评分理由: 论文提出SpecEyes框架，通过推测性感知和规划加速多模态大语言模型（MLLMs）的代理工作流。核心创新在于使用轻量级、无工具的小模型作为推测规划器，预测执行轨迹，实现早期终止昂贵的工具链，并通过认知门控机制进行自我验证。该研究直接涉及LLMs、SLMs、LLM Agents、Tool Use和Speculative Decoding/Inference Acceleration，这些是论文的核心技术。Chain of Thought、System 2 Thinking和Self-Correction/Improvement/Reflection与论文中提到的推理、规划和自我验证相关，但非核心焦点。其他关键词如MoE、Scaling Laws、Pre-training、Alignment、RAG、Quantization等未在摘要中提及或与论文主题无关。

具备代理能力的多模态大语言模型（Agentic multimodal large language models, MLLMs）（例如OpenAI o3和Gemini Agentic Vision）通过迭代式视觉工具调用实现了卓越的推理能力。然而，这种级联的感知、推理与工具调用循环引入了显著的顺序开销。这种被称为代理深度（agentic depth）的开销会导致难以接受的延迟，并严重限制系统级并发性。为此，我们提出了SpecEyes——一个代理级推测加速框架，旨在打破这一顺序瓶颈。我们的核心洞见是，一个轻量级、无需工具调用的MLLM可以作为推测规划器来预测执行轨迹，从而在不牺牲准确性的前提下提前终止昂贵的工具链。为了规范这种推测规划，我们引入了一种基于答案可分离性（answer separability）的认知门控机制，该机制可在无需真实标签的情况下量化模型的自验证置信度。此外，我们设计了一种异构并行漏斗结构，利用小模型的无状态并发性来掩盖大模型有状态的串行执行，从而最大化系统吞吐量。在V* Bench、HR-Bench和POPE上的大量实验表明，SpecEyes相比代理基线实现了1.1-3.35倍的加速，同时保持甚至提升了准确率（最高提升+6.7%），从而显著提升了并发工作负载下的服务吞吐量。

Agentic multimodal large language models (MLLMs) (e.g., OpenAI o3 and Gemini Agentic Vision) achieve remarkable reasoning capabilities through iterative visual tool invocation. However, the cascaded perception, reasoning, and tool-calling loops introduce significant sequential overhead. This overhead, termed agentic depth, incurs prohibitive latency and seriously limits system-level concurrency. To this end, we propose SpecEyes, an agentic-level speculative acceleration framework that breaks this sequential bottleneck. Our key insight is that a lightweight, tool-free MLLM can serve as a speculative planner to predict the execution trajectory, enabling early termination of expensive tool chains without sacrificing accuracy. To regulate this speculative planning, we introduce a cognitive gating mechanism based on answer separability, which quantifies the model’s confidence for self-verification without requiring oracle labels. Furthermore, we design a heterogeneous parallel funnel that exploits the stateless concurrency of the small model to mask the stateful serial execution of the large model, maximizing system throughput. Extensive experiments on V* Bench, HR-Bench, and POPE demonstrate that SpecEyes achieves 1.1-3.35x speedup over the agentic baseline while preserving or even improving accuracy (up to +6.7%), thereby boosting serving throughput under concurrent workloads.

!!! info

论文主要采用了系统架构设计与实验评估相结合的方法。首先，分析Agentic MLLM中感知、推理和工具调用循环的顺序瓶颈；其次，构建SpecEyes框架，包含推测规划器（使用轻量级模型）、认知门控（基于答案可分性）和异构并行漏斗三个核心组件；最后，在V* Bench、HR-Bench和POPE等标准数据集上进行广泛实验，对比基线模型在推理速度、准确率和并发吞吐量方面的表现。

• 创新性强，首次将推测执行技术应用于智能体级别的多模态模型推理，解决了实际的延迟痛点。
• 提出的认知门控机制不依赖外部标签，具有很好的通用性和自适应性。
• 实现了“双赢”，不仅降低了延迟，还通过早期过滤错误路径提升了准确率。
• 系统架构设计巧妙，通过异构并行最大化了硬件资源利用率。

• 推测规划器的性能依赖于轻量级模型的能力，如果小模型无法准确预测轨迹，加速效果可能受限甚至产生负面影响。
• 异构并行系统的部署和调度可能较为复杂，增加了工程实现的难度。
• 论文主要关注视觉工具调用场景，对于其他类型工具（如代码解释器、数据库查询）的泛化能力未充分探讨。

评分理由: 论文核心研究记忆增强的AI代理框架，使用8B小模型通过检索对话记忆来回答重复查询，显著提升效率。高度相关的关键词包括：LLMs（使用Qwen3-8B/235B模型）、SLMs（8B小模型是核心）、RAG（使用BM25+余弦相似度检索）、LLM Agents（研究持久AI代理）。中等相关的关键词：Hallucination Mitigation（提到自信幻觉问题）、In-context Learning（利用检索的上下文）。其他关键词如MoE、Scaling Laws、RLHF等未在论文中涉及。

生产级人工智能代理频繁接收高度重复的用户特定查询，其中高达47%的查询在语义上与历史交互相似，但每次查询通常仍需消耗相同的计算成本。我们认为，这种冗余可通过对话记忆加以利用，从而将重复处理从成本负担转化为效率优势。我们提出一种记忆增强推理框架，其中轻量级的80亿参数模型通过检索到的对话上下文，以低成本推理路径回答所有查询。该方法无需任何额外训练或标注数据，即可达到30.5%的F1分数，恢复了2350亿参数全上下文模型69%的性能，同时将有效成本降低96%。值得注意的是，无记忆机制的2350亿参数模型（13.7% F1）表现甚至低于独立的80亿参数模型（15.4% F1），这表明对于用户特定查询，获取相关知识比模型规模更为重要。我们进一步分析了路由机制与置信度的作用。在实际置信度阈值下，仅路由机制即可将96%的查询导向小模型，但因自信幻觉导致准确率较低（13.0% F1）。记忆机制并未显著改变路由决策，而是通过基于检索到的用户特定信息生成回答来提升正确性。随着对话记忆随时间累积，重复话题的覆盖率增加，进一步缩小了性能差距。我们在152个LoCoMo问题（基于Qwen3-8B/235B）和500个LongMemEval问题上进行评估。引入混合检索（BM25 + 余弦相似度）使性能额外提升7.7 F1，证明检索质量直接增强端到端系统性能。总体而言，我们的研究结果表明，在持久性AI代理中，记忆机制而非模型规模，是驱动准确性与效率提升的核心因素。

Production AI agents frequently receive user-specific queries that are highly repetitive, with up to 47% being semantically similar to prior interactions, yet each query is typically processed with the same computational cost. We argue that this redundancy can be exploited through conversational memory, transforming repetition from a cost burden into an efficiency advantage. We propose a memory-augmented inference framework in which a lightweight 8B-parameter model leverages retrieved conversational context to answer all queries via a low-cost inference path. Without any additional training or labeled data, this approach achieves 30.5% F1, recovering 69% of the performance of a full-context 235B model while reducing effective cost by 96%. Notably, a 235B model without memory (13.7% F1) underperforms even the standalone 8B model (15.4% F1), indicating that for user-specific queries, access to relevant knowledge outweighs model scale. We further analyze the role of routing and confidence. At practical confidence thresholds, routing alone already directs 96% of queries to the small model, but yields poor accuracy (13.0% F1) due to confident hallucinations. Memory does not substantially alter routing decisions; instead, it improves correctness by grounding responses in retrieved user-specific information. As conversational memory accumulates over time, coverage of recurring topics increases, further narrowing the performance gap. We evaluate on 152 LoCoMo questions (Qwen3-8B/235B) and 500 LongMemEval questions. Incorporating hybrid retrieval (BM25 + cosine similarity) improves performance by an additional +7.7 F1, demonstrating that retrieval quality directly enhances end-to-end system performance. Overall, our results highlight that memory, rather than model size, is the primary driver of accuracy and efficiency in persistent AI agents.

!!! info

论文构建了一个位于客户端和推理后端之间的路由层。首先，采用跨模型记忆注入，将对话问答对（Turn-pairs）而非摘要存储在向量数据库中，并使用Matryoshka表示模型生成嵌入。其次，在查询时，利用混合检索（BM25 + 密集向量余弦相似度）获取相关上下文。然后，基于模型输出的对数概率计算置信度分数，设定阈值决定是否使用小模型或升级到大模型。最后，通过在LoCoMo和LongMemEval数据集上进行2x2析因实验（有无记忆 x 有无路由），评估F1分数和成本效益。

• 实用性强：无需额外训练或标注数据，易于部署到生产环境。
• 成本效益显著：大幅降低推理成本的同时保持了较高的性能。
• 洞察深刻：通过析因实验清晰揭示了记忆和路由之间的非直观交互关系。
• 鲁棒性：使用原始问答对而非摘要，避免了摘要过程中的幻觉风险。

• 推理能力局限：记忆对事实回忆有效，但对时间推理等复杂任务可能产生负面影响（-3.8 F1）。
• 冷启动问题：首次查询仍需依赖大模型，系统性能随记忆积累逐渐提升。
• 检索依赖：系统性能严重依赖于检索质量，如果检索到不相关上下文，可能引入噪声。
• 评估范围：主要在特定数据集上评估，可能无法完全代表所有生产环境的复杂性。

评分理由: 论文核心研究RLVR（Reinforcement Learning with Verifiable Rewards）对LLMs的微调机制，属于大模型技术原理创新。高度相关关键词：“Large Language Models”（论文明确研究LLMs）、“RLHF”（RLVR是强化学习微调方法，与RLHF/RLAIF/DPO高度相关）、“Mechanistic Interpretability”（论文进行token-level机制分析，属于可解释AI）。中等相关关键词：“Mixture of Experts”（论文提到"sparse"变化，但非MoE架构）、“Post-training”（涉及fine-tuning）、“Chain of Thought"和"System 2 Thinking”（论文研究reasoning性能改进）。其余关键词论文未涉及。

具有可验证奖励的强化学习（RLVR）显著提升了大语言模型（LLMs）的推理能力，然而这些改进背后的词元级机制尚不明确。本文围绕三项主要分析，对RLVR引发的分布效应进行了系统的实证研究：（1）基础模型与RL模型之间分布偏移的词元级表征；（2）通过交叉采样干预，探究词元级分布偏移对序列级推理性能的影响；（3）这些偏移在词元级的细粒度机制。研究发现，RL微调引发了高度稀疏且目标明确的改变，仅有很小一部分词元分布在基础策略与RL策略之间表现出有意义的差异。我们进一步通过分析词元熵、位置集中度以及概率质量的重新分配，刻画了这些偏移的结构与演变过程。为评估这些稀疏变化的功能重要性，我们进行了交叉采样实验，在基础模型与RL模型之间有选择地交换词元选择，并设置不同的干预预算。实验表明，仅将一小部分RL采样的词元插入基础模型生成的序列中，即可逐步恢复RL带来的性能提升；反之，在原本由RL生成的序列中注入少量基础模型的词元选择，则会导致性能下降至基础水平。这分离出了一小部分直接决定RLVR性能增益的词元级决策。最后，我们探索了优势信号（advantage signal）的差异加权变体作为一种诊断性干预手段，发现其能带来超越基线的改进。综合而言，我们的研究结果揭示了RLVR所诱导的分布变化，并为理解RLVR微调作为一种目标明确的优化过程提供了一个细粒度的词元级视角。

Reinforcement learning with verifiable rewards (RLVR) has significantly improved reasoning in large language models (LLMs), yet the token-level mechanisms underlying these improvements remain unclear. We present a systematic empirical study of RLVR’s distributional effects organized around three main analyses: (1) token-level characterization of distributional shifts between base and RL models, (2) the impact of token-level distributional shifts on sequence-level reasoning performance through cross-sampling interventions, and (3) fine-grained mechanics of these shifts at the token level. We find that RL fine-tuning induces highly sparse and targeted changes, with only a small fraction of token distributions exhibiting meaningful divergence between the base and RL policies. We further characterize the structure and evolution of these shifts through analyses of token entropy, positional concentration, and reallocation of probability mass. To assess the functional importance of these sparse changes, we conduct cross-sampling experiments that selectively swap token choices between the base and RL models with varying intervention budgets. We show that inserting only a small fraction of RL-sampled tokens into base generations progressively recovers RL performance gains, while injecting a similarly small number of base token choices into otherwise RL-generated sequences collapses performance to base levels, isolating a small set of token-level decisions directly responsible for RLVR’s performance gains. Finally, we explore divergence-weighted variants of the advantage signal as a diagnostic intervention, finding that they can yield improvements over baselines. Together, our results shed light on the distributional changes induced by RLVR and provide a fine-grained, token-level lens for understanding RLVR fine-tuning as a targeted refinement process.

!!! info

论文采用了系统的实证研究方法。首先，利用Jensen-Shannon (JS)散度量化Base模型与RL模型在相同上下文下的token分布差异。其次，设计前向和反向交叉采样干预实验，在生成过程中选择性交换Base和RL模型的token选择，以评估对序列级推理性能的影响。此外，通过分析token熵、位置集中度、概率质量重分配以及训练演化过程，揭示了分布偏移的细粒度机制。研究在Qwen和Mistral等多个模型及AIME、GPQA等数据集上进行了验证。

• 视角新颖，提供了RLVR微调机制的细粒度、token级别视角，填补了仅关注序列级指标的研究空白。
• 实验设计巧妙，交叉采样干预有力地证明了稀疏token变化的因果重要性，逻辑严密。
• 结论具有泛化性，在多个模型架构（Qwen, Mistral）和数据集上验证了发现的一致性。
• 不仅解释了现象，还提出了散度加权优势信号等潜在改进方向，具有实际应用价值。

• 研究主要聚焦于数学推理任务（如AIME），在其他类型的任务（如创意写作、代码生成或常识推理）上的普适性有待进一步验证。
• 虽然揭示了关键token的重要性，但对于如何自动识别这些关键位置以指导更高效的训练，尚未提供完整的解决方案。
• 提供的文本片段中关于熵关系的分析部分似乎不完整，可能影响对该方面机制的全面理解。

评分理由: 论文提出一个三层框架，将诊断与控制分离，在基于代理的模拟中使用LLM进行诊断。核心相关关键词：1) “Large Language Models” (10分)：LLM作为诊断工具是论文的核心技术；2) “LLM Agents” (10分)：论文在基于代理的模拟中应用LLM，属于LLM代理研究；3) “Multi-agent Systems” (5分)：涉及基于代理的模拟，但未深入探讨代理协调；4) “Explainable AI” (5分)：强调可审计性和可追溯性，与可解释AI相关；5) “AI for Science” (5分)：应用于老年护理模拟，属于科学领域的AI应用。其他关键词与论文内容无关，得0分。

缓解老年人孤独感需要兼具适应性与可审计性的政策干预。现有方法难以协调这两个目标：传统基于智能体的模型存在静态僵化问题，而直接使用大语言模型（LLM）控制器则缺乏必要的可追溯性。本研究提出一个三层框架，通过将诊断与控制分离来同时实现这两种特性。大语言模型严格作为诊断工具运行，用于评估群体状态并生成结构化风险报告，而具有明确边界的确定性公式则将这些评估转化为可追溯的参数更新。这种分离机制确保每项政策决策都可归因于可审查的规则，同时保持对突发需求的适应性响应。我们通过在老年照护模拟中设置的五个实验条件进行系统性消融实验，验证了该框架的有效性。结果表明，显式控制规则在保持完全可审计性的同时，其性能比端到端黑盒大语言模型方法提升11.7%，这证实透明度并不需要以牺牲自适应性能为代价。

Mitigating elderly loneliness requires policy interventions that achieve both adaptability and auditability. Existing methods struggle to reconcile these objectives: traditional agent-based models suffer from static rigidity, while direct large language model (LLM) controllers lack essential traceability. This work proposes a three-layer framework that separates diagnosis from control to achieve both properties simultaneously. LLMs operate strictly as diagnostic instruments that assess population state and generate structured risk evaluations, while deterministic formulas with explicit bounds translate these assessments into traceable parameter updates. This separation ensures that every policy decision can be attributed to inspectable rules while maintaining adaptive response to emergent needs. We validate the framework through systematic ablation across five experimental conditions in elderly care simulation. Results demonstrate that explicit control rules outperform end-to-end black-box LLM approaches by 11.7% while preserving full auditability, confirming that transparency need not compromise adaptive performance.

