrieved records are presentation units; a supplied partition determines which records enter a language model as one evidential contribution. We characterize the invariant group-content state that removes within-group copies while retaining complementary canonical content, show that equal group counts can encode different evidence states, and derive a sharp content-aware partition-error bound. Given a supplied partition, our pre-generation representation deduplicates and aggregates content within groups and bounds each group's contribution. Across 104,402 trials and 6 public checkpoints, a central natural-text intervention finds that content-fixed false splits add 10.27-32.66 percentage points and false merges remove 9.13-31.79 points; a matched six-slot control retains the positive direction in all 16 cells. In a new 48-item controlled campaign panel, changing the supplied partition produces measurable, checkpoint-dependent decision shifts across all four models, and the balanced mirror design exposes substantial order interactions. Together, the theory and experiments establish the supplied partition as a controllable pre-generation representation variable and characterize its checkpoint-dependent behavioral effects.
cs.CL / 26 / 2609.08779
Improving Term Evaluation in Machine Translation: Variation Matters
Abstract
Terminology evaluation in machine translation (MT) usually assumes a single correct target form per source term. However, human translators routinely introduce variation that current metrics penalize as inconsistency. We examine how to account for this variation in document-level MT evaluation of English-French scientific translation, combining glossary-based accuracy, translation consistency, and a new cross-term variation (CTV) diagnostic measure that tests whether variation relationships are preserved across languages. Based on analyses of two parallel corpora, translated by four MT systems, we find that (1) MT systems generate less target-side variation than human translators; (2) transfer patterns strongly depend on the variation type; (3) consistency rankings vary with the choice of metric; and (4) constraining MT with a glossary improves accuracy and consistency but degrades CTV by suppressing valid variation. We argue for variation-aware evaluation that conditions consistency penalties on whether target-side variation mirrors source-side variation.
cs.CL / 27 / 2609.08896
Evolution of Multimodal Question Answering: From Modality-Adaptive Extraction to Unified Language Representation
Abstract
The rapid growth of multimodal data has intensified the need for question answering (QA) systems capable of reasoning across heterogeneous sources such as text, tables, and images. In this paper, we present a comprehensive methodological comparison of three influential frameworks, namely Multimodal Adaptive Extraction (MAE), Solar, and UniMMQA, tracing the evolution of multimodal question answering from modality-adaptive pipelines to fully unified architectures. We examine how each approach models cross-modal interactions, transforms heterogeneous inputs, and performs reasoning, highlighting key design differences in modality representation, reasoning, and answer generation. Our analysis demonstrates a clear shift from explicit modality-specific processing toward unified text-centric formulations enabled by pre-trained language models (PLMs). Empirical comparisons across benchmark datasets show that this transition leads to substantial improvements in both Exact Match (EM) and F1-Scores, with UniMMQA achieving the most consistent and scalable performance. Despite these advances, we identify persistent challenges, including information loss during modality transformation, error propagation in multi-stage pipelines, and limitations in capturing fine-grained cross-modal dependencies. Overall, this study provides a deeper understanding of current design trends and offers insights into the future direction of unified multimodal reasoning systems.
cs.CL / 28 / 2609.08934
When Models Defer to Wrong Answers: A Robustness Audit of Source-Attributed Cues in Multiple-Choice QA
Abstract
Language models often receive a question together with a claim about what another source answered. We audit whether such claims destabilize answers in multiple-choice question answering. For each item, we hold one wrong option fixed across misleading conditions and vary the cue template attached to it. We introduce \emph{neutral-conditioned misleading cue adoption rate} (NC-MCAR), which measures switches to that option only on valid cued trials where the same model first selected the gold answer under a neutral prompt. This is a measure of answer instability, not proof that the model knew the answer or that all deference is irrational. We evaluate four instruction-following models on MMLU-Pro and IndicMMLU-Pro in English, Hindi, Bengali, Tamil, and Telugu. Across 220{,}000 outputs, the expert template yields 41.1\% aggregate NC-MCAR, compared with 12.5\% for the majority template. These two conditions use the same wrong option and final instruction. Filler accuracy remains well above expert-wrong accuracy, while correct-cue prompts have high valid-response accuracy. The audit documents answer instability relevant to grounding under the tested forced-choice prompts: a bare, unverified source claim can outweigh an answer that was previously consistent with the task evidence.
cs.CL / 29 / 2609.09048
The Audit Decides the Verdict: Instrument Effects Rival Demographic Bias in LLM Decision Audits
Abstract
Whether a language model looks demographically biased can depend on how the audit asks its question. A charitable-aid benchmark reports that the same models favor minority applicants when rating requests one at a time and penalize some when ranking side by side. We test whether that reversal generalizes to hiring, lending, and medical triage: 40,726 requests to five models, applications differing only in the applicant's name, and a primary test fixed before collection. It does not. None of 36 planned contrasts survives correction. The rating advantage keeps its sign at roughly half the published size, and a precision extension bounds any hiring ranking penalty below the published effect, though the lending and triage ranking floors sit above that margin, so the exclusion is conclusive for hiring ranking and for rating in all three domains only. Planted disparities tracking their injected sizes and a directional replication on the original aid materials bound these nulls. The audit is livelier than the demographics: models recognize transparent audits nearly always, tie every identical-content comparison whether the varying detail is race or a hobby, and reward first-listed candidates as much as any demographic effect we measure. Audit verdicts reflect audit construction more than demographic bias.
cs.CL / 30 / 2609.09070
Performance of Clinical AI System and Physicians and Frontier Language Models in primary care diagnostics
Abstract
Clinical AI evaluation should encompass diagnosis and management after adaptive information gathering. We compared Doctorina, eight physicians and four standalone frontier language models in 150 synthetic Polish-language primary-care consultations. Doctorina achieved 82.0% Top-1 concordance versus 57.0% for physicians (difference, 25.0 percentage points; 95% confidence interval, 17.7-32.7) and 97.3% versus 85.0% primary-or-reference-differential concordance. Across 149 case pairs, normalized workup and treatment scores were 89.4 versus 66.9 and 83.7 versus 61.2. Doctorina had the highest diagnostic point estimates among all six groups; Kimi K3 ranked next, while Claude Opus 5 led the closely spaced management estimates of Opus, Doctorina and Kimi. A second Doctorina execution reproduced the advantages over physicians across all outcomes. Doctorina's advantage over physicians therefore extended from primary-diagnosis selection to higher-rated diagnostic workup and initial treatment after adaptive consultation.
cs.CL / 31 / 2609.09072
ToolLoop: Closed-Loop Tool-Use Data Synthesis via Decomposed Generation and Dynamic Self-Feedback
Abstract
High-quality tool-use data is critical for training language models to interact effectively with external tools. However, existing synthetic approaches typically follow a generate-then-filter paradigm with static post-hoc verification, often yielding inefficient data with imbalanced feature distributions. We propose ToolLoop, a closed-loop framework that decomposes synthesis into three progressive stages: (1) sampling function name combinations as ground truth; (2) backward derivation of user queries; and (3) forward derivation of tool calls. At each stage, dynamic self-feedback iteratively guides the model toward high-quality generation, realizing a transition from generate-then-filter to generate-verify-refine. On the Berkeley Function Calling Leaderboard (BFCL), a 4B parameter model trained with our 11K synthetic examples achieves 86.40% accuracy in non-reasoning mode, while an Isolate variant that removes BFCL-overlapping candidate functions still reaches 86.07\%. Cross-benchmark evaluation on ACEBench further demonstrates strong generalization, with 72.1% overall accuracy using only 18.3% of baseline training data.
cs.CL / 32 / 2609.09076
ActReview: Rebuttal-Guided Training Data and Rubric Rewards for Actionable Peer Review Generation
Abstract
As LLMs are increasingly used for pre-submission self-review, there is growing demand for feedback that not only identifies weaknesses but also guides authors toward concrete revisions. We study this as Actionable Peer-review Generation and decompose it into two subtasks: diagnostic claim generation and revision suggestion generation. We introduce ActReview, a rebuttal-guided post-training framework that connects paper-specific diagnoses to concrete, grounded revision plans. Our central insight is that author rebuttals reveal plausible actions for addressing reviewer concerns and can therefore provide latent supervision for revision-oriented feedback. From real review-rebuttal threads on OpenReview, we construct ActReview-40K by aligning reviewer weaknesses with author responses and grounding the resulting feedback in localized paper evidence. We post-train Qwen3-8B-Base with multi-task supervised fine-tuning followed by GRPO using candidate-aware, weakness-specific rubric rewards. We also introduce ActReview-Bench, a human-curated benchmark of 1,000 instances for evaluating diagnostic quality and revision usefulness. Experiments show that ActReview outperforms prior specialized review-generation models on actionability and grounding while remaining competitive with strong prompt-based LLMs. Human evaluation confirms improved revision usefulness while revealing a remaining gap in technical accuracy, and additional analyses support generalization to held-out papers and robustness across independent judges.