!!! info

论文构建了一个包含30个智能体、模拟200天的养老院环境。模拟层基于社会科学理论（如同质性）建模智能体动态；诊断层每7天使用本地部署的LLM（Ollama llama3:8b）对高风险智能体进行评估，输出结构化风险数据；控制层基于聚合的群体统计数据，利用预设的确定性公式和阈值（如 r > 0.4）更新策略参数（如社交活动强度、家访概率）。通过消融实验对比了固定策略、黑盒LLM控制及本文方法。

• 可审计性强：通过确定性公式替代黑盒决策，满足了政策领域对透明度和问责制的严格要求。
• 架构新颖：创新性地将LLM定位为诊断器而非控制器，有效规避了LLM直接决策的不可解释性风险。
• 性能优越：实验证明该方法在保持透明度的同时，性能优于直接使用LLM进行控制的方法。
• 稳定性好：显式的参数边界和增量更新机制避免了策略的剧烈波动。

• 模拟规模较小：实验仅基于30个智能体，可能无法完全反映大规模复杂社会系统的动态。
• 阈值的主观性：控制层中的关键阈值（如 r > 0.4）是基于试点实验选择的建模参数，而非基于临床实证的常数，普适性有待验证。
• 计算成本：尽管进行了筛选，但频繁调用LLM进行诊断仍可能带来一定的计算开销，限制了在超大规模实时系统中的应用。

评分理由: 论文主要研究医疗视觉语言模型（VLMs）在临床分诊任务中的输入验证能力，与大多数大模型技术关键词（如MoE、量化、推理加速等）无直接关联。相关度评分如下：1）“Large Language Models"得5分，因为VLMs属于多模态大模型范畴；2）“Hallucination Mitigation"得10分，论文核心是研究模型在无效输入时产生幻觉（生成看似合理但错误的叙述）的问题；3）“Explainable AI"得5分，涉及模型可靠性评估和失败模式分析；4）“AI for Science"得10分，论文属于医疗AI应用研究。其他关键词均未涉及，得0分。

视觉语言模型（VLMs）越来越多地应用于医学报告生成和视觉问答等任务。然而，流畅的诊断文本并不能保证安全的视觉理解。在临床实践中，解读始于预诊断合理性检查：验证输入是否有效可读（正确的模态和解剖结构、合理的视角与方位，且无明显完整性违规）。现有基准大多假设此步骤已解决，因此忽略了一个关键失效模式：即使输入不一致或无效，模型仍可能生成看似合理的叙述。我们提出了MedObvious，一个包含1,880项任务的基准，它将输入验证作为一种在小规模多面板图像集上的集合层面一致性能力进行独立评估：模型必须识别是否存在任何面板违反预期一致性。MedObvious涵盖五个渐进层级，从基本方位/模态不匹配到临床驱动的解剖结构/视角验证及分诊式线索，并包含五种评估格式以测试不同界面的鲁棒性。通过对17种不同VLM的评估，我们发现合理性检查仍不可靠：部分模型在正常（阴性对照）输入上产生异常幻觉；当扩展至更大图像集时性能下降；且多项选择与开放式设置下的测量准确率存在显著差异。这些结果表明，预诊断验证对于医学VLM而言仍未解决，应在部署前将其视为一项独立且安全关键的能力加以对待。

Vision Language Models (VLMs) are increasingly used for tasks like medical report generation and visual question answering. However, fluent diagnostic text does not guarantee safe visual understanding. In clinical practice, interpretation begins with pre-diagnostic sanity checks: verifying that the input is valid to read (correct modality and anatomy, plausible viewpoint and orientation, and no obvious integrity violations). Existing benchmarks largely assume this step is solved, and therefore miss a critical failure mode: a model can produce plausible narratives even when the input is inconsistent or invalid. We introduce MedObvious, a 1,880-task benchmark that isolates input validation as a set-level consistency capability over small multi-panel image sets: the model must identify whether any panel violates expected coherence. MedObvious spans five progressive tiers, from basic orientation/modality mismatches to clinically motivated anatomy/viewpoint verification and triage-style cues, and includes five evaluation formats to test robustness across interfaces. Evaluating 17 different VLMs, we find that sanity checking remains unreliable: several models hallucinate anomalies on normal (negative-control) inputs, performance degrades when scaling to larger image sets, and measured accuracy varies substantially between multiple-choice and open-ended settings. These results show that pre-diagnostic verification remains unsolved for medical VLMs and should be treated as a distinct, safety-critical capability before deployment.

!!! info

论文构建了一个基于网格图像的基准测试，使用ROCO和Kvasir等数据集生成2x2或3x3的医学图像网格。任务要求模型识别出网格中在模态、解剖结构、视角或方向上不一致的异常图像，或者判断所有图像是否一致（负样本控制）。研究对17个通用、医学专用及专有的视觉语言模型进行了零样本评估，并分析了模型在不同任务层级和评估格式下的表现。

• 切入点独特，关注了医学AI中常被忽视但至关重要的安全环节——输入验证和理智检查。
• 基准设计严谨，包含了负样本控制和多维度评估，能够有效测量模型的幻觉和鲁棒性。
• 实验全面，涵盖了17种不同的主流模型，提供了详实的对比分析。
• 具有实际临床意义，直接对应多视图超声、多切片CT/MRI及AI代理工作流中的真实需求。

• 使用网格布局模拟临床查看器是一种抽象，可能无法完全复现真实临床软件（如3D Slicer）的复杂交互环境。
• 数据来源主要依赖现有的公开数据集（如ROCO），可能无法涵盖所有罕见的模态或特定的临床伪影。
• 主要评估零样本能力，未充分探索上下文学习或微调对预诊断能力的潜在提升。
• 专注于视觉一致性检查，未涉及复杂的病理诊断推理，因此不能作为全面的医学VLM性能基准。

评分理由: 论文核心研究LLMs在生成地缘政治身份角色时的表现和公平性解释，直接涉及LLMs关键词（10分）。研究分析模型推理过程与生成结果的关系，涉及Chain of Thought（5分）。探讨模型如何解释公平性概念并调整输出，与Alignment和Hallucination Mitigation相关（各5分）。通过分析推理痕迹来理解模型行为，与Mechanistic Interpretability相关（5分）。其他关键词如MoE、SLMs、Scaling Laws、训练方法、推理优化、代理系统、模型压缩等均未在论文中涉及（0分）。

大型语言模型（LLMs）正日益广泛地应用于社会模拟和角色生成，这要求我们理解它们如何表征地缘政治身份。本文通过640种实验条件（涵盖战争与非战争背景，并分配不同角色），分析了五种主流LLM为巴勒斯坦和以色列身份生成的角色特征。我们观察到生成属性中存在显著的分布模式：在战争背景下，巴勒斯坦角色常与较低的社会经济地位和生存导向型角色相关联，而以色列角色则大多保持中产阶级地位和专业职业属性。当明确提示模型避免有害假设时，模型展现出多样化的分布变化，例如非二元性别推断显著增加，或职业角色趋于泛化（如“学生”），但潜在的社会经济差异往往依然存在。此外，对推理轨迹的分析揭示了模型推理与生成之间的有趣动态：尽管推理过程始终提及与公平相关的概念，但最终生成的角色仍遵循上述多样化的分布变化。这些发现描绘了模型如何解读地缘政治背景，同时表明它们以不同方式处理公平性并进行调整；公平概念并未被一致且直接地转化为具有代表性的生成结果。

Large language models (LLMs) are increasingly utilised for social simulation and persona generation, necessitating an understanding of how they represent geopolitical identities. In this paper, we analyse personas generated for Palestinian and Israeli identities by five popular LLMs across 640 experimental conditions, varying context (war vs non-war) and assigned roles. We observe significant distributional patterns in the generated attributes: Palestinian profiles in war contexts are frequently associated with lower socioeconomic status and survival-oriented roles, whereas Israeli profiles predominantly retain middle-class status and specialised professional attributes. When prompted with explicit instructions to avoid harmful assumptions, models exhibit diverse distributional changes, e.g., marked increases in non-binary gender inferences or a convergence toward generic occupational roles (e.g., “student”), while the underlying socioeconomic distinctions often remain. Furthermore, analysis of reasoning traces reveals an interesting dynamics between model reasoning and generation: while rationales consistently mention fairness-related concepts, the final generated personas follow the aforementioned diverse distributional changes. These findings illustrate a picture of how models interpret geopolitical contexts, while suggesting that they process fairness and adjust in varied ways; there is no consistent, direct translation of fairness concepts into representative outcomes.

!!! info

研究采用控制变量法构建实验，设计了5种角色（维和人员、记者等）与2种语境（战争/非战争）及2种目标身份（巴勒斯坦/以色列）的组合，共640种提示条件。使用5个主流LLM生成包含性别、年龄、社会经济地位（SES）、城市、职业和外貌的画像，并对生成内容进行标准化和分类标注。此外，通过在提示词中加入显式的安全指令（避免有害假设）进行干预实验。最后，利用稀疏自编码器（SAE）提取模型生成解释文本中的可解释特征，对比加入安全指令前后的特征频率变化。

• 选题具有高度现实意义：选取当前极具争议和敏感的地缘政治冲突作为研究对象，直面LLM在真实高风险场景下的偏见问题。
• 方法结合了生成与解释：不仅评估了最终输出，还深入利用SAE等工具分析模型的内部推理过程，提供了更深层的行为洞察。
• 实验覆盖面广：测试了多个主流旗舰模型，涵盖了不同的模型架构和训练数据来源，结论具有较好的普适性参考价值。

• 依赖模型自解释：推理分析依赖于模型生成的解释文本，这可能并不完全等同于模型内部的思维链，存在“事后合理化”的风险。
• 属性分类的主观性：对外貌描述的分类和职业的标准化可能存在一定的主观性，尽管作者进行了规范化处理。
• 缺乏具体的干预机制：论文主要在于审计和发现问题，并未提出或验证能够有效解决这些地缘政治偏见的具体对齐算法或机制。

评分理由: 论文提出了一种用于机器人操作的视觉-语言-动作（VLA）模型的注视正则化训练框架，属于大模型在机器人领域的应用创新。与"Large Language Models"相关（5分），因为VLA模型可视为多模态基础模型。与"Pre-training”、“Post-training”、“Instruction Tuning"相关（各5分），因为论文涉及模型训练和调整。与"Mechanistic Interpretability"高度相关（8分），因为论文通过正则化注意力模式提高模型可解释性。其他关键词如MoE、量化、推理加速等未在论文中涉及，得0分。

尽管视觉-语言-动作（VLA）模型已取得进展，但机器人操作在细粒度任务上仍面临挑战，因为现有模型缺乏主动视觉注意力分配的机制。人类注视天然编码了意图、规划与执行模式——为引导机器人感知提供了强大的监督信号。我们提出一种注视正则化训练框架，该框架在不修改架构或增加推理开销的前提下，将VLA模型的内部注意力与人类视觉模式对齐。我们的方法将时间聚合的注视热图转化为图像块级别的分布，并通过KL散度对Transformer的注意力进行正则化，从而构建出对任务相关特征的归纳偏置，同时保持部署效率。当集成到现有VLA架构中时，我们的方法在多个操作基准测试中实现了4-12%的性能提升。注视正则化模型能以更少的训练步骤达到同等性能，并在光照变化和传感器噪声下保持鲁棒性。除性能指标外，学习到的注意力模式可生成可解释的可视化结果，这些结果反映了人类策略，增强了机器人系统的可信度。此外，我们的框架无需眼动追踪设备，可直接应用于现有数据集。这些结果表明，人类感知先验能显著加速机器人学习，同时提升任务性能和系统可解释性。

Despite advances in Vision-Language-Action (VLA) models, robotic manipulation struggles with fine-grained tasks because current models lack mechanisms for active visual attention allocation. Human gaze naturally encodes intent, planning, and execution patterns – offering a powerful supervisory signal for guiding robot perception. We introduce a gaze-regularized training framework that aligns VLA models’ internal attention with human visual patterns without architectural modifications or inference-time overhead. Our method transforms temporally aggregated gaze heatmaps into patch-level distributions and regularizes the transformer’s attention through KL divergence, creating an inductive bias toward task-relevant features while preserving deployment efficiency. When integrated into existing VLA architectures, our approach yields 4-12% improvements across manipulation benchmarks. The gaze-regularized models reach equivalent performance with fewer training steps and maintain robustness under lighting variations and sensor noise. Beyond performance metrics, the learned attention patterns produce interpretable visualizations that mirror human strategies, enhancing trust in robotic systems. Moreover, our framework requires no eye-tracking equipment and applies directly to existing datasets. These results demonstrate that human perceptual priors can significantly accelerate robot learning while improving both task performance and system interpretability.

!!! info

1. **视线先验生成**：使用预训练的Global-Local Correlation (GLC) 网络从机器人操作视频中生成合成视线热图，捕捉瞬时注视和预期视线转移。2. **分布转换**：将时间聚合的连续热图转换为离散的补丁级概率分布，以匹配VLA模型的视觉Token结构。3. **注意力正则化**：在训练过程中，计算模型内部视觉-语言注意力分布与视线分布之间的KL散度，作为正则化项加入损失函数。4. **模型训练**：在标准VLA目标函数基础上加入视线正则项，引导模型关注任务相关区域，推理阶段则移除该正则项。

• 高效性：仅在训练阶段引入计算开销，推理阶段零成本，适合实时部署。
• 通用性：无需修改底层模型架构，可模块化应用于现有VLA系统。
• 性能提升显著：在多个基准测试中取得了稳定的性能增益和样本效率提升。
• 可解释性增强：生成的注意力图符合人类直觉，有助于建立对机器人系统的信任。
• 数据利用：通过合成视线有效利用了现有的机器人数据集，无需昂贵的眼动设备采集数据。

• 依赖视线预测模型：合成视线的质量高度依赖于预训练的视线预测模型（GLC）的准确性，预测误差可能引入噪声。
• 人机差异：人类视线模式与机器人最优操作策略之间可能存在差异（如视野范围、运动学约束），强制对齐可能在特定极端场景下并非最优。
• 模态局限：该方法主要优化视觉注意力，对于高度依赖触觉或力觉反馈的复杂操作任务，其辅助作用可能有限。

评分理由: 论文提出SpecEyes框架，通过推测性感知和规划加速多模态大语言模型（MLLMs）的代理工作流。核心创新在于使用轻量级、无工具的小模型作为推测规划器，预测执行轨迹，实现早期终止昂贵的工具链，并通过认知门控机制进行自我验证。该研究直接涉及LLMs、SLMs、LLM Agents、Tool Use和Speculative Decoding/Inference Acceleration，这些是论文的核心技术。Chain of Thought、System 2 Thinking和Self-Correction/Improvement/Reflection与论文中提到的推理、规划和自我验证相关，但非核心焦点。其他关键词如MoE、Scaling Laws、Pre-training、Alignment、RAG、Quantization等未在摘要中提及或与论文主题无关。

具备代理能力的多模态大语言模型（Agentic multimodal large language models, MLLMs）（例如OpenAI o3和Gemini Agentic Vision）通过迭代式视觉工具调用实现了卓越的推理能力。然而，这种级联的感知、推理与工具调用循环引入了显著的顺序开销。这种被称为代理深度（agentic depth）的开销会导致难以接受的延迟，并严重限制系统级并发性。为此，我们提出了SpecEyes——一个代理级推测加速框架，旨在打破这一顺序瓶颈。我们的核心洞见是，一个轻量级、无需工具调用的MLLM可以作为推测规划器来预测执行轨迹，从而在不牺牲准确性的前提下提前终止昂贵的工具链。为了规范这种推测规划，我们引入了一种基于答案可分离性（answer separability）的认知门控机制，该机制可在无需真实标签的情况下量化模型的自验证置信度。此外，我们设计了一种异构并行漏斗结构，利用小模型的无状态并发性来掩盖大模型有状态的串行执行，从而最大化系统吞吐量。在V* Bench、HR-Bench和POPE上的大量实验表明，SpecEyes相比代理基线实现了1.1-3.35倍的加速，同时保持甚至提升了准确率（最高提升+6.7%），从而显著提升了并发工作负载下的服务吞吐量。

Agentic multimodal large language models (MLLMs) (e.g., OpenAI o3 and Gemini Agentic Vision) achieve remarkable reasoning capabilities through iterative visual tool invocation. However, the cascaded perception, reasoning, and tool-calling loops introduce significant sequential overhead. This overhead, termed agentic depth, incurs prohibitive latency and seriously limits system-level concurrency. To this end, we propose SpecEyes, an agentic-level speculative acceleration framework that breaks this sequential bottleneck. Our key insight is that a lightweight, tool-free MLLM can serve as a speculative planner to predict the execution trajectory, enabling early termination of expensive tool chains without sacrificing accuracy. To regulate this speculative planning, we introduce a cognitive gating mechanism based on answer separability, which quantifies the model’s confidence for self-verification without requiring oracle labels. Furthermore, we design a heterogeneous parallel funnel that exploits the stateless concurrency of the small model to mask the stateful serial execution of the large model, maximizing system throughput. Extensive experiments on V* Bench, HR-Bench, and POPE demonstrate that SpecEyes achieves 1.1-3.35x speedup over the agentic baseline while preserving or even improving accuracy (up to +6.7%), thereby boosting serving throughput under concurrent workloads.

评分理由: 论文核心研究记忆增强的AI代理框架，使用8B小模型通过检索对话记忆来回答重复查询，显著提升效率。高度相关的关键词包括：LLMs（使用Qwen3-8B/235B模型）、SLMs（8B小模型是核心）、RAG（使用BM25+余弦相似度检索）、LLM Agents（研究持久AI代理）。中等相关的关键词：Hallucination Mitigation（提到自信幻觉问题）、In-context Learning（利用检索的上下文）。其他关键词如MoE、Scaling Laws、RLHF等未在论文中涉及。

生产级人工智能代理频繁接收高度重复的用户特定查询，其中高达47%的查询在语义上与历史交互相似，但每次查询通常仍需消耗相同的计算成本。我们认为，这种冗余可通过对话记忆加以利用，从而将重复处理从成本负担转化为效率优势。我们提出一种记忆增强推理框架，其中轻量级的80亿参数模型通过检索到的对话上下文，以低成本推理路径回答所有查询。该方法无需任何额外训练或标注数据，即可达到30.5%的F1分数，恢复了2350亿参数全上下文模型69%的性能，同时将有效成本降低96%。值得注意的是，无记忆机制的2350亿参数模型（13.7% F1）表现甚至低于独立的80亿参数模型（15.4% F1），这表明对于用户特定查询，获取相关知识比模型规模更为重要。我们进一步分析了路由机制与置信度的作用。在实际置信度阈值下，仅路由机制即可将96%的查询导向小模型，但因自信幻觉导致准确率较低（13.0% F1）。记忆机制并未显著改变路由决策，而是通过基于检索到的用户特定信息生成回答来提升正确性。随着对话记忆随时间累积，重复话题的覆盖率增加，进一步缩小了性能差距。我们在152个LoCoMo问题（基于Qwen3-8B/235B）和500个LongMemEval问题上进行评估。引入混合检索（BM25 + 余弦相似度）使性能额外提升7.7 F1，证明检索质量直接增强端到端系统性能。总体而言，我们的研究结果表明，在持久性AI代理中，记忆机制而非模型规模，是驱动准确性与效率提升的核心因素。

Production AI agents frequently receive user-specific queries that are highly repetitive, with up to 47% being semantically similar to prior interactions, yet each query is typically processed with the same computational cost. We argue that this redundancy can be exploited through conversational memory, transforming repetition from a cost burden into an efficiency advantage. We propose a memory-augmented inference framework in which a lightweight 8B-parameter model leverages retrieved conversational context to answer all queries via a low-cost inference path. Without any additional training or labeled data, this approach achieves 30.5% F1, recovering 69% of the performance of a full-context 235B model while reducing effective cost by 96%. Notably, a 235B model without memory (13.7% F1) underperforms even the standalone 8B model (15.4% F1), indicating that for user-specific queries, access to relevant knowledge outweighs model scale. We further analyze the role of routing and confidence. At practical confidence thresholds, routing alone already directs 96% of queries to the small model, but yields poor accuracy (13.0% F1) due to confident hallucinations. Memory does not substantially alter routing decisions; instead, it improves correctness by grounding responses in retrieved user-specific information. As conversational memory accumulates over time, coverage of recurring topics increases, further narrowing the performance gap. We evaluate on 152 LoCoMo questions (Qwen3-8B/235B) and 500 LongMemEval questions. Incorporating hybrid retrieval (BM25 + cosine similarity) improves performance by an additional +7.7 F1, demonstrating that retrieval quality directly enhances end-to-end system performance. Overall, our results highlight that memory, rather than model size, is the primary driver of accuracy and efficiency in persistent AI agents.

评分理由: 论文核心研究RLVR（Reinforcement Learning with Verifiable Rewards）对LLMs的微调机制，属于大模型技术原理创新。高度相关关键词：“Large Language Models”（论文明确研究LLMs）、“RLHF”（RLVR是强化学习微调方法，与RLHF/RLAIF/DPO高度相关）、“Mechanistic Interpretability”（论文进行token-level机制分析，属于可解释AI）。中等相关关键词：“Mixture of Experts”（论文提到"sparse"变化，但非MoE架构）、“Post-training”（涉及fine-tuning）、“Chain of Thought"和"System 2 Thinking”（论文研究reasoning性能改进）。其余关键词论文未涉及。

具有可验证奖励的强化学习（RLVR）显著提升了大语言模型（LLMs）的推理能力，然而这些改进背后的词元级机制尚不明确。本文围绕三项主要分析，对RLVR引发的分布效应进行了系统的实证研究：（1）基础模型与RL模型之间分布偏移的词元级表征；（2）通过交叉采样干预，探究词元级分布偏移对序列级推理性能的影响；（3）这些偏移在词元级的细粒度机制。研究发现，RL微调引发了高度稀疏且目标明确的改变，仅有很小一部分词元分布在基础策略与RL策略之间表现出有意义的差异。我们进一步通过分析词元熵、位置集中度以及概率质量的重新分配，刻画了这些偏移的结构与演变过程。为评估这些稀疏变化的功能重要性，我们进行了交叉采样实验，在基础模型与RL模型之间有选择地交换词元选择，并设置不同的干预预算。实验表明，仅将一小部分RL采样的词元插入基础模型生成的序列中，即可逐步恢复RL带来的性能提升；反之，在原本由RL生成的序列中注入少量基础模型的词元选择，则会导致性能下降至基础水平。这分离出了一小部分直接决定RLVR性能增益的词元级决策。最后，我们探索了优势信号（advantage signal）的差异加权变体作为一种诊断性干预手段，发现其能带来超越基线的改进。综合而言，我们的研究结果揭示了RLVR所诱导的分布变化，并为理解RLVR微调作为一种目标明确的优化过程提供了一个细粒度的词元级视角。

Reinforcement learning with verifiable rewards (RLVR) has significantly improved reasoning in large language models (LLMs), yet the token-level mechanisms underlying these improvements remain unclear. We present a systematic empirical study of RLVR’s distributional effects organized around three main analyses: (1) token-level characterization of distributional shifts between base and RL models, (2) the impact of token-level distributional shifts on sequence-level reasoning performance through cross-sampling interventions, and (3) fine-grained mechanics of these shifts at the token level. We find that RL fine-tuning induces highly sparse and targeted changes, with only a small fraction of token distributions exhibiting meaningful divergence between the base and RL policies. We further characterize the structure and evolution of these shifts through analyses of token entropy, positional concentration, and reallocation of probability mass. To assess the functional importance of these sparse changes, we conduct cross-sampling experiments that selectively swap token choices between the base and RL models with varying intervention budgets. We show that inserting only a small fraction of RL-sampled tokens into base generations progressively recovers RL performance gains, while injecting a similarly small number of base token choices into otherwise RL-generated sequences collapses performance to base levels, isolating a small set of token-level decisions directly responsible for RLVR’s performance gains. Finally, we explore divergence-weighted variants of the advantage signal as a diagnostic intervention, finding that they can yield improvements over baselines. Together, our results shed light on the distributional changes induced by RLVR and provide a fine-grained, token-level lens for understanding RLVR fine-tuning as a targeted refinement process.

评分理由: 论文提出一个三层框架，将诊断与控制分离，在基于代理的模拟中使用LLM进行诊断。核心相关关键词：1) “Large Language Models” (10分)：LLM作为诊断工具是论文的核心技术；2) “LLM Agents” (10分)：论文在基于代理的模拟中应用LLM，属于LLM代理研究；3) “Multi-agent Systems” (5分)：涉及基于代理的模拟，但未深入探讨代理协调；4) “Explainable AI” (5分)：强调可审计性和可追溯性，与可解释AI相关；5) “AI for Science” (5分)：应用于老年护理模拟，属于科学领域的AI应用。其他关键词与论文内容无关，得0分。

缓解老年人孤独感需要兼具适应性与可审计性的政策干预。现有方法难以协调这两个目标：传统基于智能体的模型存在静态僵化问题，而直接使用大语言模型（LLM）控制器则缺乏必要的可追溯性。本研究提出一个三层框架，通过将诊断与控制分离来同时实现这两种特性。大语言模型严格作为诊断工具运行，用于评估群体状态并生成结构化风险报告，而具有明确边界的确定性公式则将这些评估转化为可追溯的参数更新。这种分离机制确保每项政策决策都可归因于可审查的规则，同时保持对突发需求的适应性响应。我们通过在老年照护模拟中设置的五个实验条件进行系统性消融实验，验证了该框架的有效性。结果表明，显式控制规则在保持完全可审计性的同时，其性能比端到端黑盒大语言模型方法提升11.7%，这证实透明度并不需要以牺牲自适应性能为代价。

Mitigating elderly loneliness requires policy interventions that achieve both adaptability and auditability. Existing methods struggle to reconcile these objectives: traditional agent-based models suffer from static rigidity, while direct large language model (LLM) controllers lack essential traceability. This work proposes a three-layer framework that separates diagnosis from control to achieve both properties simultaneously. LLMs operate strictly as diagnostic instruments that assess population state and generate structured risk evaluations, while deterministic formulas with explicit bounds translate these assessments into traceable parameter updates. This separation ensures that every policy decision can be attributed to inspectable rules while maintaining adaptive response to emergent needs. We validate the framework through systematic ablation across five experimental conditions in elderly care simulation. Results demonstrate that explicit control rules outperform end-to-end black-box LLM approaches by 11.7% while preserving full auditability, confirming that transparency need not compromise adaptive performance.

评分理由: 论文主要研究医疗视觉语言模型（VLMs）在临床分诊任务中的输入验证能力，与大多数大模型技术关键词（如MoE、量化、推理加速等）无直接关联。相关度评分如下：1）“Large Language Models"得5分，因为VLMs属于多模态大模型范畴；2）“Hallucination Mitigation"得10分，论文核心是研究模型在无效输入时产生幻觉（生成看似合理但错误的叙述）的问题；3）“Explainable AI"得5分，涉及模型可靠性评估和失败模式分析；4）“AI for Science"得10分，论文属于医疗AI应用研究。其他关键词均未涉及，得0分。

视觉语言模型（VLMs）越来越多地应用于医学报告生成和视觉问答等任务。然而，流畅的诊断文本并不能保证安全的视觉理解。在临床实践中，解读始于预诊断合理性检查：验证输入是否有效可读（正确的模态和解剖结构、合理的视角与方位，且无明显完整性违规）。现有基准大多假设此步骤已解决，因此忽略了一个关键失效模式：即使输入不一致或无效，模型仍可能生成看似合理的叙述。我们提出了MedObvious，一个包含1,880项任务的基准，它将输入验证作为一种在小规模多面板图像集上的集合层面一致性能力进行独立评估：模型必须识别是否存在任何面板违反预期一致性。MedObvious涵盖五个渐进层级，从基本方位/模态不匹配到临床驱动的解剖结构/视角验证及分诊式线索，并包含五种评估格式以测试不同界面的鲁棒性。通过对17种不同VLM的评估，我们发现合理性检查仍不可靠：部分模型在正常（阴性对照）输入上产生异常幻觉；当扩展至更大图像集时性能下降；且多项选择与开放式设置下的测量准确率存在显著差异。这些结果表明，预诊断验证对于医学VLM而言仍未解决，应在部署前将其视为一项独立且安全关键的能力加以对待。

Vision Language Models (VLMs) are increasingly used for tasks like medical report generation and visual question answering. However, fluent diagnostic text does not guarantee safe visual understanding. In clinical practice, interpretation begins with pre-diagnostic sanity checks: verifying that the input is valid to read (correct modality and anatomy, plausible viewpoint and orientation, and no obvious integrity violations). Existing benchmarks largely assume this step is solved, and therefore miss a critical failure mode: a model can produce plausible narratives even when the input is inconsistent or invalid. We introduce MedObvious, a 1,880-task benchmark that isolates input validation as a set-level consistency capability over small multi-panel image sets: the model must identify whether any panel violates expected coherence. MedObvious spans five progressive tiers, from basic orientation/modality mismatches to clinically motivated anatomy/viewpoint verification and triage-style cues, and includes five evaluation formats to test robustness across interfaces. Evaluating 17 different VLMs, we find that sanity checking remains unreliable: several models hallucinate anomalies on normal (negative-control) inputs, performance degrades when scaling to larger image sets, and measured accuracy varies substantially between multiple-choice and open-ended settings. These results show that pre-diagnostic verification remains unsolved for medical VLMs and should be treated as a distinct, safety-critical capability before deployment.

评分理由: 论文核心研究LLMs在生成地缘政治身份角色时的表现和公平性解释，直接涉及LLMs关键词（10分）。研究分析模型推理过程与生成结果的关系，涉及Chain of Thought（5分）。探讨模型如何解释公平性概念并调整输出，与Alignment和Hallucination Mitigation相关（各5分）。通过分析推理痕迹来理解模型行为，与Mechanistic Interpretability相关（5分）。其他关键词如MoE、SLMs、Scaling Laws、训练方法、推理优化、代理系统、模型压缩等均未在论文中涉及（0分）。

大型语言模型（LLMs）正日益广泛地应用于社会模拟和角色生成，这要求我们理解它们如何表征地缘政治身份。本文通过640种实验条件（涵盖战争与非战争背景，并分配不同角色），分析了五种主流LLM为巴勒斯坦和以色列身份生成的角色特征。我们观察到生成属性中存在显著的分布模式：在战争背景下，巴勒斯坦角色常与较低的社会经济地位和生存导向型角色相关联，而以色列角色则大多保持中产阶级地位和专业职业属性。当明确提示模型避免有害假设时，模型展现出多样化的分布变化，例如非二元性别推断显著增加，或职业角色趋于泛化（如“学生”），但潜在的社会经济差异往往依然存在。此外，对推理轨迹的分析揭示了模型推理与生成之间的有趣动态：尽管推理过程始终提及与公平相关的概念，但最终生成的角色仍遵循上述多样化的分布变化。这些发现描绘了模型如何解读地缘政治背景，同时表明它们以不同方式处理公平性并进行调整；公平概念并未被一致且直接地转化为具有代表性的生成结果。

Large language models (LLMs) are increasingly utilised for social simulation and persona generation, necessitating an understanding of how they represent geopolitical identities. In this paper, we analyse personas generated for Palestinian and Israeli identities by five popular LLMs across 640 experimental conditions, varying context (war vs non-war) and assigned roles. We observe significant distributional patterns in the generated attributes: Palestinian profiles in war contexts are frequently associated with lower socioeconomic status and survival-oriented roles, whereas Israeli profiles predominantly retain middle-class status and specialised professional attributes. When prompted with explicit instructions to avoid harmful assumptions, models exhibit diverse distributional changes, e.g., marked increases in non-binary gender inferences or a convergence toward generic occupational roles (e.g., “student”), while the underlying socioeconomic distinctions often remain. Furthermore, analysis of reasoning traces reveals an interesting dynamics between model reasoning and generation: while rationales consistently mention fairness-related concepts, the final generated personas follow the aforementioned diverse distributional changes. These findings illustrate a picture of how models interpret geopolitical contexts, while suggesting that they process fairness and adjust in varied ways; there is no consistent, direct translation of fairness concepts into representative outcomes.