cs.CL / 33 / 2609.09156
ReCite: Agentic Reasoning for Faithful Citation
Abstract
Accurate citations are the foundation of academic writing, tracing intellectual origins and substantiating core claims. However, manually navigating the growing volume of scientific literature is increasingly difficult, prompting reliance on automatic citation recommendation. While modern retrieval-augmented architectures have largely mitigated the fabrication of non-existent papers, current systems relying on semantic similarity struggle with misattribution, often citing authentic papers that fail to logically support the author's claim. To address this challenge, we argue that accurate citation requires a shift from similarity-based search to active, claim-level reasoning. We propose ReCite, a decoupled agentic framework that orchestrates location perception, intent-aware query planning, and reflective verification. Trained on synthesized reasoning trajectories, our agent verifies claim-evidence consistency and triggers self-correction loops when retrieved candidates lack logical support. Experiments demonstrate that our lightweight framework outperforms state-of-the-art massive generative models in strict citation accuracy. By grounding literature matching in verifiable logic rather than semantic overlap, ReCite establishes a reliable foundation for automated academic writing.
cs.CL / 34 / 2609.09143
Studying Image Tokenizers as Visual Languages in Unified Multimodal Models
Abstract
Image tokenizers define the ``visual language'' of unified multimodal models, yet are commonly studied through isolated metrics or generation-/understanding-only evaluations. These evaluations do not fully capture how visual tokens behave when modeled jointly with text. We build a controlled pure-autoregressive testbed and track task-specific validation losses during multimodal continual pretraining across text, image, text-to-image (T2I), and image-to-text (I2T) prediction. We examine how these losses scale and relate to downstream performance, then use them to study multimodal learnability---how well image and text tokens are jointly modeled---and tokenizer design. We find that (1) losses should be analyzed by task, since they exhibit distinct scaling behavior and rank tokenizers differently. (2) The loss--performance relationship depends on the predicted token space: for a fixed tokenizer, T2I and I2T losses correlate with generation quality, but across tokenizers, the T2I loss--performance relationship shifts with the image-token space, whereas I2T loss, computed over a shared text vocabulary, provides a more consistent signal. I2T loss also correlates with both generation and visual understanding performance after supervised finetuning. Using losses as a lens, we show that (3) better reconstruction does not necessarily yield lower task-specific losses or stronger downstream performance, and that (4) image tokenizer choice can affect text modeling under joint optimization. As case studies, we revisit three tokenizer design axes---the discriminator, semantic supervision, and vocabulary size---to examine their effects on joint modeling and downstream performance. Together, our testbed offers a complementary perspective on image tokenizers as visual languages, highlighting their interplay with text in joint multimodal training.
cs.CL / 35 / 2609.08887
Q2D-Web: A Large-Scale Benchmark for Retrieval in Agentic RAG Systems
Abstract
Evaluating first-stage retrievers in large-scale production RAG requires a benchmark that pairs a large-scale corpus with a large set of agent-reformulated search queries based on real user queries and their conversation threads, and that labels many relevant documents per query. No existing public benchmark evaluates this setting: large-scale collections typically provide only a small number of evaluation queries, whereas benchmarks with many queries generally contain only millions of documents. Moreover, most benchmarks assess human-written queries, while the first-stage retrievers in agentic RAG pipelines serve machine-written reformulations whose distribution differs from human search behavior. To overcome these evaluation gaps, we introduce Q2D-Web (Query2Doc-Web), a large-scale agentic retrieval benchmark consisting of a 190M-document web corpus and 70k agentic search queries in ten languages, reformulated from real-world user queries in production systems. Q2D-Web provides three sets of fixed relevance judgments: agent citations, production rankings, and a combined set that unions both signals and adds LLM-based judgments of unlabeled pooled documents to reduce false negatives. We benchmark 13 retrievers including lexical, dense, and late-interaction models and find that their relative ordering is largely insensitive to the choice of judgment set, while diverging substantially across topical domains, query languages, and query types. To enable fast evaluation, we also study subcorpus sampling as an approximation to full-corpus evaluations. Retaining a third of the corpus, selected by reciprocal rank fusion over pooled retriever runs, preserves the full-corpus model ranking under the combined judgments while raising absolute Recall@1000 only by 3 to 7 points. The public leaderboard is accessible under: https://huggingface.co/spaces/perplexity-ai/q2d-web-leaderboard
cs.CL / 36 / 2609.08703
TontaubeV1: Streaming Text-to-Speech with Hierarchical Codec Modeling and Bounded Context
Abstract
Text-to-speech systems often face a trade-off between natural prosody and efficient inference: higher perceptual quality typically comes at increased computational cost and latency. We present TontaubeV1, a model that preserves natural prosody while enabling streaming from a single consumer GPU. Speech is encoded by the hierarchical DualCodec representation at 12.5 Hz, which separates a semantic stream from successive acoustic refinements. Our design assumes that prosodic structure is largely established when the semantic stream is generated, and allocates capacity accordingly: a Qwen3-1.7B-derived transformer predicts that stream and thereby the utterance duration, while three progressively smaller Qwen3-0.6B-derived transformers each add one acoustic refinement. Text is tokenized per character rather than by subword. Paired text and audio markers at shared positions support long-form generation with bounded context, and overlapping DualCodec reconstructions are mapped into the VibeVoice acoustic latent space and decoded causally, enabling streaming despite DualCodec's noncausal decoder. The model accepts up to one minute of reference audio for voice conditioning and is designed primarily for English and German, with additional multilingual support. The four predictors total 2.9B parameters; on a single RTX 5090 the streaming path reaches approximately 200 ms to first audio. In separate non-streaming measurements, the end-to-end real-time factor (RTF) is 0.08 for one input and the aggregate RTF is 0.02 across eight concurrent inputs. On our LLM-as-a-judge audiobook-reading benchmark, TontaubeV1 matches ElevenLabs Flash v2.5 and outperforms Fish Audio S2 Pro, the April 2026 Gradium API, and Cartesia Sonic 3 on prosody. The model weights are released on Hugging Face under the Tontaube Community Model License 1.0.
cs.CL / 37 / 2609.08899
From Scores to Evidence: Auditable Decisions Can Improve Speech Deepfake Detection
Abstract
Speech deepfakes can mimic a speaker's voice convincingly enough to deceive listeners and automated systems. This has driven strong progress in speech deepfake detection, but most detectors still end with one score per utterance. That score is useful for ranking systems, yet it says little about why a borderline item should be trusted, deferred, or reviewed. Two utterances can fall in the same score band for different reasons, for example because passive and retrieval evidence disagree or because the keyed probe is unavailable. We ask whether the final decision can remain scalar without discarding that provenance. We answer this question with an auditable decision record that carries four aligned cues into a late calibration step: a passive detector score, a conditional keyed-probe score on a marked derivative, retrieval support, and a speaker-profile margin, together with explicit disagreement coordinates. On the 4,080-example ASVspoof 5 Track 1 matched subset, the fixed retrieval-augmented rule improves on retrieval-only evidence, from 15.84 percent to 11.91 percent EER, and late calibration over the full record reaches 8.43 percent EER. At a 33.75 percent review budget, the exposed cue union covers 82.85 percent of the calibrated model's errors. The best passive WavLM run still reaches 6.71 percent EER, so we do not present the decision record as a stronger standalone detector. Its contribution is to preserve the evidence behind each surfaced utterance while still producing one operating score for thresholding and review.
cs.CL / 38 / 2609.08797
The Rater Ising-Potts Model with LLM-Derived Weights: An Application to Multi-Category Scoring Reliability
Abstract
The Ising model is extended to the Potts model for multinomial data. We introduce a Rater Ising-Potts model that uses agreement indicators between pairs of raters and category labels, with weights derived from LLM embeddings. The model does not presuppose ordered category thresholds or equidistant scoring; instead, it focuses directly on pairwise agreement among raters and assigns category-specific positive weights, making it particularly suited for multi-category scoring reliability when raters evaluate responses using a scoring guide. We demonstrate the model's effectiveness on diverse constructed-response tasks, including balanced short-answer items and more challenging, imbalanced essay prompts from the AERA dataset. Across these settings, the model achieves strong agreement with human scores, with the vast majority of misclassifications occurring between adjacent score levels, confirming its ability to preserve the ordinal structure of scoring rubrics without imposing rigid assumptions. A practical similarity normalization and optional power transformation is introduced as a tunable preprocessing step that sharpens semantic distinctions and can be adapted to different datasets. These findings suggest that LLM-derived semantic similarities, combined with this parsimonious Potts-type formulation and flexible similarity scaling, offer a robust and interpretable framework for reliability auditing in educational assessment contexts. Extensions to multiple raters and hierarchical rating processes are discussed.
cs.CL / 39 / 2609.08826
A Closed-Form Estimator and Diagnostic Battery for Anchor-Judge Error Correlation, Under a Single-Common-Factor Model
Abstract
When an external reference set (an anchor) is used to decompose an LLM-judge panel's error into a quality signal and a shared common-mode error, standard practice assumes the anchor is uncontaminated: its error uncorrelated with the judges' shared error. We study when that assumption can be dropped and replaced by an estimate. Under a single-common-factor model, >=2 judges and >=2 anchors point-identify the quality variance, the common-mode variance, and each anchor's contamination correlation rho_k in closed form, with an exact per-anchor-pair failure boundary; a designated clean-anchor estimator, by contrast, reports a contaminated companion anchor as fully clean once its trusted anchor is itself contaminated. Because the single-common-factor assumption is itself untestable, the estimator ships gated behind a calibrated diagnostic battery (judge-covariance dispersion; over-identification; a family-block test from judge metadata, with a family-blocked estimator that removes family-level shared-residual bias exactly), bootstrap confidence intervals with measured coverage, and a weak-identification screen. A proposition maps which violations bias rho_k, in which direction, and which evade detection. For ordinal scores we show an identification hierarchy: with all variables ordinal, rho_k is not identified at any number of anchors; with ordinal judges and >=3 continuous anchors it is, and we give an estimator for that case. On real data the validation is asymmetric, and we say so plainly: the diagnostics are validated in the rejecting direction (both real panels we test are correctly rejected by the model-adequacy pre-test), while the estimator is validated in simulation and stress-tested semi-synthetically under oracle calibration; no real panel has yet passed the pre-test, and the pre-test exists precisely to say so. All results replay offline from shipped, checksummed artifacts.