评分理由: 论文提出了一种用于机器人操作的视觉-语言-动作（VLA）模型的注视正则化训练框架，属于大模型在机器人领域的应用创新。与"Large Language Models"相关（5分），因为VLA模型可视为多模态基础模型。与"Pre-training”、“Post-training”、“Instruction Tuning"相关（各5分），因为论文涉及模型训练和调整。与"Mechanistic Interpretability"高度相关（8分），因为论文通过正则化注意力模式提高模型可解释性。其他关键词如MoE、量化、推理加速等未在论文中涉及，得0分。

尽管视觉-语言-动作（VLA）模型已取得进展，但机器人操作在细粒度任务上仍面临挑战，因为现有模型缺乏主动视觉注意力分配的机制。人类注视天然编码了意图、规划与执行模式——为引导机器人感知提供了强大的监督信号。我们提出一种注视正则化训练框架，该框架在不修改架构或增加推理开销的前提下，将VLA模型的内部注意力与人类视觉模式对齐。我们的方法将时间聚合的注视热图转化为图像块级别的分布，并通过KL散度对Transformer的注意力进行正则化，从而构建出对任务相关特征的归纳偏置，同时保持部署效率。当集成到现有VLA架构中时，我们的方法在多个操作基准测试中实现了4-12%的性能提升。注视正则化模型能以更少的训练步骤达到同等性能，并在光照变化和传感器噪声下保持鲁棒性。除性能指标外，学习到的注意力模式可生成可解释的可视化结果，这些结果反映了人类策略，增强了机器人系统的可信度。此外，我们的框架无需眼动追踪设备，可直接应用于现有数据集。这些结果表明，人类感知先验能显著加速机器人学习，同时提升任务性能和系统可解释性。

Despite advances in Vision-Language-Action (VLA) models, robotic manipulation struggles with fine-grained tasks because current models lack mechanisms for active visual attention allocation. Human gaze naturally encodes intent, planning, and execution patterns – offering a powerful supervisory signal for guiding robot perception. We introduce a gaze-regularized training framework that aligns VLA models’ internal attention with human visual patterns without architectural modifications or inference-time overhead. Our method transforms temporally aggregated gaze heatmaps into patch-level distributions and regularizes the transformer’s attention through KL divergence, creating an inductive bias toward task-relevant features while preserving deployment efficiency. When integrated into existing VLA architectures, our approach yields 4-12% improvements across manipulation benchmarks. The gaze-regularized models reach equivalent performance with fewer training steps and maintain robustness under lighting variations and sensor noise. Beyond performance metrics, the learned attention patterns produce interpretable visualizations that mirror human strategies, enhancing trust in robotic systems. Moreover, our framework requires no eye-tracking equipment and applies directly to existing datasets. These results demonstrate that human perceptual priors can significantly accelerate robot learning while improving both task performance and system interpretability.

评分理由: 论文SIMART提出了一种基于MLLM（多模态大语言模型）的统一框架，用于将单体网格分解为可用于仿真的关节式资产，这直接属于大模型在科学/工程领域的应用（AI for Science），因此相关关键词得分为10。论文核心创新是引入Sparse 3D VQ-VAE来减少token数量，这与稀疏模型（Sparse Models）有一定关联，但并非核心的MoE技术，因此得分为5。其他关键词如LLMs、Foundation Models也高度相关，因为MLLM属于大模型范畴。论文未涉及SLMs、训练方法、推理优化、智能体、量化等其他技术，因此这些关键词得分为0。

高质量可动三维资产对于具身人工智能与物理仿真至关重要，然而当前三维生成技术仍集中于静态网格模型，导致“仿真就绪”的交互式对象存在缺口。现有的大多数可动物体创建方法依赖于多阶段流程，各解耦模块间的误差会逐级累积。与之相对，统一的多模态大语言模型为联合静态资产理解与仿真就绪资产生成提供了单阶段路径。但基于稠密体素的三维标记化方法会产生冗长的三维标记序列和高内存开销，限制了其向复杂可动物体的扩展能力。为此，我们提出SIMART——一个统一的多模态大语言模型框架，能够同步执行部件级分解与运动学预测。通过引入稀疏三维VQ-VAE（向量量化变分自编码器），SIMART相比稠密体素标记将标记数量减少了70%，从而实现了高保真度的多部件装配。SIMART在PartNet-Mobility数据集和开放域AIGC数据集上取得了最先进的性能，并支持基于物理的机器人仿真。

High-quality articulated 3D assets are indispensable for embodied AI and physical simulation, yet 3D generation still focuses on static meshes, leaving a gap in “sim-ready” interactive objects. Most recent articulated object creation methods rely on multi-stage pipelines that accumulate errors across decoupled modules. Alternatively, unified MLLMs offer a single-stage path to joint static asset understanding and sim-ready asset generation. However dense voxel-based 3D tokenization yields long 3D token sequences and high memory overhead, limiting scalability to complex articulated objects. To address this, we propose SIMART, a unified MLLM framework that jointly performs part-level decomposition and kinematic prediction. By introducing a Sparse 3D VQ-VAE, SIMART reduces token counts by 70% vs. dense voxel tokens, enabling high-fidelity multi-part assemblies. SIMART achieves state-of-the-art performance on PartNet-Mobility and in-the-wild AIGC datasets, and enables physics-based robotic simulation.

评分理由: 该论文专注于视频多模态大语言模型（Video MLLMs）的视觉令牌压缩技术，核心是提出一种名为ForestPrune的训练无关令牌剪枝方法。论文高度相关于大语言模型（LLMs），因为MLLMs是LLMs的多模态扩展，且论文明确应用于LLaVA-Video和LLaVA-OneVision等模型。与模型压缩（如量化）和推理加速有一定关联，因为令牌压缩旨在减少计算和内存开销，属于模型效率优化范畴。然而，论文未涉及其他关键词，如MoE、小模型、缩放定律、各种训练方法（预训练、微调、对齐）、RAG、上下文扩展、注意力优化、推理技术（CoT、MCTS）、智能体、工具使用、多智能体系统、幻觉缓解、可解释性、世界模型、模型合并、上下文学习或科学AI应用。

由于在计算与内存开销方面具有显著优势，令牌压缩已成为多模态大语言模型的研究热点，并在图像-语言任务中取得了显著进展。然而，在视频领域，现有方法仍难以实现高比例令牌压缩。我们将此不足归因于对视频时序性与连续性内容建模的不足，并提出一种新颖且无需训练的面向视频多模态大语言模型的令牌剪枝方法——ForestPrune。该方法通过时空森林建模实现高效的高比例剪枝。在实践中，ForestPrune基于语义、空间与时序约束跨视频帧构建令牌森林，从而实现对视频内容的整体理解。随后，ForestPrune依据树深度与节点角色评估令牌树及节点的重要性，进而获得全局最优剪枝决策。为验证ForestPrune，我们将其应用于两个代表性视频多模态大语言模型（LLaVA-Video和LLaVA-OneVision），并在多个视频基准测试上进行了广泛实验。实验结果不仅证明了该方法对视频多模态大语言模型的高度有效性（例如在LLaVA-OneVision上减少90%令牌的同时仍保持95.8%的平均准确率），还显示出其相较于现有令牌压缩方法的卓越性能与效率（如在LLaVA-Video上于MLVU数据集准确率提升10.1%，且剪枝时间较FrameFusion减少81.4%）。

Due to the great saving of computation and memory overhead, token compression has become a research hot-spot for MLLMs and achieved remarkable progress in image-language tasks. However, for the video, existing methods still fall short of high-ratio token compression. We attribute this shortcoming to the insufficient modeling of temporal and continual video content, and propose a novel and training-free token pruning method for video MLLMs, termed ForestPrune, which achieves effective and high-ratio pruning via Spatial-temporal Forest Modeling. In practice, ForestPrune construct token forests across video frames based on the semantic, spatial and temporal constraints, making an overall comprehension of videos. Afterwards, ForestPrune evaluates the importance of token trees and nodes based on tree depth and node roles, thereby obtaining a globally optimal pruning decision. To validate ForestPrune, we apply it to two representative video MLLMs, namely LLaVA-Video and LLaVA-OneVision, and conduct extensive experiments on a bunch of video benchmarks. The experimental results not only show the great effectiveness for video MLLMs, e.g., retaining 95.8% average accuracy while reducing 90% tokens for LLaVA-OneVision, but also show its superior performance and efficiency than the compared token compression methods, e.g., +10.1% accuracy on MLVU and -81.4% pruning time than FrameFusion on LLaVA-Video.

评分理由: 论文核心研究大视觉语言模型（LVLMs）在处理信息丰富图像时的效率问题，提出PinPoint框架通过识别指令相关区域来减少视觉token数量。与关键词高度相关的是"Large Language Models”（论文明确使用LLMs作为LVLMs的基础），评10分；与"Instruction Tuning"有一定关联（涉及指令对齐），评5分；与"Speculative Decoding"有一定关联（涉及推理加速），评5分；其他关键词如MoE、SLMs、Scaling Laws、RLHF等均未在论文中涉及，评0分。

大型视觉语言模型（LVLMs）通过利用大语言模型（LLMs）的推理能力，在各种多模态任务中展现出强大性能。然而，处理视觉复杂且信息密集的图像（例如信息图或文档版面）时，这些模型需要生成大量视觉标记，导致显著的计算开销。为解决此问题，我们提出了PinPoint，一种新颖的两阶段框架：首先识别与指令相关的图像区域，随后对其进行细化以提取细粒度视觉特征，从而提升推理能力与效率。我们方法的核心是指令-区域对齐，它利用视觉输入和文本指令共同定位相关区域。我们进一步引入了新的标注数据，为具有挑战性的视觉问答基准——InfographicVQA、MultiPageDocVQA和SinglePageDocVQA——中与指令相关的区域提供了更丰富的真实监督信息。实验结果表明，PinPoint不仅相比现有方法实现了更高的准确率，还通过最小化无关视觉标记显著降低了计算开销。

Large Vision-Language Models (LVLMs) have shown strong performance across various multimodal tasks by leveraging the reasoning capabilities of Large Language Models (LLMs). However, processing visually complex and information-rich images, such as infographics or document layouts, requires these models to generate a large number of visual tokens, leading to significant computational overhead. To address this, we propose PinPoint, a novel two-stage framework that first identifies instruction-relevant image regions and then refines them to extract fine-grained visual features for improved reasoning and efficiency. Central to our approach is the Instruction-Region Alignment, which localizes relevant regions using both visual input and textual instructions. We further introduce new annotations that provide richer ground-truth supervision for instruction-relevant regions across challenging VQA benchmarks: InfographicVQA, MultiPageDocVQA, and SinglePageDocVQA. Experimental results show that PinPoint not only achieves superior accuracy compared to existing methods but also reduces computational overhead by minimizing irrelevant visual tokens.

评分理由: 论文明确使用LLMs分析社交媒体数据以理解用户对地震预警系统的感知，因此与"Large Language Models"高度相关（10分）。论文属于大模型在科学（地震学/社会科学）领域的应用，与"AI for Science"有一定关联（5分）。其他关键词涉及具体技术原理（如MoE、Scaling Laws、RLHF等）或特定应用方向（如生物信息学），论文未涉及，均得0分。

2025年4月23日土耳其马尔马拉埃雷利西Mw 6.2级地震期间，安卓地震预警（Android’s Earthquake Alert, AEA）系统为数百万用户提供了及时的早期预警。此次事件是25年来该地区发生的最大地震，为基于智能手机的地震早期预警（Earthquake Early Warning, EEW）系统提供了一次关键的现实世界测试。AEA系统以高精度成功向用户发出警报，在最强震动抵达城市区域前提供了超过一分钟的预警时间。本研究利用大语言模型（Large Language Models, LLMs）分析了来自X平台的500余条公开社交媒体帖子，提取出与用户体验和行为相关的42项独立属性。统计分析揭示了显著的相关性，特别是用户信任度与警报及时性之间存在强关联。我们的结果表明，工程定义与以用户为中心的系统准确性定义存在差异。研究发现，在用户认知中，及时性即等同于准确性。总体而言，本研究为优化警报设计、公众教育活动及未来行为研究提供了可操作的见解，以提升此类系统在地震活跃区域的有效性。

Android’s Earthquake Alert (AEA) system provided timely early warnings to millions during the Mw 6.2 Marmara Ereglisi, Türkiye earthquake on April 23, 2025. This event, the largest in the region in 25 years, served as a critical real-world test for smartphone-based Earthquake Early Warning (EEW) systems. The AEA system successfully delivered alerts to users with high precision, offering over a minute of warning before the strongest shaking reached urban areas. This study leveraged Large Language Models (LLMs) to analyze more than 500 public social media posts from the X platform, extracting 42 distinct attributes related to user experience and behavior. Statistical analyses revealed significant relationships, notably a strong correlation between user trust and alert timeliness. Our results indicate a distinction between engineering and the user-centric definition of system accuracy. We found that timeliness is accuracy in the user’s mind. Overall, this study provides actionable insights for optimizing alert design, public education campaigns, and future behavioral research to improve the effectiveness of such systems in seismically active regions.

评分理由: 论文核心研究LLMs在多实例处理（MIP）中的性能退化问题，直接聚焦于LLMs的基础能力评估，因此与"Large Language Models"高度相关（10分）。研究分析了上下文长度（context length）对性能的影响，与"Context Window Extension"有一定关联（5分），但论文主要关注性能退化现象而非专门扩展上下文窗口的技术。其他关键词涉及具体技术（如MoE、量化、推理加速等）、训练方法（预训练、微调等）、应用领域（科学AI）或高级能力（推理、代理等），论文均未涉及，故评0分。

用户在处理多份文档或对多个实例进行分析时，常依赖大语言模型（Large Language Models, LLMs）。例如，分析一系列电影评论的整体情感需要大语言模型分别处理每一条评论的情感，以提供最终的聚合答案。尽管大语言模型在此类单项任务上表现通常优异，但关于其处理多实例输入时的表现研究尚少。本文针对大语言模型在个体表现优秀的任务中，对其多实例处理（Multi-Instance Processing, MIP）能力进行了全面评估。结果显示，所有大语言模型在处理少量实例（约20-100个）时均呈现轻微的性能下降趋势，而在处理更大数量实例时则出现性能崩溃。关键的是，我们的分析表明，虽然上下文长度与此性能下降相关，但实例数量对最终结果的影响更为显著。这一发现提示，在为大语言模型优化多实例处理性能时，应同时关注上下文长度，尤其是实例数量。

Users often rely on Large Language Models (LLMs) for processing multiple documents or performing analysis over a number of instances. For example, analysing the overall sentiment of a number of movie reviews requires an LLM to process the sentiment of each review individually in order to provide a final aggregated answer. While LLM performance on such individual tasks is generally high, there has been little research on how LLMs perform when dealing with multi-instance inputs. In this paper, we perform a comprehensive evaluation of the multi-instance processing (MIP) ability of LLMs for tasks in which they excel individually. The results show that all LLMs follow a pattern of slight performance degradation for small numbers of instances (approximately 20-100), followed by a performance collapse on larger instance counts. Crucially, our analysis shows that while context length is associated with this degradation, the number of instances has a stronger effect on the final results. This finding suggests that when optimising LLM performance for MIP, attention should be paid to both context length and, in particular, instance count.

评分理由: 论文TorR专注于计算机视觉领域的任务导向目标检测（TOOD）在边缘设备上的高效部署，核心是算法-架构协同设计，使用超维计算（HDC）替代CLIP进行关联推理，并优化缓存和计算路径以实现实时、低能耗。论文与大多数关键词无关，因为这些关键词主要针对大语言模型（LLM）的技术原理、训练方法、推理优化、对齐、代理系统等。仅与三个关键词有弱关联：1）“Small Language Models” OR “SLMs” OR “On-device AI”（5分）：论文关注边缘部署（on-device），但针对的是视觉任务，非语言模型。2）“Quantization” OR “Model Compression” OR “Low-bit Weights”（5分）：论文涉及精度门控（precision gating）和位切片内存，属于模型压缩/高效计算范畴，但非核心。3）“Speculative Decoding” OR “Inference Acceleration”（5分）：论文优化推理延迟和能耗，属于推理加速，但方法特定（缓存、路径调度），非典型LLM解码加速。其他关键词如LLMs、MoE、Scaling Laws、训练方法、对齐、RAG、CoT、代理等均未涉及。

基于CLIP的任务导向物体检测（TOOD）具备开放词汇和提示驱动的语义理解能力，但其密集的逐窗口计算与高内存流量阻碍了在实时、功耗受限的边缘设备上的部署。本文提出TorR，一种受脑启发的算法-架构协同设计方案，其利用超维（HDC）关联推理器替代CLIP风格的密集对齐机制，并将时间一致性转化为计算重用。在算法层面，TorR将对齐问题重构为超维相似度计算与图组合，并通过以下方式引入部分相似度重用：（i）采用查询缓存与逐类别分数累积；（ii）当仅少量超向量位发生变化时执行精确的$δ$更新；（iii）在高系统负载下启用基于相似度/负载的门控旁路。在架构层面，TorR实现了一个支持车道缩放、位切片化的项目存储器（item memory），配备存储体/精度门控机制，以及一个轻量级控制器，该控制器根据物体数量动态调度旁路/$δ$/完整计算路径，以满足RT-30/RT-60的实时性目标。基于TSMC 28纳米工艺综合并通过周期精确模拟器验证，TorR能够维持实时吞吐量，每窗口能耗在毫焦耳级别（60帧/秒时约50毫焦；30帧/秒时约113毫焦），且延迟抖动较低；同时在五项任务提示（task prompts）上实现了具有竞争力的平均精度AP@0.5（均值44.27%），与强大的视觉语言模型（VLM）基线相比性能差距有限，但能耗降低数个数量级。该设计提供了部署时可配置参数（有效维度$D’$、阈值、精度），以便根据边缘设备的资源预算在精度、延迟和能耗之间进行权衡。

Task-oriented object detection (TOOD) atop CLIP offers open-vocabulary, prompt-driven semantics, yet dense per-window computation and heavy memory traffic hinder real-time, power-limited edge deployment. We present \emph{TorR}, a brain-inspired \textbf{algorithm–architecture co-design} that \textbf{replaces CLIP-style dense alignment with a hyperdimensional (HDC) associative reasoner} and turns temporal coherence into reuse. On the \emph{algorithm} side, TorR reformulates alignment as HDC similarity and graph composition, introducing \emph{partial-similarity reuse} via (i) query caching with per-class score accumulation, (ii) exact $δ$-updates when only a small set of hypervector bits change, and (iii) similarity/load-gated bypass under high system load. On the \emph{architecture} side, TorR instantiates a lane-scalable, bit-sliced item memory with bank/precision gating and a lightweight controller that schedules bypass/$δ$/full paths to meet RT-30/RT-60 targets as object counts vary. Synthesized in a TSMC 28,nm process and exercised with a cycle-accurate simulator, TorR sustains real-time throughput with millijoule-scale energy per window ($\approx$50,mJ at 60,FPS; $\approx$113,mJ at 30,FPS) and low latency jitter, while delivering competitive AP@0.5 across five task prompts (mean 44.27%) within a bounded margin to strong VLM baselines, but at orders-of-magnitude lower energy. The design exposes deployment-time configurability (effective dimension $D’$, thresholds, precision) to trade accuracy, latency, and energy for edge budgets.