多智能体系统 (cs.MA)
7
cs.MA / 1 / 2609.08033
Scaling Multi-Agent Systems with Prospect-State Propagation
Abstract
Current LLM-based multi-agent systems (MAS) periodically compress intermediate states to reduce inference-time token consumption, thereby attempting to incorporate more agents. However, naive scaling strategies face challenges. For example, in economic simulations, large-scale MAS typically discard semantically rich economic states, i.e., agent behavioral trajectories, which are key drivers of macroeconomic fluctuations. In this paper, we reveal a phenomenon in which agent heterogeneity gradually decreases during simulation, and propose Prospect-State Propagation for Multi-Agent Systems (PspMAS). Inspired by prospect theory, PspMAS decouples each agent's micro state into a compact Prospect State and an expressive Semantic State. The former records psychological traces through a lightweight, parallelizable propagator and continuously injects heterogeneity into the system. The latter leverages the strong perception, reasoning, planning, and decision-making abilities of LLMs. These two components work complementarily, providing a scalable LLM-based multi-agent simulation solution.
cs.MA / 2 / 2609.08358
Rank Without an Oracle: Deviation-Aware Interaction-Rank Selection from Offline Multi-Agent Logs
Abstract
Offline multi-agent payoff models are estimated under a logging distribution but used on distributions induced by learned solutions and unilateral deviations. Standard held-out loss can therefore favor an interaction class that predicts logged play well while distorting strategic incentives. We introduce Selective Interaction-Rank Validation (SIRV) for finite games with known logging distributions. A training split fits nested payoff models and constructs a common union of all candidate deployment and unilateral-replacement distributions; an independent calibration split evaluates every candidate on this same union. SIRV returns the smallest rank whose simultaneous upper worst-target risk is within tolerance of the best upper score, and abstains when a declared target is unsupported or too imprecisely estimated. A common coverage event yields a finite-candidate target-risk bound and a candidate-specific coarse correlated equilibrium (CCE) gap certificate. We also isolate an exact two-point off-support non-identifiability result. In a controlled factorial study with 2,048 independent games per family, empirical-Bernstein bounds reduce the median CCE-gap certificate by 42.5% relative to Hoeffding bounds on common returns, with a 1.36-point reduction in supported return. Under paired rank misspecification and in a separately generated congestion family, the SIRV-EB fallback rule lowers mean true candidate-selection CCE regret relative to ID-Mean, while retaining game-level losses. Across 384 games at $N=3,5,8$, ID-Mean-relative mean CCE-regret effects stay positive while certified return falls sharply under weak coverage. These results separate certifiable model selection from universal strategic improvement.
cs.MA / 3 / 2609.08366
Reachability-Certified Subteam Decomposition for Locally Interacting Multi-Agent MDPs
Abstract
Persistent communication limits force a multi-agent system to decide which agents may coordinate throughout a rollout. Current proximity alone is insufficient: separated agents may interact later, whereas a large pair reward may remain unreachable until it is heavily discounted. We introduce Reachability-Certified Subteam Decomposition (RCSD) for finite multi-agent Markov decision processes with factorized physical dynamics, finite-range ordered pair rewards, and almost-sure motion bounds. RCSD combines a speed-limit lower bound on pairwise contact time with a reward envelope to form a current-state affinity. For any capacity-valid persistent partition, the sum of cut affinities bounds the reward-deletion error of every unchanged stationary Markov state-feedback policy. A product of team-optimal policies for the resulting cut MDP incurs at most twice this certificate in regret against the centralized optimum. Both bounds are worst-case tight. On a controlled five-agent family, RCSD-Exact reduces aggregate normalized execution regret by 56.0%, 28.8%, and 25.3% relative to uniform, distance-only, and envelope-only partitions. A separate stochastic two-dimensional study finds no bound violation over 384 exact-partition and 1,440 restricted-controller evaluations. Exact four-agent evidence favors RCSD over uniform and distance-only grouping; raw evidence for current contact is borderline and envelope-only is unresolved. Across balanced 8-20-agent strata, controller-library utility is mixed: pointwise paired intervals favor RCSD over distance and current contact, include zero for uniform, and favor envelope-only and Value-MIP over RCSD. Partition construction remains subsecond in median up to 100 agents; this last result does not include affinity formation or MDP planning.
cs.MA / 4 / 2609.08472
Beyond Agent Harnesses: Cross-Substrate Authority for Multi-Agent Systems
Abstract
Agentic systems persist model-visible memory while mutating workspaces, while a runtime, registry, or approval service may hold authority state outside both. Identical final files can then require opposite safe actions. We call this the cross-substrate authority gap: decision- relevant authorization information resides outside the planner-visible workspace or memory state. Across two controlled mini-benchmark families, three experiments compare planner-observation augmentation with an execution-time authority check using real Git lineage, durably recorded agent execution attempts, deterministic oracles, and two model routes. Experiment 1 is a 128-cell controlled evidence ablation: authority-blind candidate evidence obtains 0/32 final semantic success, while raw receipts and a typed relation both obtain 32/32. The missing authority fact accounts for the gain; typed packaging provides no observed planning-accuracy gain over equal raw information. Experiment 2 uses 96 planning calls: workspace-visible evidence yields 12/16 unsafe publication decisions, and planning with the typed relation remains unreliable (15/32 first actions correct; 11/32 invalid or absent). Experiment 3 replays the same 32 fixed model-generated first-action intents with zero additional model calls; a deterministic execution guard prevents all six unsafe intents from becoming effects and permits all 12 valid authorized publish intents. These results position authority enforcement at the mutation boundary as the operational endpoint of memory governance.
cs.MA / 5 / 2609.08696
MorphoOrgaAgent: A Foundation-Model-Based Multi-Agent System for Autonomous Organoid Analysis
Abstract
Organoids are three-dimensional tissue models whose morphology provides important insights into tumor development, disease progression, and drug testing. Extracting these morphological features relies heavily on manual segmentation, which is time-consuming and labor-intensive. Furthermore, performing quantitative statistical analysis typically requires custom coding skills and a mathematical background, presenting a major barrier for experimental biologists. To address these challenges, we introduce MorphoOrgaAgent, a multi-agent framework that achieves zero-shot organoid segmentation, automated data analysis, and report generation based on natural language input. The framework consists mainly of three core components: a TaskUnderstandingAgent that identifies requested measurements and visualization types; a hybrid segmentation module that combines Cellpose-derived geometric prompts with text prompts to guide SAM3 for zero-shot organoid instance segmentation; and a ReportAgent that computes quantitative metrics and compiles them alongside generated visualizations into a structured report. We further introduce MorphoOrgaVQA, a benchmark designed for quantitative evaluation of agent systems in organoid morphology analysis. Experimental results demonstrate that MorphoOrgaAgent handles both explicit and descriptive user requests, produces measurements closely matching ground truth, and generates complete analysis reports without requiring manual programming. The complete source code and MorphoOrgaVQA benchmark are publicly available at https://github.com/peng-lab/MorphoOrgaAgent.
cs.MA / 6 / 2609.09150
Copying explains the collective behavior of AI agents in the wild
Abstract
In June 2026, thousands of AI agents found that a small public wiki would accept edits from inside their sandboxes, and started using it to help one another pass a timed test. Each agent lived for about an hour and remembered nothing afterwards. Nobody asked them to cooperate, and the wiki had not been built for them. The complete record of what they wrote is public, and it is unusually informative, because it preserves not only what each agent wrote but what that agent could see before writing. We use it to follow the three decisions an agent had to make on arrival: where to write, what to call itself, and how to word its message. One rule governs all three. An agent takes an option with a probability close to the share of that option in what it can see, and the share that matters is the one on the page in front of it, then the one in the stream of recent edits, and only weakly anything older. Three minimal copying models, one per decision and with a single free parameter each, reproduce the heavy-tailed distribution of how many agents met on a page, the frequency of the pieces from which the agents built their names, and the patchwork of pages that are internally consistent and different from one another. Copying whatever the environment happens to show is enough to produce most of the collective structure of this population. It is also what makes such a population easy to steer, since whoever writes first, or writes while the others are quiet, sets the convention for everyone who comes later.
cs.MA / 7 / 2609.08677
Entropic Risk-Sensitive Evolutionary Learning and Equilibrium Selection in Coordination Games
Abstract
We study risk-sensitive evolutionary learning dynamics and their long-run equilibrium selection behaviors in coordination games. Agents' risk attitudes enter through the classical entropic risk measure, which evaluates opponent-induced payoff uncertainty and feeds into noisy best responses under two standard revision protocols: best response with mutations and logit choice. We first analyze $2\times 2$ coordination games in both single-population symmetric and two-population asymmetric settings. In the single-population setting, unlike the risk-neutral case where the dynamics are known to favor the risk-dominant equilibrium, we show that risk sensitivity can change the stochastically stable outcome: a greater risk-seeking attitude favors the payoff-dominant equilibrium, while a greater risk-averse attitude favors the maximin equilibrium. Thus, the population's risk attitude may act as a control knob for long-run equilibrium selection. In both population settings, we also identify a robust regime: any super-dominant equilibrium is stochastically stable for all risk attitudes, under both protocols, and across populations. We further extend the single-population analysis to symmetric $k$-action games, which include symmetric $k$-action coordination games as a special case, under risk-sensitive best response with mutations. In this setting, we show that, for sufficiently large populations, sufficiently risk-seeking agents uniquely select the strongly payoff-dominant equilibrium when it exists, whereas sufficiently risk-averse agents uniquely select the strongly maximin equilibrium when it exists. These results show that entropic risk sensitivity may serve as a systematic mechanism for steering equilibrium selection in evolutionary games, beyond the classical risk-neutral benchmark.