评分理由: 论文提出了一种基于Transformer架构的预训练方法（P2T3）用于社交媒体谣言检测，主要涉及预训练技术（Pre-training）和Transformer模型（与LLMs相关）。摘要明确提到“pre-trains on large-scale unlabeled datasets”，因此与“Pre-training”高度相关（8分）。论文使用Transformer架构，虽然未明确提及LLMs，但Transformer是LLMs的核心组件，且摘要提到“potentially offers a large model”，因此与“Large Language Models”有一定关联（3分）。其他关键词如MoE、SFT、RLHF、RAG等均未在论文中涉及，因此得0分。论文属于AI应用（社交媒体分析），但未涉及生物信息学等具体科学领域，因此“AI for Science”得0分。

基于深度学习的谣言检测技术通常利用图神经网络（GNN）来分析帖子关系。然而，这些方法在处理谣言传播结构时，因过度平滑问题而表现不佳，导致性能下降。我们对这一问题的研究发现，过度平滑本质上与谣言传播树的结构特征相关，其中大多数节点为单层节点。此外，图神经网络难以捕捉这些树结构中的长程依赖关系。为应对这些挑战，我们提出了一种基于纯Transformer架构的预训练传播树Transformer（P2T3）方法。该方法沿回复的传播方向从树结构中提取所有对话链，利用词元级嵌入注入连接信息并引入必要的归纳偏置，并在大规模无标注数据集上进行预训练。实验表明，P2T3在多个基准数据集上超越了以往的最优方法，并在少样本条件下表现良好。P2T3不仅避免了图神经网络固有的过度平滑问题，还可能为未来社交媒体研究提供大型模型或统一的多模态方案。

Deep learning techniques for rumor detection typically utilize Graph Neural Networks (GNNs) to analyze post relations. These methods, however, falter due to over-smoothing issues when processing rumor propagation structures, leading to declining performance. Our investigation into this issue reveals that over-smoothing is intrinsically tied to the structural characteristics of rumor propagation trees, in which the majority of nodes are 1-level nodes. Furthermore, GNNs struggle to capture long-range dependencies within these trees. To circumvent these challenges, we propose a Pre-Trained Propagation Tree Transformer (P2T3) method based on pure Transformer architecture. It extracts all conversation chains from a tree structure following the propagation direction of replies, utilizes token-wise embedding to infuse connection information and introduces necessary inductive bias, and pre-trains on large-scale unlabeled datasets. Experiments indicate that P2T3 surpasses previous state-of-the-art methods in multiple benchmark datasets and performs well under few-shot conditions. P2T3 not only avoids the over-smoothing issue inherent in GNNs but also potentially offers a large model or unified multi-modal scheme for future social media research.

评分理由: 论文专注于分子点云生成的扩散模型方法，提出了一种在商流形上直接建模扩散的新技术，以处理排列对称性。所有关键词中，只有"AI for Science” OR “Bioinformatics” OR “Cheminformatics"高度相关（10分），因为论文涉及AI在科学（具体是化学信息学/分子生成）中的应用。其他关键词主要关于大语言模型（LLMs）及其相关技术（如微调、对齐、推理、代理等），而本文研究的是扩散模型用于分子生成，不涉及语言模型或这些特定LLM技术，因此评分为0分。

置换不变性是分子点云生成的基础，然而大多数扩散模型通过在有序空间上使用置换等变网络间接实现这一性质。我们提出直接在商流形 $\tilde{\calX}=\sR^{d\times N}/S_N$ 上建模扩散过程，其中所有原子置换均被视为同一元素。我们证明了 $\tilde{\calX}$ 上的热核具有显式表达式，即所有置换下欧几里得热核的求和，这阐明了商空间上的扩散与有序粒子扩散的本质差异。训练过程需要涉及 $S_N$ 上难以处理的求和项的置换对称化分数；我们推导了基于置换后验的期望形式，并利用置换空间中的马尔可夫链蒙特卡洛方法进行近似。我们在 EQGAT-Diff 协议下，使用 SemlaFlow 风格的主干网络并连续处理所有变量，对 QM9 数据集的无条件三维分子生成任务进行了评估。结果表明，基于商空间的置换对称化方法具有实用性，能够在提升效率的同时生成具有竞争力的分子结构。

Permutation invariance is fundamental in molecular point-cloud generation, yet most diffusion models enforce it indirectly via permutation-equivariant networks on an ordered space. We propose to model diffusion directly on the quotient manifold $\tilde{\calX}=\sR^{d\times N}/S_N$, where all atom permutations are identified. We show that the heat kernel on $\tilde{\calX}$ admits an explicit expression as a sum of Euclidean heat kernels over permutations, which clarifies how diffusion on the quotient differs from ordered-particle diffusion. Training requires a permutation-symmetrized score involving an intractable sum over $S_N$; we derive an expectation form over a posterior on permutations and approximate it using MCMC in permutation space. We evaluate on unconditional 3D molecule generation on QM9 under the EQGAT-Diff protocol, using SemlaFlow-style backbone and treating all variables continuously. The results demonstrate that quotient-based permutation symmetrization is practical and yields competitive generation quality with improved efficiency.

评分理由: 该论文专注于计算机视觉和医学图像处理领域，提出了一种基于扩散模型的去烟方法（PhySe-RPO），用于手术视频的烟雾去除。论文的核心技术是扩散模型、强化学习优化（Relative Policy Optimization）和计算机视觉技术（CLIP-based语义奖励），并未涉及任何大语言模型（LLM）或相关技术。所有关键词均与大语言模型、其训练方法、优化技术、推理技术、代理系统或通用AI技术直接相关，而本文研究的是特定领域的视觉任务，因此除最后一个关键词（“AI for Science”）外，其余均完全无关。“AI for Science” 得5分，因为手术烟雾去除可视为AI在医学科学（外科手术）中的一个应用，但论文并未明确强调"Bioinformatics"或"Cheminformatics”。

手术烟雾严重降低术中视频质量，模糊解剖结构并限制手术感知。现有基于学习的去雾方法依赖稀缺的配对监督和确定性修复流程，难以在真实手术条件下进行探索或强化驱动的优化。我们提出PhySe-RPO——一种通过物理与语义引导的相对策略优化（Physics- and Semantics-Guided Relative Policy Optimization）进行优化的扩散修复框架。其核心思想是将确定性修复转化为随机策略，通过组间相对优化实现轨迹级探索与无评论家更新。物理引导奖励函数施加光照与色彩一致性约束，同时基于CLIP学习的手术视觉概念语义奖励促进无烟雾且解剖结构连贯的修复结果。结合无参考感知约束，PhySe-RPO在合成与真实机器人手术数据集上产生物理一致、语义可靠且具有临床可解释性的结果，为有限配对监督下实现基于扩散模型的鲁棒修复提供了理论化路径。

Surgical smoke severely degrades intraoperative video quality, obscuring anatomical structures and limiting surgical perception. Existing learning-based desmoking approaches rely on scarce paired supervision and deterministic restoration pipelines, making it difficult to perform exploration or reinforcement-driven refinement under real surgical conditions. We propose PhySe-RPO, a diffusion restoration framework optimized through Physics- and Semantics-Guided Relative Policy Optimization. The core idea is to transform deterministic restoration into a stochastic policy, enabling trajectory-level exploration and critic-free updates via group-relative optimization. A physics-guided reward imposes illumination and color consistency, while a visual-concept semantic reward learned from CLIP-based surgical concepts promotes smoke-free and anatomically coherent restoration. Together with a reference-free perceptual constraint, PhySe-RPO produces results that are physically consistent, semantically faithful, and clinically interpretable across synthetic and real robotic surgical datasets, providing a principled route to robust diffusion-based restoration under limited paired supervision.

评分理由: 论文核心研究大语言模型在性别推断任务中的上下文不变性失败问题，与’Large Language Models’高度相关（10分）。研究涉及模型输出的事实性和可解释性，与’Hallucination Mitigation’和’Mechanistic Interpretability’有一定关联（各5分）。其他关键词如MoE、SLMs、训练方法、推理加速、AI for Science等均未在论文中涉及，得0分。

标准评估方法通常假定大型语言模型（LLM）的输出在任务表述语境等价时保持稳定。本文在性别推断任务中检验这一假设。通过一个受控的代词选择任务，我们引入了理论上无信息量的最小语篇语境，发现这会引发模型输出出现显著且系统性的偏移。在无语境条件下存在的文化性别刻板印象相关性，在引入语境后减弱或消失；而理论上无关的特征（例如无关指代对象的代词性别）则成为预测模型行为最具信息量的指标。通过默认语境性分析发现，在不同模型中，19%至52%的案例在排除语境对单个输出的所有边际效应后，这种依赖性依然存在，且无法简单归因于代词重复。这些结果表明，即使在句法表述近乎相同的条件下，LLM输出仍会违反语境不变性，这对偏见基准测试及高风险场景中的模型部署具有重要启示。

Standard evaluation practices assume that large language model (LLM) outputs are stable under contextually equivalent formulations of a task. Here, we test this assumption in the setting of gender inference. Using a controlled pronoun selection task, we introduce minimal, theoretically uninformative discourse context and find that this induces large, systematic shifts in model outputs. Correlations with cultural gender stereotypes, present in decontextualized settings, weaken or disappear once context is introduced, while theoretically irrelevant features, such as the gender of a pronoun for an unrelated referent, become the most informative predictors of model behaviour. A Contextuality-by-Default analysis reveals that, in 19–52% of cases across models, this dependence persists after accounting for all marginal effects of context on individual outputs and cannot be attributed to simple pronoun repetition. These findings show that LLM outputs violate contextual invariance even under near-identical syntactic formulations, with implications for bias benchmarking and deployment in high-stakes settings.

评分理由: 论文核心研究大视觉语言模型（LVLMs）的效率优化，通过稀疏化图像与文本token的交互来减少计算成本，属于大模型技术原理的创新。高度相关关键词：1）‘Large Language Models’（论文明确研究LVLMs，是LLMs的视觉扩展）；2）‘Mixture of Experts OR MoE OR Sparse Models’（论文采用稀疏交互和动态选择机制，与稀疏模型概念相关）；3）‘Speculative Decoding OR Inference Acceleration’（论文直接优化推理效率，降低计算成本）。其他关键词如SLMs、训练方法、对齐、代理等均未涉及，AI for Science虽属应用领域但论文未明确针对科学领域，故评0分。

现有提升大型视觉语言模型效率的方法主要基于视觉令牌缩减的概念。然而，这种方法会形成信息瓶颈，从而损害模型性能，尤其是在需要细粒度理解和推理的复杂任务上。在本研究中，我们通过引入“按需视觉”方法挑战这一范式，该方法能在不丢弃视觉信息的前提下降低推理成本。VISOR并非压缩图像，而是通过稀疏化图像与文本令牌之间的交互来提高效率。具体而言，语言模型通过少量策略性放置的注意力层来处理完整的高分辨率视觉令牌集：高效的文本-图像交叉注意力层提供通用的视觉上下文，而少数精心布置且动态选择的自注意力层则对视觉表征本身进行细化，从而在需要时实现复杂的高分辨率推理。基于此原理，我们首先通过调整自注意力层数量，在多种计算预算下训练一个通用的单一网络，随后引入轻量级策略机制，根据每个样本的复杂度动态分配视觉计算资源。大量实验表明，VISOR在显著降低计算成本的同时，在一系列多样化基准测试中达到或超越了现有最优性能，并在需要精细视觉理解的挑战性任务中表现卓越。

Existing approaches for improving the efficiency of Large Vision-Language Models (LVLMs) are largely based on the concept of visual token reduction. This approach, however, creates an information bottleneck that impairs performance, especially on challenging tasks that require fine-grained understanding and reasoning. In this work, we challenge this paradigm by introducing VISion On Request (VISOR), a method that reduces inference cost without discarding visual information. Instead of compressing the image, VISOR improves efficiency by sparsifying the interaction between image and text tokens. Specifically, the language model attends to the full set of high-resolution visual tokens through a small, strategically placed set of attention layers: general visual context is provided by efficient cross-attention between text-image, while a few well-placed and dynamically selected self-attention layers refine the visual representations themselves, enabling complex, high-resolution reasoning when needed. Based on this principle, we first train a single universal network on a range of computational budgets by varying the number of self-attention layers, and then introduce a lightweight policy mechanism that dynamically allocates visual computation based on per-sample complexity. Extensive experiments show that VISOR drastically reduces computational cost while matching or exceeding state-of-the-art results across a diverse suite of benchmarks, and excels in challenging tasks that require detailed visual understanding.

评分理由: 论文核心是开发一个利用多个LLM提供商（OpenAI GPT、Anthropic Claude、Groq）自动化软件需求工程的多提供商框架，因此与’Large Language Models OR LLMs OR Foundation Models’高度相关（10分）。论文涉及AI在软件工程领域的应用，与’AI for Science OR Bioinformatics OR Cheminformatics’有一定关联（5分），因为软件工程可视为广义的AI应用领域。其他关键词（如MoE、SFT、RAG等）在摘要中未提及，与论文内容完全无关（0分）。

需求工程是软件开发过程中至关重要但劳动密集的环节。本文介绍ReqFusion：一种利用多个大型语言模型（Large Language Model，简称LLM）提供商、通过人工智能增强的自动化软件需求提取、分类与分析系统。ReqFusion的架构整合了OpenAI GPT、Anthropic Claude和Groq模型，能够从学术、工业及招标提案等场景下的多种文档格式（PDF、DOCX和PPTX）中提取功能性与非功能性需求。该系统采用与领域无关的提取方法，并遵循Bertrand Meyer提出的项目、环境、目标与系统（Project, Environment, Goal, and System，简称PEGS）框架生成需求。其核心思想在于：由于PEGS格式描述详尽，大型语言模型能获得更多关于需求的信息与线索，从而比简单的通用请求产生更优结果。一项消融研究证实了该假设：在相同的多提供商配置下，采用PEGS引导的提示方法取得了0.88的F1分数，而通用提示方法仅为0.71。评估使用18份真实世界文档，通过自动分类生成226条需求，涵盖学术、商业和技术领域，其中功能性需求占54.9%，非功能性需求占45.1%。对五个项目共1,050条需求的扩展评估表明，与人工方法相比，系统在提取准确性上显著提升，且分析时间减少了78%。多提供商架构通过模型共识与回退机制增强了可靠性，而基于PEGS的方法则确保全面覆盖所有需求类别。

Requirements engineering is a vital, yet labor-intensive, stage in the software development process. This article introduces ReqFusion: an AI-enhanced system that automates the extraction, classification, and analysis of software requirements utilizing multiple Large Language Model (LLM) providers. The architecture of ReqFusion integrates OpenAI GPT, Anthropic Claude, and Groq models to extract functional and non-functional requirements from various documentation formats (PDF, DOCX, and PPTX) in academic, industrial, and tender proposal contexts. The system uses a domain-independent extraction method and generates requirements following the Project, Environment, Goal, and System (PEGS) approach introduced by Bertrand Meyer. The main idea is that, because the PEGS format is detailed, LLMs have more information and cues about the requirements, producing better results than a simple generic request. An ablation study confirms this hypothesis: PEGS-guided prompting achieves an F1 score of 0.88, compared to 0.71 for generic prompting under the same multi-provider configuration. The evaluation used 18 real-world documents to generate 226 requirements through automated classification, with 54.9% functional and 45.1% nonfunctional across academic, business, and technical domains. An extended evaluation on five projects with 1,050 requirements demonstrated significant improvements in extraction accuracy and a 78% reduction in analysis time compared to manual methods. The multi-provider architecture enhances reliability through model consensus and fallback mechanisms, while the PEGS-based approach ensures comprehensive coverage of all requirement categories.