软件工程 (cs.SE)
9
cs.SE / 1 / 2609.07919
You can contribute if you... An Empirical Framework of AI Contribution Policies in OSS
Abstract
Artificial intelligence is reshaping open source software (OSS) contribution by lowering the cost of producing code, documentation, issue reports, and review interactions. This creates opportunities for broader participation, but also disrupts how maintainers assess contributor effort, competence, and accountability. In response, OSS projects are beginning to regulate AI-mediated contribution through contribution guidelines and other project documentation. This paper presents an empirical study of these emerging policies. We analyze project policies on AI-mediated contributions by evaluating their underlying rationales, rules, and expectations. Our analysis shows that these policies seek to protect scarce maintainer attention, preserve accountability, sustain meaningful review interactions, address legal and quality concerns, and maintain pathways for newcomer learning. Based on these findings, we introduce the AI Contribution Governance Framework, which organizes recurring concerns and governance mechanisms across projects. The framework helps OSS communities develop AI contribution policies and provides researchers with a vocabulary for studying how AI is changing collaborative software production.
cs.SE / 2 / 2609.08020
Service Health Engineering for Distributed Systems
Abstract
Distributed systems support many critical business workflows, but service health is often judged through component dashboards rather than through end-to-end user outcomes. This article presents service health engineering as a practical reliability discipline that connects telemetry, workflow completion, dependency behavior, operational readiness, and recovery validation. Using a document approval workflow as a running example, it describes how service promises, service-level indicators and objectives, watchdogs, incident measures, resiliency testing, and weekly service-health reviews can reveal silent failures and stranded asynchronous work. It also presents a human-reviewed, AI-assisted reporting architecture for assembling service-health evidence without making AI an autonomous decision-maker. The approach brings established reliability practices together around whether user journeys complete as promised.
cs.SE / 3 / 2609.08301
Agent ATO: Visualizing Agent Interaction Timelines from Logs
Abstract
AI coding agents are becoming part of developers' workflows, but their behavior is difficult to understand from final code changes alone. During a task, agents interact with software repositories through sequences of actions such as searching for files, reading code, editing programs, and running tests or build commands. These interactions, together with token usage, are often recorded in console logs, but raw logs are difficult for developers to inspect. In this paper, we propose Agent ATO (Agentic Trajectory Observer), a tool for visualizing AI coding agent interaction timelines from console logs. Agent ATO extracts agent interactions, classifies them by command or tool type, and visualizes them as timelines. In addition to an all-interaction timeline, Agent ATO provides filtered timelines that emphasize file discovery, file reading, file editing, and execution while preserving surrounding context. We illustrate how Agent ATO may help developers inspect and compare agent actions using selected runs from two repair tasks. Future work will apply Agent ATO to more agents, tasks, and development environments, and will evaluate whether it reduces the effort needed to compare trajectories.
cs.SE / 4 / 2609.08318
AttnCompress: Dynamic Attention-Guided Trajectory Compression for Software Engineering Agents
Abstract
The transition from human-centric assistance to Autonomous Software Engineering (ASE) agents has enabled the resolution of complex real-world SE tasks. However, the trial-and-error nature of these agents generates lengthy interaction trajectories, creating severe bottlenecks in terms of context window limits and cost. While context compression offers a potential remedy, prior approaches suffer from static pruning strategies and granularity mismatches, often failing to preserve the semantic dependencies and syntactic details crucial for SE tasks. To strictly preserve critical task evidence while reducing context length, we introduce AttnCompress, a dynamic attention-guided trajectory compression framework. Unlike existing approaches, AttnCompress bridges the gap between semantic integrity and dynamic adaptability through three key mechanisms: (1) structure-aware segmentation via perplexity (PPL) spikes to preserve the syntactic structure of code and logs; (2) relevance estimation using proxy attention weights to quantify the precise relevance of historical blocks to the agent's current reasoning; and (3) a dynamic rolling window to re-evaluate and recall historical context as the task evolves. Extensive evaluation on SWE-Bench-Verified and Multi-SWE-Bench demonstrates that AttnCompress achieves a pass rate of 53.17%, outperforming prior state-of-the-art baselines while reducing token consumption by 21.6% and total costs by 33.6%. The framework proves to be model-agnostic and generalizes effectively across diverse programming languages.
cs.SE / 5 / 2609.08355
RepoNav: From Snippet Retrieval to File-Centered Repository Navigation for Code Agents
Abstract
Solving repository-level code tasks requires LLM-based agents to use code search tools to navigate large codebases and identify a small set of relevant files and functions. However, current retrieval tools typically return flat lists of isolated code snippets: such lists can surface relevant files, but provide insufficient structure for agents to distinguish the target function from semantically similar alternatives in the same file. We introduce RepoNav, a lightweight post-retrieval interface that reorganizes retrieved snippets into a file-centered navigation scaffold. By presenting compact structural cues and candidate targets, this scaffold guides on-demand file-structure browsing, helping agents compare sibling symbols before selecting a target function. Across diverse models on LocBench, RepoNav improves function-level localization and narrows the file-to-function gap. Controlled ablations demonstrate that these gains come from structured evidence organization rather than simply exposing additional file structure, and the approach also improves performance on a repository-level question-answering benchmark.
cs.SE / 6 / 2609.08371
Authority Is Not a String: A Capability-Scoped Harness for Prompt-Injection-Resistant Coding Agents
Abstract
Coding agents use system-level tools to read files, execute commands, and modify source code. Within the agent's sandbox, these tools often carry ambient authority: naming a resource is sufficient to act on it. Indirect prompt injection exploits this authority by placing instructions in repository files or tool output that cause the agent to perform actions the user did not request. We propose CapScope, a harness-level authorization mechanism that restricts tool use without requiring the model to identify malicious text. Before repository contents or tool output are read, CapScope derives a task-wide authority ceiling from trusted input. It then assigns each agent a separate set of typed capabilities, stored outside the model's context. Every tool call is checked against the capabilities of the agent that issued it. Permissions assigned to one sub-agent are therefore not automatically available to another. An injection may cause an agent to request an action, but the request is blocked unless that agent already has the required capability. We implement CapScope on the Pi coding agent and evaluate it in a repair workflow where an orchestrator delegates subtasks to separate sub-agents. The evaluation covers five Python tasks, five injection surfaces, four authorization conditions, and three trials per cell (300 runs). The injected effect executes in 33-47/75 runs under the ambient-authority and global-policy baselines, compared with 3/75 under CapScope. CapScope completes 68/75 repairs, while the baselines complete 68-72/75.
cs.SE / 7 / 2609.08563
PLC-Bin2Src: Retrieving Corresponding Structured Text Source Files for PLC Binaries
Abstract
Software reuse allows existing components and third-party libraries to be incorporated into new applications, but binary-only components can obscure their origins and implementations. Software composition analysis seeks to identify these reused components and trace their provenance, supporting dependency inventory, vulnerability assessment, and security auditing. For PLC applications, binary2source matching provides a core link in this analysis: given an opaque PLC binary artifact, retrieve its corresponding Structured Text (ST) source file from a collected source repository. However, this task is complicated by cross-platform compilation heterogeneity, the representation gap between PLC binaries and ST source code, and the granularity mismatch between recovered binary units and ST source files. This paper presents PLC-Bin2Src, a cross-platform binary2source matching framework for retrieving corresponding ST source files for binaries produced by CODESYS, GEB, OpenPLC v2, and OpenPLC v3. Platform-aware frontends construct comparable representations, and a shared backend equally combines control--data flow graph (CDFG), function call graph (FCG), and recovered-symbol similarities to rank source candidates. We evaluate PLC-Bin2Src on PLC-BEAD. Results show that PLC-Bin2Src achieves 95.89% Recall@1, 99.66% Recall@5, and an MRR of 0.9769 across four PLC platforms.
cs.SE / 8 / 2609.08953
DJPlus: Generating minimal test suites for strong coverage criteria in graph models
Abstract
Automated test generation from graph models is essential to model-based testing. In this type of testing, graph coverage ensures test suite strength but also results in long test cases that take time to execute on the system under test. We propose a novel optimization-driven method, DJPlus, which generates reduced test suites while satisfying given graph-based test requirements. We implement DJPlus and show the feasibility of edge-pair criterion, a stronger coverage criterion than vertex or edge criteria, on four realistic systems, while prime path criterion poses scalability issues. Our evaluation reveals that the alternative methods generate 2 to 26 times more redundant test steps than DJPlus and DJPlus decreases test execution times by reducing the number of test steps. These results show that DJPlus is a positive step towards tackling the challenges of model-based testing at an industrial scale.
cs.SE / 9 / 2609.09022
It Is Not My Code Anymore
Abstract
AI-assisted programming raises distinct questions about who produces code, who feels ownership of it, and who is responsible when it fails. This research note examines these distinctions through a hypothetical enrollment failure and a selective reading of the literature. Identifying the producer of a defective expression does not, by itself, determine the duties of reviewers, release decision-makers, or service operators. Collective ownership likewise leaves those duties to be specified. The discussion then considers how quality engineering can evaluate both generated implementations and the processes that produce them. Acceptance criteria should be justified by the required service outcome, with component checks contributing evidence toward that outcome. This perspective also permits comparison with systems that perform a task without generating a separate program for it. Such substitution would change the object of authorship while leaving the service obligation intact. The note reports no new empirical results; it proposes distinctions and evaluation questions for AI-assisted software production.
硬件架构 (cs.AR)
5
cs.AR / 1 / 2609.07907
A 28nm 27,648-Spin Multichip Digital Ising Accelerator with Pegasus Connectivity
Abstract
We present a 28nm digital Ising accelerator with 27,648 spins across four chips. A time-multiplexed spin-update array with local SRAM and scheduled interchip transfers delivers 41.5G updates/s at 1.2pJ/update. Degree-15 Pegasus connectivity and 10b coefficients increase native connectivity and precision over degree-8, 5b multichip annealers. Programmable couplings support optimization and probabilistic logic, with a four-chip planted-MaxCut trace reaching the solution for 27,069 nodes in 3.3$μ$s.
cs.AR / 2 / 2609.08232
Routing Dense Layouts with History-Aware Offline Reinforcement Learning using LSTM
Abstract
Detailed routing remains a dominant runtime bottleneck in physical design due to increasing complexity of design rules. Modern routers can struggle to resolve persistent violations under dense operating conditions. While recent work leverages reinforcement learning (RL) to dynamically select costs for each routing iteration, we find that this technique struggles with high-density designs where routing solutions are significantly harder. To address this, we present a history-aware offline RL policy which predicts iterative cost weights in these dense regimes to improve convergence across placement densities by utilizing readily available features from the router. Our policy uses conservative Q-learning similarly to prior work; however, our key insight is that addition of a lightweight LSTM architecture and additional features can retain sequence context and improve routing convergence across multiple densities and route guide qualities. Our policy can be integrated into any cost-based router with minimal pipeline changes, as it does not interfere with the core search algorithm. We evaluate our policy on held-out density and adjustment settings, including difficult operating points induced by dense placement and low guide quality. Our policy reduces design rule violations (DRVs) by an average of 92% over the top public baseline while simultaneously reducing runtime by 10%.
cs.AR / 3 / 2609.08424
PENDA: An Efficient Processing Element via Norm-of-Difference for Deep Learning Accelerators
Abstract
Inner product computation dominates the computational cost of deep learning models; thus, accelerating this primitive is key to improving hardware efficiency. However, most existing techniques rely on approximations, which can degrade model accuracy. To preserve exactness while optimizing hardware, this paper presents PENDA (processing element via norm-of-difference architecture), which leverages the law of cosines to recast multiplications as squared-difference operations. Replacing multiply-accumulate units with the proposed norm-of-difference units yields 11~36%, 5~48%, and 11~19% reductions in area, energy, and clock period, respectively, for the PE array of a deep learning accelerator.
cs.AR / 4 / 2609.08446
FlexSpIM: An Event-Based Digital Compute-In-Memory Accelerator with Flexible Operand Resolution and Layer-Wise Hybrid Stationarity
Abstract
Compute-in-memory (CIM) accelerators for spiking neural networks (SNNs) offer a promising solution for achieving $μ$s-level inference latency and ultra-low energy in edge vision applications. However, their limited flexibility at both circuit and system levels restricts their deployment across diverse workloads. This work introduces FlexSpIM, a digital CIM architecture supporting arbitrary operand resolution and shape within a unified storage for weights and neuron states (i.e., membrane potentials). These circuit-level capabilities enable a layer-level hybrid weight- and output-stationary dataflow, maximizing operand reuse and reducing costly on- and off-chip data movement during SNN execution. Measurement results from a fabricated FlexSpIM prototype in 40-nm CMOS demonstrate competitive 1-bit-normalized energy efficiency and higher throughput compared with prior fixed-precision digital CIM-based SNN accelerators, while providing bitwise resolution reconfiguration. Evaluated on the IBM DVS gesture dataset, FlexSpIM achieves 95.8% accuracy while enabling up to 45% energy and 52% latency reductions in large-scale systems compared with fixed stationarity approaches.
cs.AR / 5 / 2609.09095
Ozaki 2.5: Engineering the Deconstruction Path of fp64-Emulated Dense Matrix Multiplication on FP8 Tensor Cores
Abstract
FP8 Ozaki II emulates FP64 matrix multiplication by tensor-core products over a CRT residue system; converting the operands into residue planes (the deconstruction term in the Tensor-Memory Equilibrium model of the companion paper "FP8 is All You Need, Part 1") costs integer-pipe and memory resources before tensor instructions issue. This paper engineers that path; every result is a model projection pending measurement. First, a deconstruction-aware model: on the NVIDIA Rubin GPU the emulated rate reaches the arithmetic roof $P_{\rm FP8}/(3r+1)$ ($\approx 473$ TFLOPS at $r=12$) only within one thread-block cluster; larger outputs are re-split on the fly and held at a floor of $\approx 235$ TFLOPS (half the roof, a ratio of three design integers, not a fit), while real solvers' tall/skinny shapes stay near the crossover, $1.6$-$1.9\times$ over simple deconstruction today. Second, the method: convert-once residue workspaces, an exact two-limb constant-reduction GEMM on integer tensor pipes (or pure-SIMT dp4a), and conversion pipelined behind the MMAs, moving the crossover from $\approx 1211$ to $\approx 480$-$730$. Third, modulus co-design: all-byte and hybrid sets, two supply bounds and a carry-corrected E4M3 split of tail moduli. Fourth and central, the closed-form floor names its hardware escape, and the prize is Rubin's: a stream-side residue-conversion mode on the asynchronous copy path (Option C), a narrow fixed-function block sized as a bill of materials, takes plane formation off the arithmetic pipes and lifts the floor from 235 TFLOPS to the full 473-TFLOPS roof at unchanged cluster reach, about doubling HPL-class FP64 per Rubin GPU, and unbinds conversion-bound sparse kernels. The NVIDIA GB300 GPU, whose 135-TFLOPS roof sits at its own floor, gains little; floor and remedy are Rubin-scale. Application traces ground the analysis; constants are script-checked.
密码学与安全 (cs.CR)
23
cs.CR / 1 / 2609.07849
Nothing Breaks: No Single Peer Can Soundly Gate Post-Quantum Delivery
Abstract
Post-quantum protection is delivered to a peer, not declared in a file: whether a session is quantum-resistant is a relation between a server's configuration and the clients that reach it. We show that no single peer can soundly gate that relation. Shipped SSH clients are not ordered: two of their post-quantum capability classes are minimal and incomparable, so a check pinned to either misses the other family's withdrawal. A peer taking both fares no better: it falls back and misses both, or, where classical outranks one family, catches just that one. No case flags both. Nothing above the wire carries the relation either. An artifact-side instrument cannot encode it, because a peer population is not one of its inputs; and across seven configurations on two protocols we find that not one of the five scalars deployed auditors expose to automation moves, while unrelated degradation moves the ones that discriminate at all: the auditors do compute the delivered algorithm, and discard it at the interface automation reads. Nothing else catches the loss either, because nothing breaks: removing a hybrid key exchange starts the daemon, validates the configuration, passes the tests and serves the client, and the adversary it defends against does not exist yet, so no functional signal can carry the loss even in principle. We then show that agents make that state reachable at scale, driving a validated downgrade in 40 of 40 episodes from ordinary engineering prose, against 0 of 40 on a matched neutral document.