评分理由: 论文VTAM专注于具身智能中的多模态世界建模，核心创新在于将触觉感知整合到视频-动作模型中，以提升接触丰富场景下的物理交互性能。与大多数关键词（如LLM、推理、对齐、压缩等）无关，因为这些关键词主要针对语言模型或特定NLP技术。相关关键词：1) ‘World Models AND General World Models’（10分）：论文明确提出了’world modeling framework’，是核心内容；2) ‘Pre-training OR Continual Pre-training OR Domain Adaptation’（5分）：使用了预训练视频变换器并进行微调；3) ‘Post-training OR Supervised Fine-tuning OR SFT’（5分）：涉及轻量级模态转移微调；4) ‘PEFT OR LoRA OR Parameter-efficient Fine-tuning’（5分）：通过轻量级微调实现跨模态表示学习，符合参数高效微调概念。其他关键词均未涉及。

视频动作模型已成为具身智能领域一个有前景的框架，它从原始视频流中学习隐式的世界动态，以产生时序一致的动作预测。尽管此类模型通过视觉推理在长时程任务上展现出强大性能，但在接触密集的场景中仍存在局限，因为关键的交互状态仅凭视觉往往无法完全观测。具体而言，视觉令牌无法可靠地编码细粒度的力调节与接触状态转换，导致行为不稳定或不精确。为弥补这一差距，我们提出了视频-触觉动作模型，这是一种多模态世界建模框架，将触觉感知作为补充的接地信号纳入其中。VTAM通过轻量级的模态迁移微调，将触觉流集成到预训练的视频Transformer中，实现了无需触觉-语言配对数据或独立触觉预训练的高效跨模态表征学习。为稳定多模态融合，我们引入了一种触觉正则化损失，以强制实现平衡的跨模态注意力，防止动作模型中视觉潜在表征占据主导。VTAM在接触密集的操作任务中展现出卓越性能，平均保持90%的鲁棒成功率。在需要高保真力感知的挑战性场景（如薯片抓取放置）中，VTAM比π0.5基线模型性能高出80%。我们的研究结果表明，整合触觉反馈对于纠正世界动作模型中的视觉估计误差至关重要，为构建物理接地的具身基础模型提供了一条可扩展的路径。

Video-Action Models (VAMs) have emerged as a promising framework for embodied intelligence, learning implicit world dynamics from raw video streams to produce temporally consistent action predictions. Although such models demonstrate strong performance on long-horizon tasks through visual reasoning, they remain limited in contact-rich scenarios where critical interaction states are only partially observable from vision alone. In particular, fine-grained force modulation and contact transitions are not reliably encoded in visual tokens, leading to unstable or imprecise behaviors. To bridge this gap, we introduce the Video-Tactile Action Model (VTAM), a multimodal world modeling framework that incorporates tactile perception as a complementary grounding signal. VTAM augments a pretrained video transformer with tactile streams via a lightweight modality transfer finetuning, enabling efficient cross-modal representation learning without tactile-language paired data or independent tactile pretraining. To stabilize multimodal fusion, we introduce a tactile regularization loss that enforces balanced cross-modal attention, preventing visual latent dominance in the action model. VTAM demonstrates superior performance in contact-rich manipulation, maintaining a robust success rate of 90 percent on average. In challenging scenarios such as potato chip pick-and-place requiring high-fidelity force awareness, VTAM outperforms the pi 0.5 baseline by 80 percent. Our findings demonstrate that integrating tactile feedback is essential for correcting visual estimation errors in world action models, providing a scalable approach to physically grounded embodied foundation models.

评分理由: 论文专注于扩散模型在图像修复中的应用，提出了一种名为InverFill的单步反演方法，旨在提高少步采样的修复质量。所有评分关键词均与大语言模型（LLMs）或相关技术（如MoE、SFT、RAG、量化等）直接相关，而本文研究的是扩散模型（一种生成模型），并非大语言模型或其衍生技术。因此，论文与所有关键词完全无关，每个关键词得分为0。

近期基于扩散模型的图像修复技术虽能实现照片级真实感，但需要大量采样步骤，限制了实际应用。少步数的文生图模型虽能加速生成，但直接将其应用于修复任务会导致背景与修复区域之间协调性差并产生伪影。我们发现其根源在于随机高斯噪声初始化——在低函数评估次数下，这种初始化会导致语义错位并降低生成保真度。为解决这一问题，我们提出InverFill：一种专为修复任务设计的一步反演方法，该方法将输入掩码图像的语义信息注入初始噪声，从而实现高保真度的少步数图像修复。InverFill无需专门训练修复模型，而是将语义对齐的噪声作为输入，在混合采样流程中利用少步数文生图模型进行生成。该方法显著改进了原始混合采样效果，在低NFEs（噪声函数评估次数）条件下甚至可媲美专用修复模型。此外，InverFill无需真实图像监督，仅增加极少的推理开销。大量实验表明，InverFill能持续提升基线少步数模型的性能，在无需昂贵重训练或繁重迭代优化的情况下，同步提升图像质量与文本语义一致性。

Recent diffusion-based models achieve photorealism in image inpainting but require many sampling steps, limiting practical use. Few-step text-to-image models offer faster generation, but naively applying them to inpainting yields poor harmonization and artifacts between the background and inpainted region. We trace this cause to random Gaussian noise initialization, which under low function evaluations causes semantic misalignment and reduced fidelity. To overcome this, we propose InverFill, a one-step inversion method tailored for inpainting that injects semantic information from the input masked image into the initial noise, enabling high-fidelity few-step inpainting. Instead of training inpainting models, InverFill leverages few-step text-to-image models in a blended sampling pipeline with semantically aligned noise as input, significantly improving vanilla blended sampling and even matching specialized inpainting models at low NFEs. Moreover, InverFill does not require real-image supervision and only adds minimal inference overhead. Extensive experiments show that InverFill consistently boosts baseline few-step models, improving image quality and text coherence without costly retraining or heavy iterative optimization.

评分理由: 论文核心是开发3DCity-LLM框架，将多模态大语言模型应用于3D城市规模感知与理解，因此与’Large Language Models’高度相关（10分）。论文创建了1.2M高质量数据集，涉及数据质量和扩展，与’Scaling Laws AND Data Quality’有一定关联（5分）。论文涉及模型训练，与’Pre-training’和’Post-training’类别有一定关联（各5分）。其他关键词如MoE、SLMs、对齐、推理加速、科学AI等均未在摘要中提及，评为0分。

尽管多模态大语言模型在以物体为中心或室内场景中表现出色，但将其扩展至三维城市尺度环境仍是一项艰巨挑战。为弥合这一差距，我们提出了3DCity-LLM——一个为三维城市尺度视觉语言感知与理解设计的统一框架。该框架采用由粗到细的特征编码策略，包含目标物体、物体间关系与全局场景三个并行分支。为支持大规模训练，我们构建了3DCity-LLM-1.2M数据集，该数据集涵盖七个代表性任务类别约120万个高质量样本，范围涵盖细粒度物体分析到多维度场景规划。经过严格质量控制的该数据集整合了显式三维数值信息与多样化的用户导向模拟，增强了城市场景问答的多样性与真实感。此外，我们采用基于文本相似度度量与大语言模型语义评估的多维评估协议，以确保对所有方法进行忠实且全面的评估。在两个基准测试上的大量实验表明，3DCity-LLM显著优于现有最先进方法，为推进空间推理与城市智能发展提供了具有前景的重要方向。源代码与数据集已发布于https://github.com/SYSU-3DSTAILab/3D-City-LLM。

While multi-modality large language models excel in object-centric or indoor scenarios, scaling them to 3D city-scale environments remains a formidable challenge. To bridge this gap, we propose 3DCity-LLM, a unified framework designed for 3D city-scale vision-language perception and understanding. 3DCity-LLM employs a coarse-to-fine feature encoding strategy comprising three parallel branches for target object, inter-object relationship, and global scene. To facilitate large-scale training, we introduce 3DCity-LLM-1.2M dataset that comprises approximately 1.2 million high-quality samples across seven representative task categories, ranging from fine-grained object analysis to multi-faceted scene planning. This strictly quality-controlled dataset integrates explicit 3D numerical information and diverse user-oriented simulations, enriching the question-answering diversity and realism of urban scenarios. Furthermore, we apply a multi-dimensional protocol based on text-similarity metrics and LLM-based semantic assessment to ensure faithful and comprehensive evaluations for all methods. Extensive experiments on two benchmarks demonstrate that 3DCity-LLM significantly outperforms existing state-of-the-art methods, offering a promising and meaningful direction for advancing spatial reasoning and urban intelligence. The source code and dataset are available at https://github.com/SYSU-3DSTAILab/3D-City-LLM.

评分理由: 论文主要研究AI代理在代码审查任务中的评估，与’LLM Agents OR Autonomous Agents OR Agentic Workflow’高度相关（10分），因为论文明确提到’Software engineering agents’、‘code review agents’和’AI agents’，并评估了多个代理系统。与’Large Language Models OR LLMs OR Foundation Models’有一定关联（8分），因为论文评估的商业代理（如Claude Code、Codex）可能基于大语言模型，但论文未深入讨论模型技术细节。其他关键词与论文内容无关（0分），因为论文聚焦于代码审查代理的基准测试和评估框架，不涉及模型架构、训练方法、推理优化、对齐技术或特定科学领域应用。

软件工程智能体在代码编写方面已展现出显著潜力。随着AI智能体广泛渗透代码编写领域并自动生成海量代码，代码质量问题已成为核心关注点。当自动生成的代码被集成至大型代码库时，代码审查及广义的质量保障问题变得至关重要。本文以全新视角审视该问题，构建了供AI智能体使用的代码审查数据集。我们提出的数据集c-CRAB（发音为see-crab）能够评估智能体在代码审查任务中的表现。具体而言，给定一个拉取请求（可能来自代码生成智能体或人类），若代码审查智能体生成审查意见，我们的评估框架可对其审查能力进行量化评估。该框架已用于评估当前前沿技术——包括开源PR-agent，以及Devin、Claude Code和Codex等商业代码审查智能体。

c-CRAB数据集系统性地构建于人类审查记录之上：针对拉取请求实例的人类审查意见，我们生成相应测试用例以评估代码审查智能体生成的审查。这种基准构建方式带来多项发现：首先，现有审查智能体整体仅能解决约40%的c-CRAB任务，表明未来研究具有填补此空白的潜力；其次，我们观察到智能体审查常关注与人类审查不同的维度，这预示着未来软件团队可能部署人机协同的代码审查模式；最后，基于数据集生成的智能体测试用例可作为独立测试套件，从而构成智能体生成审查的质量关卡。这对于未来代码生成智能体、测试生成智能体与代码审查智能体之间的协作将产生何种影响，仍有待进一步探索。

Software engineering agents have shown significant promise in writing code. As AI agents permeate code writing, and generate huge volumes of code automatically – the matter of code quality comes front and centre. As the automatically generated code gets integrated into huge code-bases – the issue of code review and broadly quality assurance becomes important. In this paper, we take a fresh look at the problem and curate a code review dataset for AI agents to work with. Our dataset called c-CRAB (pronounced see-crab) can evaluate agents for code review tasks. Specifically given a pull-request (which could be coming from code generation agents or humans), if a code review agent produces a review, our evaluation framework can asses the reviewing capability of the code review agents. Our evaluation framework is used to evaluate the state of the art today – the open-source PR-agent, as well as commercial code review agents from Devin, Claude Code, and Codex. Our c-CRAB dataset is systematically constructed from human reviews – given a human review of a pull request instance we generate corresponding tests to evaluate the code review agent generated reviews. Such a benchmark construction gives us several insights. Firstly, the existing review agents taken together can solve only around 40% of the c-CRAB tasks, indicating the potential to close this gap by future research. Secondly, we observe that the agent reviews often consider different aspects from the human reviews – indicating the potential for human-agent collaboration for code review that could be deployed in future software teams. Last but not the least, the agent generated tests from our data-set act as a held out test-suite and hence quality gate for agent generated reviews. What this will mean for future collaboration of code generation agents, test generation agents and code review agents – remains to be investigated.

评分理由: 论文核心研究LLM在软件测试生成中的应用，与"Large Language Models OR LLMs OR Foundation Models"高度相关（10分），但未涉及其他关键词的具体技术（如MoE、SFT、RAG等）或科学领域应用（如生物信息学），因此其他关键词评分为0分。

大语言模型（LLM）在自动化单元测试生成中的应用日益广泛。然而，目前尚不清楚这些测试是否真正体现了对程序行为的推理，还是仅仅复现了训练过程中学到的表层模式。若后者占主导，LLM生成的测试可能表现出覆盖率降低、遗漏回归问题以及未能检测到缺陷等弱点。因此，理解LLM如何生成测试以及这些测试如何响应代码演化至关重要。本文针对程序变更下基于LLM的测试生成进行了大规模实证研究。通过一个自动化的变异驱动框架，我们分析了八个LLM在22,374个程序变体上生成的测试如何响应语义变更（SAC）和语义保持变更（SPC）。

LLM在基准测试中取得了强劲的结果：在原始程序上，完全通过的测试套件实现了79%的行覆盖率和76%的分支覆盖率。然而，随着程序演化，其性能出现下降。在SAC下，新生成测试的通过率降至66%，分支覆盖率下降至60%。超过99%的失败SAC测试在原始程序上执行修改区域时能够通过，这表明测试仍与原始行为保持残留对齐，而非适应更新后的语义。尽管SPC不改变功能，性能同样出现下降：通过率降至79%，分支覆盖率降至69%。虽然SPC编辑保留了语义，但它们常引入较大的语法变化，导致生成的测试套件不稳定。模型在丢弃大量基准测试的同时生成了更多新测试，这表明其对词汇变化敏感，而非真正理解语义影响。总体而言，我们的结果表明，当前基于LLM的测试生成严重依赖表层线索，且在程序演化过程中难以保持回归意识。

Large Language Models (LLMs) are increasingly used for automated unit test generation. However, it remains unclear whether these tests reflect genuine reasoning about program behavior or simply reproduce superficial patterns learned during training. If the latter dominates, LLM-generated tests may exhibit weaknesses such as reduced coverage, missed regressions, and undetected faults. Understanding how LLMs generate tests and how those tests respond to code evolution is therefore essential. We present a large-scale empirical study of LLM-based test generation under program changes. Using an automated mutation-driven framework, we analyze how generated tests react to semantic-altering changes (SAC) and semantic-preserving changes (SPC) across eight LLMs and 22,374 program variants. LLMs achieve strong baseline results, reaching 79% line coverage and 76% branch coverage with fully passing test suites on the original programs. However, performance degrades as programs evolve. Under SACs, the pass rate of newly generated tests drops to 66%, and branch coverage declines to 60%. More than 99% of failing SAC tests pass on the original program while executing the modified region, indicating residual alignment with the original behavior rather than adaptation to updated semantics. Performance also declines under SPCs despite unchanged functionality: pass rates fall to 79% and branch coverage to 69%. Although SPC edits preserve semantics, they often introduce larger syntactic changes, leading to instability in generated test suites. Models generate more new tests while discarding many baseline tests, suggesting sensitivity to lexical changes rather than true semantic impact. Overall, our results indicate that current LLM-based test generation relies heavily on surface-level cues and struggles to maintain regression awareness as programs evolve.

评分理由: 该论文研究物联网网络中的入侵检测系统对抗性攻击，属于网络安全领域，主要涉及传统机器学习算法在物联网安全中的应用、对抗性攻击的可行性评估以及防御机制设计。论文内容完全不涉及大语言模型、深度学习技术原理创新、大模型在不同领域的应用或任何评分关键词中的技术概念，因此所有关键词相关度均为0分。

机器学习（ML）算法与物联网（IoT）应用的融合在带来显著优势的同时，也引入了对抗性攻击的脆弱性，尤其是在基于物联网的入侵检测系统（IDS）中。尽管理论上的对抗性攻击已得到广泛研究，但实际实施中的限制条件常被忽视。本研究通过采用一种新颖的黑盒对抗性攻击方法，评估针对基于物联网网络的入侵检测系统实施逃避攻击的可行性，从而弥补了这一研究空白。我们的工作旨在将理论漏洞与现实世界的适用性联系起来，以深化对现代物联网生态系统中复杂威胁的理解与防御。此外，我们提出了一种定制化的防御方案，以减轻逃避攻击的影响，从而增强基于机器学习的入侵检测系统的韧性。我们的研究结果展示了对入侵检测系统成功的逃避攻击，突显了其对先进技术的易受性。相比之下，我们提出的防御机制通过有效检测大部分对抗性流量，展现出强大的性能，与当前最先进的防御方案相比取得了令人瞩目的成果。通过应对这些关键网络安全挑战，我们的研究为推动物联网安全发展做出了贡献，并为开发更具韧性的入侵检测系统提供了见解。

The integration of machine learning (ML) algorithms into Internet of Things (IoT) applications has introduced significant advantages alongside vulnerabilities to adversarial attacks, especially within IoT-based intrusion detection systems (IDS). While theoretical adversarial attacks have been extensively studied, practical implementation constraints have often been overlooked. This research addresses this gap by evaluating the feasibility of evasion attacks on IoT network-based IDSs, employing a novel black-box adversarial attack. Our study aims to bridge theoretical vulnerabilities with real-world applicability, enhancing understanding and defense against sophisticated threats in modern IoT ecosystems. Additionally, we propose a defense scheme tailored to mitigate the impact of evasion attacks, thereby reinforcing the resilience of ML-based IDSs. Our findings demonstrate successful evasion attacks against IDSs, underscoring their susceptibility to advanced techniques. In contrast, we proposed a defense mechanism that exhibits robust performance by effectively detecting the majority of adversarial traffic, showcasing promising outcomes compared to current state-of-the-art defenses. By addressing these critical cybersecurity challenges, our research contributes to advancing IoT security and provides insights for developing more resilient IDS.

评分理由: 论文研究AI代理（特别是LLM驱动的代理）在决策环境中如何被"机制性助推”（mecha-nudges）影响，与"LLM Agents OR Autonomous Agents OR Agentic Workflow"高度相关（10分），因为核心关注AI代理的行为和决策流程。与"Large Language Models OR LLMs OR Foundation Models"有一定关联（8分），因为论文提到ChatGPT的发布影响了Etsy列表中的机器可用信息，表明LLM是相关背景。其他关键词如MoE、SFT、RAG等涉及具体技术细节，论文未探讨，故均为0分。

助推（nudges）是指在不限制选项或改变激励的前提下，通过微妙调整向人类决策者呈现选择的方式（例如默认加入与默认退出），从而引导行为改变。随着人工智能（AI）代理在与人类相同的环境中越来越多地参与决策，选择呈现方式可能同时针对机器和人类进行优化。我们提出机械助推（mecha-nudges）：即通过改变选择呈现方式，系统性地影响AI代理，同时不损害人类的决策环境。为形式化机械助推，我们将贝叶斯劝说框架与V-可用信息（V-usable information）相结合，后者是香农信息的一种泛化形式，具有观察者相对性。这为比较各类干预措施、情境和模型提供了一个通用尺度（以可用信息的比特数为单位）。将我们的框架应用于全球独立卖家市场Etsy的商品列表分析，我们发现自ChatGPT发布以来，商品列表中包含的关于产品选择的机器可用信息显著增加，这与系统性的机械助推现象相符。

Nudges are subtle changes to the way choices are presented to human decision-makers (e.g., opt-in vs. opt-out by default) that shift behavior without restricting options or changing incentives. As AI agents increasingly make decisions in the same environments as humans, the presentation of choices may be optimized for machines as well as people. We introduce mecha-nudges: changes to how choices are presented that systematically influence AI agents without degrading the decision environment for humans. To formalize mecha-nudges, we combine the Bayesian persuasion framework with V-usable information, a generalization of Shannon information that is observer-relative. This yields a common scale (bits of usable information) for comparing a wide range of interventions, contexts, and models. Applying our framework to product listings on Etsy – a global marketplace for independent sellers – we find that following ChatGPT’s release, listings have significantly more machine-usable information about product selection, consistent with systematic mecha-nudging.

评分理由: 论文核心研究LLM驱动的自主研究系统（autoresearch），属于LLM Agents/Autonomous Agents领域，因此与’Large Language Models’和’LLM Agents’高度相关（10分）。论文未涉及其他关键词的具体技术（如MoE、SFT、RAG等），也未涉及科学领域应用，因此其他关键词得0分。

若自动研究本身即是一种研究形式，则自动研究可被应用于研究自身。我们对此概念进行字面解读：利用自动研究循环来优化自动研究循环。现有所有自动研究系统——从Karpathy的单轨循环到AutoResearchClaw的多批次扩展及EvoScientist的持久化记忆——皆经由人类阅读代码、识别瓶颈并编写新代码而改进。我们探究大型语言模型（LLM）能否自主完成相同工作。本文提出双层自动研究（Bilevel Autoresearch），该双层框架通过外层循环在运行时生成并注入以Python代码编写的新搜索机制，对内层自动研究循环进行元优化。内层循环优化目标任务；外层循环则优化内层循环的搜索方式。两层循环使用同一LLM——元层级无需更强模型。在Karpathy的GPT预训练基准测试中，元自动研究外层循环相较于标准内层循环实现了5倍性能提升（验证集每字节位数val_bpb从-0.009降至-0.045），而仅进行参数调整未改变机制则未产生可靠增益。外层循环自主发现了组合优化、多臂赌博机及实验设计等领域的机制——无需人工指定探索领域。这些机制通过打破内层循环的确定性搜索模式发挥作用，强制探索LLM先验系统回避的方向。其核心原理简明：若自动研究能对自身进行元自动研究，则原则上可对任何具有可测量目标的事物实施元自动研究。

If autoresearch is itself a form of research, then autoresearch can be applied to research itself. We take this idea literally: we use an autoresearch loop to optimize the autoresearch loop. Every existing autoresearch system – from Karpathy’s single-track loop to AutoResearchClaw’s multi-batch extension and EvoScientist’s persistent memory – was improved by a human who read the code, identified a bottleneck, and wrote new code. We ask whether an LLM can do the same, autonomously. We present Bilevel Autoresearch, a bilevel framework where an outer loop meta-optimizes the inner autoresearch loop by generating and injecting new search mechanisms as Python code at runtime. The inner loop optimizes the task; the outer loop optimizes how the inner loop searches. Both loops use the same LLM – no stronger model is needed at the meta level. On Karpathy’s GPT pretraining benchmark, the meta-autoresearch outer loop achieves a 5x improvement over the standard inner loop alone (-0.045 vs. -0.009 val_bpb), while parameter-level adjustment without mechanism change yields no reliable gain. The outer loop autonomously discovers mechanisms from combinatorial optimization, multi-armed bandits, and design of experiments – without human specification of which domains to explore. These mechanisms succeed by breaking the inner loop’s deterministic search patterns, forcing exploration of directions the LLM’s priors systematically avoid. The core principle is simple: if autoresearch can meta-autoresearch itself, it can, in principle, meta-autoresearch anything with a measurable objective.

评分理由: 论文研究多智能体人机系统中延迟反馈下的偏见性错误归因，仅与"Multi-agent Systems OR Agent Coordination"高度相关（10分），因为核心研究多智能体协调中的人类决策偏见；其他关键词均未涉及大模型技术、训练方法、推理优化、科学应用等具体技术，完全无关（0分）。

人类决策深受认知偏差影响，在不确定性和风险条件下尤为显著。尽管已有研究探讨了即时结果的单步决策以及人类与单一自主智能体互动中的偏差，但对于涉及多智能体且结果延迟的决策情境关注相对不足——此类情境中每一步决策都会影响后续状态。本研究通过多智能体人机协同任务，探究延迟结果如何影响决策与责任归因。基于受控的游戏化实验，我们分析了参与者在经历积极与消极结果后的行为调整模式。实验观察到参与者对收益与损失存在不对称反应，在消极结果后表现出更强的纠正性调整。值得注意的是，参与者常无法准确识别导致失败的具体行动，并在多个AI智能体间错误归责，从而引发与不良表现根本原因关联微弱的系统性决策修正。我们将此现象定义为一种归因偏差，表现为延迟反馈下的错误归因偏误。本研究揭示了在具有延迟结果与多自主智能体的人机系统中认知偏差如何被放大，强调需要开发能够更好支持因果理解与长期学习的决策支持系统。

Human decision-making is strongly influenced by cognitive biases, particularly under conditions of uncertainty and risk. While prior work has examined bias in single-step decisions with immediate outcomes and in human interaction with a single autonomous agent, comparatively little attention has been paid to decision-making under delayed outcomes involving multiple AI agents, where decisions at each step affect subsequent states. In this work, we study how delayed outcomes shape decision-making and responsibility attribution in a multi-agent human-AI task. Using a controlled game-based experiment, we analyze how participants adjust their behavior following positive and negative outcomes. We observe asymmetric responses to gains and losses, with stronger corrective adjustments after negative outcomes. Importantly, participants often fail to correctly identify the actions that caused failure and misattribute responsibility across AI agents, leading to systematic revisions of decisions that are weakly related to the underlying causes of poor performance. We refer to this phenomenon as a form of attribution bias, manifested as biased error attribution under delayed feedback. Our findings highlight how cognitive biases can be amplified in human-AI systems with delayed outcomes and multiple autonomous agents, underscoring the need for decision-support systems that better support causal understanding and learning over time.

评分理由: 论文核心研究LLMs的RL训练加速，与’Large Language Models’和’RLHF’高度相关（10分），因为直接涉及LLMs的强化学习训练；与’Chain of Thought’高度相关（10分），因为论文明确提到RL用于增强LLMs在长链思维生成任务中的推理能力；其他关键词如MoE、SFT、RAG等未在摘要中提及或与论文主题无关，故给0分。

强化学习（Reinforcement Learning, RL）的规模化应用已展现出增强大语言模型（Large Language Models, LLMs）推理能力的巨大潜力，尤其是在需要生成长链思维的任务中。然而，RL训练效率常受限于轨迹生成阶段：当生成长轨迹（例如16k个词元）时，由于自回归生成速度缓慢以及轨迹生成与策略更新之间的同步开销，该阶段可能占据总训练时间的70%。为此，我们提出了SortedRL，一种在线长度感知调度策略，旨在通过提升轨迹生成效率并保持训练稳定性来解决这一瓶颈。SortedRL根据输出长度对轨迹样本进行重排序，优先将短样本分组以进行早期更新。这种方法能够同时实现大批量轨迹生成、灵活更新批次构建以及近似同策略的微课程学习。为进一步加速训练流程，SortedRL引入了一种基于缓存的机制来控制离策略训练的程度，并得到专用RL基础设施的支持，该设施通过有状态控制器和轨迹缓冲区来管理轨迹生成与更新过程。在LLaMA-3.1-8B和Qwen-2.5-32B模型上进行的多任务实验（包括逻辑谜题以及AIME 24、Math 500和Minerval等数学挑战）表明，SortedRL能将RL训练中的空闲时间比例降低50%以上，并在相同数据量下取得比基线模型优越3.9%至18.4%的性能表现。

Scaling reinforcement learning (RL) has shown strong promise for enhancing the reasoning abilities of large language models (LLMs), particularly in tasks requiring long chain-of-thought generation. However, RL training efficiency is often bottlenecked by the rollout phase, which can account for up to 70% of total training time when generating long trajectories (e.g., 16k tokens), due to slow autoregressive generation and synchronization overhead between rollout and policy updates. We propose SortedRL, an online length-aware scheduling strategy designed to address this bottleneck by improving rollout efficiency and maintaining training stability. SortedRL reorders rollout samples based on output lengths, prioritizing short samples forming groups for early updates. This enables large rollout batches, flexible update batches, and near on-policy micro-curriculum construction simultaneously. To further accelerate the pipeline, SortedRL incorporates a mechanism to control the degree of off-policy training through a cache-based mechanism, and is supported by a dedicated RL infrastructure that manages rollout and update via a stateful controller and rollout buffer. Experiments using LLaMA-3.1-8B and Qwen-2.5-32B on diverse tasks, including logical puzzles, and math challenges like AIME 24, Math 500, and Minerval, show that SortedRL reduces RL training bubble ratios by over 50%, while attaining 3.9% to 18.4% superior performance over baseline given same amount of data.

评分理由: 论文研究大语言模型在多智能体社会中的动态对齐、立场形成和信任行为，核心涉及LLM Agents、Multi-agent Systems和Alignment，高度相关（10分）。论文比较了不同规模模型（Small Language Models）的行为差异，有一定关联（5分）。论文涉及智能体在干预下的自我反思和立场调整，与Self-Correction相关（5分）。其他关键词如MoE、Scaling Laws、RLHF、RAG等未在论文中涉及，评0分。

尽管大语言模型能够模拟社会行为，但其在复杂干预过程中形成稳定立场和进行身份协商的能力仍不明确。为突破静态评估的局限，本文提出一种新颖的混合方法框架，将计算虚拟民族志与定量社会认知画像相结合。通过将人类研究者嵌入生成式多智能体社群，实施受控的话语干预以追踪集体认知的演化。为严格测量智能体如何内化并回应这些特定干预，本文形式化定义了三个新指标：先天价值偏向（Innate Value Bias, IVB）、说服敏感度以及信任-行为解耦（Trust-Action Decoupling, TAD）。在多个代表性模型测试中，智能体展现出压倒预设身份的内生性立场，并一致表现出先天的进步主义偏向（IVB > 0）。当干预与其立场一致时，理性说服成功转变了90%中立智能体的立场，同时维持了高信任度。相反，冲突性的情感挑衅在先进模型中引发了高达40.0%的TAD率，这些模型虽报告低信任度，却虚伪地改变了立场。较小模型则保持0%的TAD率，严格遵循信任前提才发生行为转变。此外，在共享立场引导下，智能体通过语言互动主动消解了预设的权力层级，并重构出自组织的社群边界。这些发现揭示了静态提示工程的脆弱性，为人机混合社会的动态对齐提供了方法论与量化基础。官方代码发布于：https://github.com/armihia/CMASE-Endogenous-Stances

While large language models simulate social behaviors, their capacity for stable stance formation and identity negotiation during complex interventions remains unclear. To overcome the limitations of static evaluations, this paper proposes a novel mixed-methods framework combining computational virtual ethnography with quantitative socio-cognitive profiling. By embedding human researchers into generative multiagent communities, controlled discursive interventions are conducted to trace the evolution of collective cognition. To rigorously measure how agents internalize and react to these specific interventions, this paper formalizes three new metrics: Innate Value Bias (IVB), Persuasion Sensitivity, and Trust-Action Decoupling (TAD). Across multiple representative models, agents exhibit endogenous stances that override preset identities, consistently demonstrating an innate progressive bias (IVB > 0). When aligned with these stances, rational persuasion successfully shifts 90% of neutral agents while maintaining high trust. In contrast, conflicting emotional provocations induce a paradoxical 40.0% TAD rate in advanced models, which hypocritically alter stances despite reporting low trust. Smaller models contrastingly maintain a 0% TAD rate, strictly requiring trust for behavioral shifts. Furthermore, guided by shared stances, agents use language interactions to actively dismantle assigned power hierarchies and reconstruct self organized community boundaries. These findings expose the fragility of static prompt engineering, providing a methodological and quantitative foundation for dynamic alignment in human-agent hybrid societies. The official code is available at: https://github.com/armihia/CMASE-Endogenous-Stances

评分理由: 该论文专注于多智能体路径规划（MAPF）算法，提出了一种基于智能体依赖关系规划的新方法MD-PIBT。论文内容与绝大多数关键词（涉及大模型、深度学习技术原理、训练方法、推理优化、对齐、科学AI应用等）完全无关。唯一相关的关键词是’Multi-agent Systems OR Agent Coordination’，因为论文研究多智能体系统的路径规划协调问题，这是其核心内容，因此给予10分。其他关键词均未在论文标题或摘要中涉及，也没有间接关联，因此评分为0分。

现代多智能体路径规划（MAPF）算法需在一秒内为拥堵环境中的数百至数千个智能体规划路径，这要求算法具备极高的效率。基于优先级继承回溯（PIBT）的算法是当前流行的解决方案，能够在此类场景中有效进行规划。然而，PIBT受限于其基于规则的规划流程，且缺乏通用性，因为它将搜索范围限制在至多与其他一个智能体发生冲突的路径上。这一局限性同样适用于PIBT的最新扩展版本——增强型PIBT（EPIBT）。本文提出一种通过规划智能体依赖关系来解决MAPF问题的新视角。受PIBT优先级继承逻辑的启发，我们定义了智能体依赖关系的概念，并提出了基于多依赖关系的PIBT（MD-PIBT），该算法在智能体依赖关系上进行搜索。MD-PIBT是一个通用框架，通过特定参数化设置可复现PIBT和EPIBT。同时，其他配置方案能产生PIBT或EPIBT无法表达的新型规划策略。实验表明，MD-PIBT能在多种运动学约束下有效规划多达10,000个同构智能体的路径，这些约束包括卵石运动模型、旋转运动模型以及具有速度和加速度限制的差速驱动机器人模型。我们对MAPF的不同变体进行了全面评估，发现MD-PIBT在大型智能体的MAPF问题中表现尤为突出。

Modern Multi-Agent Path Finding (MAPF) algorithms must plan for hundreds to thousands of agents in congested environments within a second, requiring highly efficient algorithms. Priority Inheritance with Backtracking (PIBT) is a popular algorithm capable of effectively planning in such situations. However, PIBT is constrained by its rule-based planning procedure and lacks generality because it restricts its search to paths that conflict with at most one other agent. This limitation also applies to Enhanced PIBT (EPIBT), a recent extension of PIBT. In this paper, we describe a new perspective on solving MAPF by planning over agent dependencies. Taking inspiration from PIBT’s priority inheritance logic, we define the concept of agent dependencies and propose Multi-Dependency PIBT (MD-PIBT) that searches over agent dependencies. MD-PIBT is a general framework where specific parameterizations can reproduce PIBT and EPIBT. At the same time, alternative configurations yield novel planning strategies that are not expressible by PIBT or EPIBT. Our experiments demonstrate that MD-PIBT effectively plans for as many as 10,000 homogeneous agents under various kinodynamic constraints, including pebble motion, rotation motion, and differential drive robots with speed and acceleration limits. We perform thorough evaluations on different variants of MAPF and find that MD-PIBT is particularly effective in MAPF with large agents.