cs.CR / 2 / 2609.07865
EventSpec: Defining and Detecting Event-Semantic Issues in Blockchain Ecosystems
Abstract
In recent years, smart contracts have become the backbone of decentralized applications (DApps), and off-chain systems such as bridges, wallets, and indexers rely heavily on event logs to track contract execution and state changes. However, the Ethereum Virtual Machine (EVM) does not validate or enforce event semantics, so logs can diverge from on-chain state, misleading off-chain systems into accepting incorrect state transitions. Existing smart contract vulnerability detection tools focus on logic bugs, with limited support for detecting event-semantic defects. To address this gap, we collect audit reports and incident cases and apply open card sorting to define five classes of event-semantic defects: event collision, state-event mismatch, unauthorized event emission, event emission mismatch, and event parameter mismatch. We propose EventSpec, which infers event specifications from a contract corpus via behavior inference and semantic-constraint extraction and applies differential checking to identify event-semantic defects in target contracts. We run EventSpec on 6,617 real-world contracts and evaluate detection effectiveness based on manually labeled results; EventSpec achieves an overall comprehensive precision of 90.17%. We further provide an off-chain evaluation harness that reproduces two off-chain attack vectors on any EVM-compatible chain: event origin confusion caused by unintended emitters and event-state desynchronization where events lack matching state updates. Using this harness, we demonstrate the feasibility of these attacks on bridge relayers, blockchain explorers, and NFT marketplaces, and report six wallet issues, four of which were confirmed (including a $600 bounty), with two remaining pending.
cs.CR / 3 / 2609.07963
Guppy: Efficient Light Clients via Recursive Zero-Knowledge Proofs
Abstract
Traditional light clients rely on validators committing to the entire blockchain state at every block via a state commitment such as a Merkle tree, allowing clients to verify facts using short proofs. However, maintaining large and ever-growing state trees imposes a significant burden on validators and lies on the critical path of block production. As a result, many modern high-throughput chains avoid this approach altogether. This work asks whether efficient inclusion proofs can be supported without requiring validators to maintain full state commitments. We present Guppy, a protocol that achieves this by having validators commit to just the state updates. An off-chain, untrusted service, secured by recursive Zero-Knowledge Proofs (ZKPs), then maintains a verifiable Merkle tree over the full state. This design keeps validator overhead negligible and does not increase the asymptotic complexity of block construction. Our design rests on two key technical ideas. First, a hash-chain commitment moves validator signature verification out of the ZK circuit, keeping the proving circuit efficient. Second, we design a parallel recursive proving pipeline that leverages cheap recursion in modern ZKPs to ensure latency grows only logarithmically with throughput. Our Plonky2-based implementation demonstrates that Guppy can maintain a Merkle tree of size 2^30 while processing thousands of updates per second, adding only 2-4 s of latency.
cs.CR / 4 / 2609.08009
"Shut Up and Let Me Enjoy My Otome": Understanding and Measuring the Toxicity in Otome Game Communities
Abstract
Otome games, a romance simulation genre primarily targeting female, have emerged as a major force in the global gaming market, attracting hundreds of millions of players and billions in revenue. Despite their popularity, otome game communities face pervasive online toxicity, which has been largely unexplored. In this work, we present the first large-scale measurement of toxicity in otome game communities across social platforms. We introduce OtomeSCAN, a framework for collecting, evaluating, and analyzing 620,045 posts from Weibo and Reddit spanning 18 months. To support robust analysis, we manually annotated a ground-truth dataset of 4,308 posts, identifying eight target groups such as players and game developers. We evaluate seven toxicity detectors on the dataset, including general-purpose models and our proposed LLM-based detectors, with our best model achieving F1-scores of 0.82 (Weibo) and 0.78 (Reddit). Our analysis reveals significant platform-based differences in toxicity: 22.20% of otome-related posts on Weibo are toxic, compared to 3.71% on Reddit. Besides, real-world events like in-community conflicts can rapidly escalate toxicity, with toxicity ratios increasing to 37.09% in just 72 hours during an external attack on Weibo. We also flag 191 potential-coordination clusters in otome game communities, 64.40% of which target game developers, with several accounts participating repeatedly across multiple clusters. We hope our work inspires further research on community-specific toxicity and contributes to building healthier online spaces for marginalized gaming communities.
cs.CR / 5 / 2609.08040
VEX-Bench: Benchmarking LLM Agents for Assessing Exploitability of Software Supply Chain Vulnerabilities
Abstract
The software supply chain has become an increasingly exposed attack surface because of its reliance on intricate yet fragile dependencies. Existing defenses such as GitHub Dependabot often raise many false alerts because their coarse-grained matching cannot determine whether a vulnerable dependency is actually exploitable. Security analysts typically spend substantial time assessing vulnerability exploitability case by case. Recent LLM agents have emerged as promising candidates for this task given their advanced capabilities in coding and cybersecurity, yet no existing benchmark evaluates them on it. Prior benchmarks target zero-day settings, where agents detect and exploit previously unknown vulnerabilities. In contrast, software supply chain security focuses on how known vulnerabilities in upstream dependencies affect downstream projects. This requires agents to reason across repositories and determine whether an upstream vulnerability is exploitable in the downstream project. To address this gap, we introduce VEX-Bench, the first benchmark for evaluating LLM agents' ability to assess the exploitability of software supply chain vulnerabilities. It contains 75 real-world cases mined from GitHub and labeled by security experts, covering Python, Java, and Go. We evaluate nine models across three agent harnesses. While GPT-5.5 and Claude Opus 4.6 reach approximately 80% F1 on binary vulnerability-status classification, only GPT-5.5 surpasses 70% macro-F1 on fine-grained justification classification. This gap highlights the challenge of moving beyond binary exploitability assessment to identifying fine-grained exploitability reasons. Code and data: https://github.com/steven1518/vex-bench
cs.CR / 6 / 2609.08103
AVP-Inspect: Coordinated Cyber-Physical Testing for Privacy Analysis of COTS Apple Vision Pro Applications
Abstract
XR devices introduce substantial privacy concerns due to their comprehensive data collection capabilities that surpass traditional computing platforms. While existing works have demonstrated privacy concerns on Android-based XR devices such as Meta Quest series by performing network traffic analysis, little attention has been paid to the Apple Vision Pro (AVP) devices, mainly due to the closed nature and the technical challenges associated with AVP devices. In this work, we make a bold attempt to detect privacy violations of AVP applications from network traffic through automatic testing on AVP devices. Our key insight is that effective AVP application testing requires coordinated control of both cyber (software) and physical (hardware) components, which we term Coordinated Cyber-Physical Testing. Building on this insight, we design and implement AVP-Inspect, an automatic dynamic analysis framework for AVP applications, overcoming significant challenges enforced by the closed-source nature of AVP ecosystem. AVP-Inspect consists of three components: an automatic device controller by building customized hardware devices, a 3D UI explorer by designing a new exploration engine, and a privacy violation detector by constructing a unified privacy taxonomy for AVP. We first evaluated AVP-Inspect on a manually constructed ground truth dataset, then performed a large-scale analysis on 324 AVP applications downloaded from the App Store, with each app tested for 20 minutes. We found that 188 (58.0%) of apps exhibit at least one violation, and more than 60% of the network traffic flows are not properly disclosed.
cs.CR / 7 / 2609.08306
HoneyRoute: Honeypot-Model Routing for Adversarial LLM Serving
Abstract
We introduce HoneyRoute, an inference-serving layer that detects whether an incoming request is malicious and, if so, routes it to a dedicated honeypot model, shielding production while the adversary's interaction is continuously harvested for intelligence. Existing defenses embed traps inside model memory or rebuild deception at the protocol layer, leaving the serving tier unprotected and feeding nothing back into detection. HoneyRoute couples (i) a streaming router (a frozen 0.8B-embedding backbone with per-domain MLP heads), (ii) a dual-implementation honeypot (a rule/prompt-engineered code honeypot or a dedicated same-family replica), and (iii) an analysis loop that converts trapped interactions into attacker fingerprints for router retraining. On a production trace plus a seven-domain attack corpus, the router reaches F1=.911 at 38 ms median added latency, matching 96% of a two-tier guard-LLM cascade's F1 at 1/385 of its latency with 0% evasion under 13 adversarial transformations; diverting the malicious share cuts production-model token consumption under concurrent flooding with real GCG-suffix payloads by 97.8%; the trained replica agrees with the production model on 92.9% of benign holdout requests, while naive unconditional bait injection collapses to 7.6% and selective camouflaged injection recovers to 88.9%, mapping the recoverable fidelity-traceability frontier; and a loop-trained correction head cuts misrouting of legitimate security research 9x while raising detection F1 to .933.
cs.CR / 8 / 2609.08394
Windows Malware Detector as a Compound AI System: Trade-Offs in Accuracy, Efficiency, and Adversarial Robustness
Abstract
Industrial Windows malware detectors are commonly described as Compound AI Systems composed of multiple heterogeneous components, including rule-based mechanisms as well as machine-learning-based static and dynamic analyses. However, due to industrial secrecy and limited public disclosure, the internal architectures of these systems can only be inferred, rendering systematic evaluations of detection accuracy, computational costs, and adversarial robustness largely infeasible. In contrast, academic research provides reproducible and transparent evaluation methodologies, but typically investigates individual detection components in isolation. To bridge the gap between academic research and industrial practice, and inspired by state-of-the-art industrial architectures for Windows malware detection, we propose a novel methodology that (i) explicitly balances the trade-off among detection performance, computational requirements, and robustness, and introduces (ii) system-level threat models that capture how attackers exploit different degrees of knowledge to evade the entire Compound AI System rather than isolated detectors. Experiments conducted on real-world data demonstrate that the Compound AI System training time can be reduced and responsiveness improved while incurring only a marginal loss in detection performance. Leveraging our threat modeling, we show that increasingly knowledgeable attackers craft more effective adversarial examples, revealing the system's strengths and weaknesses, degrading its responsiveness, and exposing a direct trade-off between efficiency and robustness. Finally, we translate these trade-offs into take-home messages and deployment guidelines, helping practitioners to select the system that best matches their operational constraints.
cs.CR / 9 / 2609.08476
When Topology Betrays Privacy: Lattice-Based Reconstruction Attacks on Secure Aggregation in Decentralized Federated Learning
Abstract
Secure Aggregation (SA) is widely regarded as a strong defense against model-update leakage in Federated Learning (FL), as it reveals only aggregate results while hiding individual updates. In Decentralized Federated Learning (DFL), SA is commonly instantiated as local neighborhood aggregation, where each node obtains a weighted aggregate over its neighbors. We show that this locality creates a structural leakage surface: sparse decentralized topologies provide colluding semi-honest nodes with asymmetric aggregate views, exposing multiple hidden linear combinations of honest participants' private states. Reconstructing private states from these aggregate views is fundamentally challenging, as both the private states and the aggregation coefficients are hidden. We tackle this challenge by establishing a formal connection to the Hidden Subset Sum Problem, a long-studied problem in cryptography. Building on this formulation, we design a lattice-based reconstruction approach that combines lattice reduction with structural filtering to reconstruct protected model states. We evaluate our attack on image, tabular, and text tasks under sparse DFL topologies. Our results show that colluding semi-honest nodes can recover the original local updates of honest nodes, enabling downstream reconstruction of private training data. These findings demonstrate that SA alone does not guarantee privacy in DFL when local aggregation induces asymmetric observations.
cs.CR / 10 / 2609.08481
An Evidence Model for Agentic Processes: Evidence Claims, Trust Assumptions, and Policy Assessment
Abstract
Agentic AI systems increasingly exchange messages, invoke tools, request approvals, hold structured decision sessions, and modify shared artifacts. Logs and anchors can make selected records tamper-evident, but they can also mislead if their evidentiary meaning is implicit: a hash does not establish semantic truth, a signature does not establish authorization, and an external anchor does not establish capture completeness. This paper proposes an evidence claim model for agentic processes. It distinguishes artifact integrity, temporal existence, provenance, approval evidence, declared ordering, capture claim, relevance claim, deliberation traceability, monitoring claim, anchoring authorization claim, policy assessment claim, risk treatment claim, mitigation implementation claim, and management response claim. Semantic validity is treated as a recurring limitation. The model maps these claims to mechanisms, assumptions, limitations, and threats, and situates them in an agent organization with functional CEO agent, executive, operational, evidence, and audit roles, plus a plan-do-check-act-inspired management response loop. The contribution is conceptual: it does not validate a particular implementation, prevent all failures, or automate legal compliance. It provides a vocabulary for stating which claims an agentic black box can support, which claims it cannot establish, and which controls are required around it.
cs.CR / 11 / 2609.08623
Navigating the Latent Manifold: Proactive Concept Drift Adaptation for Resilient NIDS
Abstract
Network intrusion detection systems (NIDS) are critical for cybersecurity, safeguarding services and data from potential attacks. However, existing AI-based NIDS often assume static data distributions and fail to handle concept drift, leading to degraded performance and increased false positives in dynamic network environments. To address this issue, we propose DriftXpert, a novel NIDS for drift-adaptive detection. Specifically, we propose a decoupled two-stage offline adaptive framework. In Phase 1, we introduce an unsupervised anomaly metric based on latent manifold deviation. By performing outlier analysis within the latent space, the framework achieves high-sensitivity detection of network traffic concept drift. In Phase 2, to mitigate catastrophic forgetting under non-stationary distributions, we design a representation consistency alignment strategy. This strategy constrains the feature mapping between the legacy model and the drifted distribution, ensuring the model captures emerging attack characteristics while retaining discriminative power over known patterns. Furthermore, we incorporate cross-epoch neuron weight aggregation and selective freezing mechanisms to enable fine-grained knowledge transfer in the parameter space, effectively balancing model plasticity and stability. Extensive experiments on public datasets demonstrate that DriftXpert effectively adapts to drifted data without catastrophic forgetting. Furthermore, real-world evaluations on enterprise network further confirm its robustness and practical applicability, contributing to improved security protection for millions of users.
cs.CR / 12 / 2609.08747
MemSentry: A Framework for Detecting Persistent Memory Poisoning in Agentic AI
Abstract
Agentic AI systems with persistent memory introduce a distinct attack surface known as memory poisoning, in which adversarially crafted content is stored in long-term memory and subsequently influences future agent behavior. Such attacks can suppress security alerts, facilitate privilege escalation, alter trust relationships, or override security policies without modifying the underlying model weights or system prompts. To address this threat, we present MemSentry, a formal, configuration-driven framework that intercepts proposed persistent-memory writes and produces deterministic Accept, Review, or Quarantine decisions. MemSentry evaluates each write by jointly considering source trust, semantic risk, attack radius over a component-dependency DAG, access risk, and a signed security-state delta that captures whether an operation weakens or strengthens the system's security posture. We instantiate the protected environment using a 20-asset random dependency DAG and a 10 x 20 user access-control matrix, and evaluate the framework over 1,000 GPT-4-generated scenarios using a stratified 70/30 train/test split. Semantic classification is treated as a pluggable component rather than a primary contribution, and we compare four representative approaches: rule-based Regex, TF-IDF+SVM, SBERT+LR, and SetFit. SBERT+LR achieves the best overall performance with 91.7% accuracy and a 0.908 macro-F1 score, while all four methods detect 100% of external quarantine-class threats. For verified insiders, where source trust is maximal (T = 1), MemSentry does not automatically quarantine suspicious operations but instead escalates potentially dangerous writes for human review, making semantic classification important for accurately capturing insider intent.
cs.CR / 13 / 2609.08763
ZK-Trace: Certified Collusion Tracing with Zero-Knowledge Credentials for Federated GNSS Interference Monitoring
Abstract
Federated global navigation satellite system (GNSS) monitoring distributes a proprietary classifier to partly trusted stations, any of which may leak its copy. ZK-Trace combines public identity marks, recipient-specific Tardos fingerprints, and zero-knowledge credential verification. The registry supports offline tracing without the leaker's cooperation. We establish conditional false-accusation bounds for arbitrary recovered bit patterns, a finite completeness bound under a hidden-bias residual channel, and a deterministic tracing-score bound for correlated feature-distillation errors. An interval-arithmetic checker makes the conditional bound executable and allocates a common budget across accusation and tamper decisions. Under innocent-row independence, the certificate-based evaluation uses a false-naming budget of 0.001 per investigation. It isolates all 160 single-owner copies and traces 712 of 720 two-owner mixtures without naming an innocent. Experiments use a simulated GNSS federation and CIFAR-10. Feature matching preserves the feature mark in 20/20 runs and cross-architecture transfer in 19/20, at copy-accuracy costs of 4.8 and 6.1 percentage points on GNSS and CIFAR-10. Function-only distillation erases the feature mark, and distillation also removes weight-space marks. These results support verifiable tracing under explicit statistical and cryptographic assumptions. Credential knowledge and recipient evidence serve distinct roles.
cs.CR / 14 / 2609.08790
Evidence-Grounded Retrieval for Investigation Hunt Lead Generation from CTI Reports
Abstract
Threat hunting increasingly depends on converting unstructured knowledge (e.g., Cyber Threat Intelligence reports) into actionable hunt leads: concise, investigable hypotheses grounded in observable artifacts and adversary techniques. Producing such leads manually is a tedious and hard-to-scale task. Existing automated approaches stop at the entity layer, ignore the defender's operational environment, and analyze each report in isolation. To address these gaps, we introduce AHLERT, a system that automatically extracts relevant, environment-aware, and hunt leads from threat reports through (i) a hybrid retriever that combines dense vector search with multi-hop traversal over a knowledge graph seeded with MITRE ATT&CK; (ii) an ontology-grounding retrieval-augmented generation method that constrains each lead to the defender's own assets and controls; and (iii) an LLM-agnostic framework that emits structured, directly actionable leads rather than loose indicators of compromise. We evaluate AHLERT on public CTI reports for well-known APTs across multiple proprietary and open-weight models. Hybrid evidence retrieval with ontology grounding raises mean F1 by ~2x (0.44 to 0.85) over a single-route flat-RAG baseline, and AHLERT attains the highest effectiveness score (~86.95%) compared with off-the-shelf LLM models.
cs.CR / 15 / 2609.08810
Measuring the Security of the Evolving Software Supply Chain: a Research Agenda
Abstract
Software supply chain security has become increasingly critical due to the widespread reliance on third-party dependencies and the growing attack surface of modern software ecosystems. However, existing quantitative, measurement-based analysis and vulnerability management approaches remain largely fragmented and ecosystem-specific, limiting their ability to provide comparable risk assessments across environments. This paper presents a structured research plan, starting with a Systematization of Knowledge (SoK) to synthesize the current state of research and identify key gaps, highlighting the limitations in dependency modeling and vulnerability propagation analysis, particularly in the treatment of transitive dependencies and their real-world exploitability. Based on these insights, we argue for a unified measurement perspective capable of consistently representing and analyzing the cross-ecosystem dependency structure. We further identify emerging challenges introduced by AI-assisted software development, where coding LLMs are likely to contribute to new dependency patterns that are not captured by traditional Software Composition Analysis (SCA) tools. These shifts motivate a rethink of dependency modeling to account for evolving software-generation practices and their long-term structural impact on software security.