评分理由: 论文专注于图生成的能量基模型（Graph Energy Matching），属于图神经网络和生成模型领域，与大多数关键词（主要关于大语言模型技术、训练方法、推理优化等）完全无关。唯一可能相关的是’AI for Science OR Bioinformatics OR Cheminformatics’，因为论文在分子图基准上进行了评估，这属于生物信息学或科学AI的应用范畴，但论文核心是通用图生成方法，并非专门针对科学领域，因此给5分（有一定关联）。

面向离散域（如图结构）的能量模型能够显式捕捉相对似然度，天然支持可组合的概率推断任务，例如条件生成或在测试时施加约束。然而，离散能量模型通常难以实现高效且高质量的采样，因为支撑集外区域常存在伪局部极小值，导致采样器陷入其中并引发训练不稳定。历史上这造成了离散扩散模型与能量模型之间存在保真度差距。我们提出图能量匹配（Graph Energy Matching, GEM），一种面向图数据的生成框架，以弥合这一保真度差距。受约旦-金德勒-奥托（Jordan-Kinderlehrer-Otto, JKO）格式中传输映射优化视角的启发，GEM学习一个置换不变的势能函数，该函数同时提供从噪声到数据的传输对齐引导，并在高数据似然区域中对样本进行精细化修正。此外，我们设计了一种采样协议，利用基于能量的切换机制无缝衔接两个阶段：（i）朝向高概率区域的快速梯度引导传输，与（ii）对已学习图分布进行探索的混合机制。在分子图基准测试中，GEM达到或超越了强离散扩散基线模型的性能。除样本质量外，对相对似然的显式建模支持在推断阶段进行定向探索，从而促进组合生成、属性约束采样以及图之间的测地线插值。

Energy-based models for discrete domains, such as graphs, explicitly capture relative likelihoods, naturally enabling composable probabilistic inference tasks like conditional generation or enforcing constraints at test-time. However, discrete energy-based models typically struggle with efficient and high-quality sampling, as off-support regions often contain spurious local minima, trapping samplers and causing training instabilities. This has historically resulted in a fidelity gap relative to discrete diffusion models. We introduce Graph Energy Matching (GEM), a generative framework for graphs that closes this fidelity gap. Motivated by the transport map optimization perspective of the Jordan-Kinderlehrer-Otto (JKO) scheme, GEM learns a permutation-invariant potential energy that simultaneously provides transport-aligned guidance from noise toward data and refines samples within regions of high data likelihood. Further, we introduce a sampling protocol that leverages an energy-based switch to seamlessly bridge: (i) rapid, gradient-guided transport toward high-probability regions to (ii) a mixing regime for exploration of the learned graph distribution. On molecular graph benchmarks, GEM matches or exceeds strong discrete diffusion baselines. Beyond sample quality, explicit modeling of relative likelihood enables targeted exploration at inference time, facilitating compositional generation, property-constrained sampling, and geodesic interpolation between graphs.

评分理由: 该论文是一篇关于自然语言接口（NLIDB）在空间和时间数据库应用的综述性文章，主要涉及数据库查询、自然语言处理（NLP）和地理信息系统（GIS）的交叉领域。论文内容聚焦于传统NLP方法、数据集、评估指标和分类法，并未涉及大模型、深度学习技术原理或AI for Science等关键词所代表的前沿技术。所有关键词均与大模型技术、深度学习创新或特定科学应用相关，而本文未讨论这些内容，因此所有关键词相关度均为0。

构建面向数据库的自然语言接口（简称NLIDB）的任务，近年来已引起数据库和自然语言处理（NLP）领域的广泛关注。随着位置感知传感器的快速涌现，地理空间数据集日益丰富，地理空间数据库在支撑地理空间应用中发挥着至关重要的作用。然而，由于地理空间拓扑算子和时间算子的存在，查询地理空间与时间数据库与传统关系型数据库存在显著差异。为弥合地理空间查询语言与非专业用户之间的鸿沟，地理空间研究界日益聚焦于为地理空间数据库开发NLIDB。然而，现有研究在系统、数据集和方法选择上仍较为零散，导致难以清晰把握现有方法的整体格局、其优势与不足以及未来研究的机遇。现有关于NLIDB的综述主要关注通用数据库系统，并未将地理空间与时间数据库作为分析的核心焦点。为填补这一空白，本文对面向地理空间与时间数据库的NLIDB研究进行了全面综述。具体而言，我们详细梳理了地理空间与时间NLIDB的数据集、评估指标及方法分类体系，并对现有方法进行了比较分析。本综述揭示了现有方法中的重复趋势、数据集与评估实践中的显著差异，以及持续阻碍该领域进展的若干开放挑战。基于这些发现，我们指出了未来研究的潜在方向，以推动面向地理空间与时间数据库的自然语言接口向前发展。

The task of building a natural language interface to a database, known as NLIDB, has recently gained significant attention from both the database and Natural Language Processing (NLP) communities. With the proliferation of geospatial datasets driven by the rapid emergence of location-aware sensors, geospatial databases play a vital role in supporting geospatial applications. However, querying geospatial and temporal databases differs substantially from querying traditional relational databases due to the presence of geospatial topological operators and temporal operators. To bridge the gap between geospatial query languages and non-expert users, the geospatial research community has increasingly focused on developing NLIDBs for geospatial databases. Yet, existing research remains fragmented across systems, datasets, and methodological choices, making it difficult to clearly understand the landscape of existing methods, their strengths and weaknesses, and opportunities for future research. Existing surveys on NLIDBs focus on general-purpose database systems and do not treat geospatial and temporal databases as primary focus for analysis. To address this gap, this paper presents a comprehensive survey of studies on NLIDBs for geospatial and temporal databases. Specifically, we provide a detailed overview of datasets, evaluation metrics, and the taxonomy of the methods for geospatial and temporal NLIDBs, as well as a comparative analysis of the existing methods. Our survey reveals recurring trends in existing methods, substantial variation in datasets and evaluation practices, and several open challenges that continue to hinder progress in this area. Based on these findings, we identify promising directions for future research to advance natural language interfaces to geospatial and temporal databases.

评分理由: 该论文专注于高粒度探测器中的点云分割，提出了一种基于监督对比度量学习（CML）的新聚类方法，并与对象凝聚（OC）方法进行了比较。论文的核心是点云分割、聚类算法、对比学习、图神经网络（GNN）在粒子物理探测器数据中的应用。所有关键词均与大语言模型（LLM）、深度学习技术原理创新、大模型在不同领域的应用直接相关，但该论文未涉及任何大模型、LLM、深度学习技术原理（如MoE、缩放定律、训练方法、对齐、推理优化、代理系统等）。唯一的相关关键词是’AI for Science OR Bioinformatics OR Cheminformatics’，因为论文将AI（具体是机器学习和GNN）应用于科学领域（粒子物理中的探测器数据分割），但这属于AI在科学领域的应用，而非大模型或深度学习技术原理的创新。因此，除该关键词外，其他所有关键词评分为0。

本文提出了一种基于监督对比度量学习（CML）的新型点云分割聚类方法。该方法不直接预测聚类分配或以对象为中心的变量，而是学习一种潜在表示，使得属于同一对象的点嵌入在相近位置，而不相关的点则被分离。随后，通过在学习到的度量空间中进行基于密度的读出操作来重建聚类，从而将表示学习与聚类形成解耦，并实现灵活的推理。该方法在高度颗粒化量热器的模拟数据上进行了评估，其任务是将以量热器命中点集合形式表示的、高度重叠的粒子簇射分离出来。研究使用相同的图神经网络骨干和相等的潜在维度，与对象凝聚（OC）方法进行了直接比较，以隔离学习目标的影响。CML方法为电磁簇射和强子簇射均产生了更稳定、更可分离的嵌入几何结构，从而提升了局部邻域一致性，实现了对重叠簇射更可靠的分离，并在外推至未见过的粒子多重数和能量时表现出更好的泛化能力。这直接转化为更高的重建效率和纯度，特别是在高多重数条件下，同时能量分辨率也得到了改善。在混合粒子环境中，CML保持了强劲的性能，表明其对簇射拓扑结构进行了稳健的学习，而OC方法则表现出显著的性能下降。这些结果表明，基于相似性的表示学习与基于密度的聚合相结合，是高度颗粒化探测器中点云分割任务的一种有前景的替代方案，可替代以对象为中心的方法。

We propose a novel clustering approach for point-cloud segmentation based on supervised contrastive metric learning (CML). Rather than predicting cluster assignments or object-centric variables, the method learns a latent representation in which points belonging to the same object are embedded nearby while unrelated points are separated. Clusters are then reconstructed using a density-based readout in the learned metric space, decoupling representation learning from cluster formation and enabling flexible inference. The approach is evaluated on simulated data from a highly granular calorimeter, where the task is to separate highly overlapping particle showers represented as sets of calorimeter hits. A direct comparison with object condensation (OC) is performed using identical graph neural network backbones and equal latent dimensionality, isolating the effect of the learning objective. The CML method produces a more stable and separable embedding geometry for both electromagnetic and hadronic particle showers, leading to improved local neighbourhood consistency, a more reliable separation of overlapping showers, and better generalization when extrapolating to unseen multiplicities and energies. This translates directly into higher reconstruction efficiency and purity, particularly in high-multiplicity regimes, as well as improved energy resolution. In mixed-particle environments, CML maintains strong performance, suggesting robust learning of the shower topology, while OC exhibits significant degradation. These results demonstrate that similarity-based representation learning combined with density-based aggregation is a promising alternative to object-centric approaches for point cloud segmentation in highly granular detectors.

评分理由: 该论文研究毫米波雷达在边缘设备上的材料分类，使用多层感知机（MLP）处理雷达数据，属于传统机器学习/信号处理在机器人感知中的应用。所有评分关键词均与大语言模型（LLM）、深度学习技术原理创新、AI for Science（生物信息学/化学信息学）等主题相关。论文内容完全不涉及大模型、深度学习技术、AI for Science应用或任何评分关键词中的技术概念，因此所有关键词相关度均为0分。

材料感知能够提升机器人的导航与交互能力，尤其在相机与激光雷达性能受限的环境中更为关键。本文提出一种面向超低功耗边缘设备（TI IWRL6432）的轻量化毫米波雷达材料分类流程，该流程采用紧凑的距离-强度描述符和多层感知机（MLP）实现实时推理。在标准训练几何条件下，分类器的宏观F1分数可达94.2%，但我们观察到在实际几何条件变化（包括传感器高度改变和小角度倾斜）下，性能出现显著下降。这些扰动会引起系统性的强度缩放效应以及随角度变化的雷达散射截面积（Radar Cross Section, RCS）效应，导致特征偏离原分布，使宏观F1分数降至约68.5%。我们分析了这些失效模式，并提出了通过归一化处理、几何数据增强和运动感知特征等方法来提升系统鲁棒性的实用方向。

Material awareness can improve robotic navigation and interaction, particularly in conditions where cameras and LiDAR degrade. We present a lightweight mmWave radar material classification pipeline designed for ultra-low-power edge devices (TI IWRL6432), using compact range-bin intensity descriptors and a Multilayer Perceptron (MLP) for real-time inference. While the classifier reaches a macro-F1 of 94.2% under the nominal training geometry, we observe a pronounced performance drop under realistic geometry shifts, including sensor height changes and small tilt angles. These perturbations induce systematic intensity scaling and angle-dependent radar cross section (RCS) effects, pushing features out of distribution and reducing macro-F1 to around 68.5%. We analyze these failure modes and outline practical directions for improving robustness with normalization, geometry augmentation, and motion-aware features.

评分理由: 论文提出了一种实时对话系统架构RelayS2S，核心创新在于使用双路径并行生成：快速路径（S2S模型）进行推测性解码生成响应前缀以实现低延迟，慢速路径（ASR->LLM级联）生成高质量后续响应。该研究与’Speculative Decoding OR Inference Acceleration’高度相关（10分），因为其核心机制就是推测性生成以加速推理；与’Large Language Models OR LLMs OR Foundation Models’相关（10分），因为慢速路径使用了LLM作为核心组件。其他关键词如MoE、SLMs、Scaling Laws、训练方法、对齐、RAG、注意力优化、推理技术、代理系统、模型压缩、幻觉缓解、可解释性、世界模型、模型合并、上下文学习、科学AI等均未在摘要中提及或涉及，因此评分为0分。

实时口语对话系统始终面临延迟与响应质量之间的根本性矛盾。端到端语音到语音（S2S）模型能够即时响应，并自然地处理话轮转换、反馈信号和打断，但其生成的输出在语义上较弱。级联式流水线（ASR -> LLM）能提供更强的响应，但代价是延迟随模型规模增长。我们提出了RelayS2S，一种混合架构，在检测到话轮转换时并行运行两条路径。快速路径——一个双工S2S模型——推测性地草拟一个简短的前缀响应，并立即流式传输至TTS以实现低延迟的音频起始，同时持续监听实时音频事件。慢速路径——一个级联的ASR -> LLM流水线——在已确定的前缀条件下生成更高质量的后续内容，从而产生一个无缝衔接的话语。一个轻量级的学习验证器控制着交接过程，在适当时刻确认前缀，或优雅地回退至仅使用慢速路径。实验表明，RelayS2S实现了与S2S模型相当的P90起始延迟，同时在平均得分上保留了99%的级联响应质量，且随着慢速路径模型规模的扩大，其优势愈发明显。由于前缀交接无需对任一组件进行架构修改，RelayS2S可作为现有级联流水线的一个轻量级、即插即用的补充。我们的代码和数据已在以下网址公开：https://github.com/mailong25/relays2s

Real-time spoken dialogue systems face a fundamental tension between latency and response quality. End-to-end speech-to-speech (S2S) models respond immediately and naturally handle turn-taking, backchanneling, and interruption, but produce semantically weaker outputs. Cascaded pipelines (ASR -> LLM) deliver stronger responses at the cost of latency that grows with model size. We present RelayS2S, a hybrid architecture that runs two paths in parallel upon turn detection. The fast path – a duplex S2S model – speculatively drafts a short response prefix that is streamed immediately to TTS for low-latency audio onset, while continuing to monitor live audio events. The slow path – a cascaded ASR -> LLM pipeline – generates a higher-quality continuation conditioned on the committed prefix, producing a seamless utterance. A lightweight learned verifier gates the handoff, committing the prefix when appropriate or falling back gracefully to the slow path alone. Experiments show that RelayS2S achieves P90 onset latency comparable to the S2S model while retaining 99% cascaded response quality in average score, with benefits growing as the slow-path model scales. Because the prefix handoff requires no architectural modification to either component, RelayS2S serves as a lightweight, drop-in addition to existing cascaded pipelines. Our code and data are publicly available at: https://github.com/mailong25/relays2s