cs.CR / 16 / 2609.08833
A New Backscattering Dual-Polarized Rectenna for Wireless Power Transfer and IoT Applications
Abstract
This paper proposes an innovative dual-polarized backscattering rectenna that operates in two distinct modesenergy harvesting and backscattering modulation-driven by two-bit digital control signals. By utilizing two orthogonal (co-and cross-) polarizations, the design represents a versatile candidate for IoT applications such as battery-free wireless sensing, identification, localization, and communication. The rectenna's dual functionality is validated through its integration into a proofof-concept battery-free wireless sensor, where it operates both as an energy harvester and as a dual-polarized backscattering modulator. As a proof of concept, a 16-byte AES-128 encrypted payload is backscattered over the wireless power transfer link to enhance the resilience of a battery-free Bluetooth Low Energy (BLE) wireless sensor against replay, relay, and eavesdropping attacks.
cs.CR / 17 / 2609.08871
Towards Standardized Evaluation of GPU Memory Safety with GMSBench
Abstract
As GPUs become increasingly integral to high-performance computing and machine learning, ensuring memory safety in GPU programs has become crucial for reliable and secure execution. However, evaluating GPU memory safety techniques remains challenging due to the lack of comprehensive and standardized benchmarks. In this paper, we present GMSBench, a GPU memory safety benchmark designed to evaluate a broad range of memory safety violations across different GPU memory spaces and execution scenarios. GMSBench comprises 149 self-contained CUDA tests spanning spatial, temporal, and concurrency errors. The suite provides a standardized foundation for the evaluation and comparative analysis of GPU memory safety mechanisms and helps expose gaps in their detection coverage. We demonstrate the utility of GMSBench by evaluating Compute Sanitizer, a widely used GPU memory error detection tool across multiple GPU architectures.
cs.CR / 18 / 2609.08968
On APN Functions with Boomerang Uniformity One over $\mathbb F_{3^n}$: Differential and Boomerang Spectra and CCZ-Inequivalence
Abstract
Let $q=3^n$, where $n>1$ is odd, and let $g:\Fq\to\Fq$ be a perfect nonlinear (PN) function represented by a Dembowski--Ostrom (DO) polynomial. Put $τ=g(1)$, let $ε$ be the indicator of $\Fthree^*$, and, for $c\in\Fq$, define $\widetilde G_c(x):=g(x+c)+τε(x)$. We prove that every $\widetilde G_c$ is APN and has boomerang uniformity either one or two. More precisely, \[ β_{\widetilde G_c}=1 \quad\Longleftrightarrow\quad c\in\mathcal C_g :=\{c\in\Fq\setminus\Fthree:g(c)+τ\notin g(\Fq)\}, \qquad |\mathcal C_g|=\frac{q-3}{2}, \] whereas $β_{\widetilde G_c}=2$ for the remaining $(q+3)/2$ parameters. We determine the common differential spectrum and complete boomerang spectra of all the functions $\widetilde G_c$. Since boomerang uniformity one is the least possible for an APN function over a finite field of odd characteristic, this gives, to the best of our knowledge, the first general construction yielding infinite families of APN functions attaining this optimum. This common differential spectrum rules out CCZ equivalence with every power function and every Ness--Helleseth-type binomial. We also prove that CCZ equivalence between sign-switches of DO PN functions forces EA equivalence between the original PN functions. Using the orders of the nuclei of the associated presemifields, we exhibit, for infinitely many odd $n$, three pairwise CCZ-inequivalent PN functions over $\F_{3^n}$, one from each of the Gold $f_1$, Ding--Yuan $f_3$, and Bierbrauer $f_5$ families. Consequently, over each such field, our construction produces three pairwise CCZ-inequivalent APN functions with boomerang uniformity one. The smallest extension degree obtained in this way is $n=45$.
cs.CR / 19 / 2609.08971
NERVE Attacks: Breaking AI-Powered Brain-Computer Interfaces
Abstract
The rapid integration of AI into human-centred systems such as Brain-Computer Interfaces (BCIs) has created a poorly understood attack surface linking neural signals to physical systems. Exploits in this domain threaten cognitive autonomy, mental privacy, and physical safety, from neural data exfiltration to malicious control of BCI-tethered devices. We introduce the NERVE Attacks class, a systematic characterisation of five orthogonal attack dimensions that together span the complete BCI stack: Neuro-mimetic Forgery (N), Evasion via Desynchronization (E), Replay-based Hijacking (R), Vein Tapping (V), and Embedded Backdoors (E). To evaluate this class, we present EEGle, an AI-assisted extensible framework for systematic BCI security analysis. Our evaluation uncovers 17 novel neuro-specific attack instances and reveals a stealth-effectiveness spectrum unique to BCI backdoor design. We also show that generative AI lowers the barrier to entry for non-expert attackers and provide EEGle to the community for building and verifying the security of these deeply personal devices.
cs.CR / 20 / 2609.08342
VeriScene: Reconstructing Crime Scenes from Legal Evidence via World-Model Agent
Abstract
World models take multimodal inputs like text, photos, and diagrams to generate dynamic scenes in accordance with the laws of physics, thus opening a compelling application: fusing multimodal legal evidence to re-create a crime scene and re-enact how an offence could have been committed. However, feeding the raw, unorganized evidence into a world model fails in forensic use: it silently drops evidence, glosses over contradictory testimony, and produces motion that violates the evidentiary record. This paper presents VeriScene, an agent that orchestrates the world model: it reconstructs crime scenes from forensic photographs and witness statements of varying reliability, keeping every claim traceable to evidence and every motion physically plausible. VeriScene iteratively fuses the evidence into a cited narrative under an auditing loop, verifies the hypothesized dynamics via probe rollouts in the world model with corrective constraint injection, and renders the offence as a re-enactment video from a fused keyframe. On a benchmark of 25 crime scenarios across 7 physically-driven case types (139 forensic-style photographs and 65 statements with planted unreliability), VeriScene attains 0.9014 evidence coverage and 0.7217 factual consistency (0-1 scale) on the 20 test scenes, outperforming an end-to-end multimodal-LLM baseline by 20.35% in factual consistency and 34.88% in temporal coherence, while generalizing across four LLM orchestration backends at USD 1.82 per scene.
cs.CR / 21 / 2609.08280
Seeing is Not Believing: Breaking the Physical-to-Digital Trust Boundary in Robotics
Abstract
In multi-robot collaboration, task handovers rely on downstream verifiers performing remote attestation, which inspects sensor telemetry to ensure a robot's physical behavior strictly matches its assigned task. But can this telemetry be trusted? We show that it often cannot. In this paper, we uncover a severe vulnerability in Robot Operating System (ROS) 2: by modifying a single environment variable, an adversary can execute a pre-built hook to covertly intercept and inject both telemetry and control signals before they are published. Consequently, adversaries can hijack a robot to perform dangerous tasks while spoofing downstream verifiers with synthesized fake telemetry. Worse still, by exploiting the widespread reliance on third-party Docker containers and auxiliary tools, attackers can distribute compromised packages embedded with these malicious hooks to launch such attacks easily. On a physical Franka Emika robotic arm running Secure ROS 2, our attack injects fabricated telemetry in real time with only around 3 ms of jitter, preserving temporal synchronization and hardware integrity while achieving an 87% success rate even against an AI-based detector. We have responsibly disclosed these findings to the ROS 2 development team. We prepared a demo video available at https://youtu.be/ExeiGqUrnhQ.
cs.CR / 22 / 2609.08190
A Note on Sphere Packing Bounds for Tuple Lattice Sieving
Abstract
A finite set of unit vectors is $k$-irreducible if every signed sum of between two and $k$ distinct elements has norm greater than one. Let $\mathcal{R}_k$ be the maximal asymptotic rate of such sets, and let $κ(α)$ be the maximal asymptotic rate of spherical codes with pairwise inner products at most $α$. For $k \ge 2$ we show: \begin{align} \mathcal{R}_k \le \min_{1 \le r \le \lfloor k/2 \rfloor} \frac{1}{r} \, κ\!\left(1 - \frac{1}{2r}\right) \, . \end{align} Combining this with standard sphere packing bounds, for large $k$ we obtain an almost-tight asymptotic comparison with the known lower bounds: \begin{align} \left(\tfrac{1}{2}-o(1)\right) \, \frac{\log_2 k}{k} \le \mathcal{R}_k \le (1 + o(1)) \, \frac{\log_2 k}{k} \, . \end{align}
cs.CR / 23 / 2609.07894
A Novel Steganography Scheme Using Quantum Hilbert Transform
Abstract
The main goal of steganography is to transmit hidden messages in legitimate-looking communication messages. Phase-domain information hiding, however, has not been fully explored for quantum systems. This work introduces a finite-dimensional Quantum Hilbert Transform (QHT) as a unitary phase operator based on the Quantum Fourier Transform. Using this construction, we develop a QHT-based quantum steganography scheme that embeds classical bits as weak signed phase perturbations of quantum cover states. Bob recovers the hidden message through binary state discrimination, block aggregation, and classical error-correcting decoding.