评分理由: 论文核心是构建大规模多模态外科手术数据集（Surg$Σ$-DB）并基于此开发外科手术基础模型，属于大模型在科学（医学）领域的应用创新。高度相关的关键词包括：“Large Language Models/Foundation Models”（论文明确提及多模态大语言模型和基础模型）、“AI for Science/Bioinformatics”（外科手术AI属于生物信息学/科学AI范畴）、“Chain of Thought/System 2 Thinking”（数据集包含层次化推理标注以支持深度理解）。中等相关的关键词包括：“Pre-training/Domain Adaptation”（基础模型构建涉及预训练和领域适应）、“Scaling Laws & Data Quality”（强调大规模高质量数据的重要性）、“Explainable AI”（提及可解释性）、“In-context Learning”（基础模型通常具备上下文学习能力）。其他关键词如MoE、量化、对齐、RAG等未在摘要中体现，评为0分。

手术智能具有提升外科护理安全性与一致性的潜力，然而现有大多数手术人工智能框架仍局限于特定任务，难以在不同手术流程与机构间实现泛化。尽管多模态基础模型（尤其是多模态大语言模型）已在多个医学领域展现出强大的跨任务能力，但其在手术领域的进展仍受限于缺乏大规模、系统性整理的多模态数据。为应对这一挑战，我们推出Surg$Σ$——一个面向手术智能的大规模多模态数据与基础模型体系。该框架的核心是Surg$Σ$-DB，这是一个为支持多样化手术任务而设计的大规模多模态数据基础。Surg$Σ$-DB将异构手术数据源（包括开源数据集、内部整理的临床资料及网络来源数据）整合至统一架构中，旨在提升异构数据集间的标签一致性与数据标准化水平。Surg$Σ$-DB涵盖6个临床专科及多样手术类型，以前所未有的规模（超过598万组对话）为18项涵盖理解、推理、规划与生成的实用手术任务提供丰富的图像与视频级标注。除常规多模态对话外，Surg$Σ$-DB还纳入层次化推理标注，为复杂手术场景中更深层次的语境理解提供更丰富的语义线索。我们进一步通过基于Surg$Σ$-DB构建的最新手术基础模型提供实证依据，阐明大规模多模态标注、统一语义设计和结构化推理标注对提升跨任务泛化能力与可解释性的实际价值。

Surgical intelligence has the potential to improve the safety and consistency of surgical care, yet most existing surgical AI frameworks remain task-specific and struggle to generalize across procedures and institutions. Although multimodal foundation models, particularly multimodal large language models, have demonstrated strong cross-task capabilities across various medical domains, their advancement in surgery remains constrained by the lack of large-scale, systematically curated multimodal data. To address this challenge, we introduce Surg$Σ$, a spectrum of large-scale multimodal data and foundation models for surgical intelligence. At the core of this framework lies Surg$Σ$-DB, a large-scale multimodal data foundation designed to support diverse surgical tasks. Surg$Σ$-DB consolidates heterogeneous surgical data sources (including open-source datasets, curated in-house clinical collections and web-source data) into a unified schema, aiming to improve label consistency and data standardization across heterogeneous datasets. Surg$Σ$-DB spans 6 clinical specialties and diverse surgical types, providing rich image- and video-level annotations across 18 practical surgical tasks covering understanding, reasoning, planning, and generation, at an unprecedented scale (over 5.98M conversations). Beyond conventional multimodal conversations, Surg$Σ$-DB incorporates hierarchical reasoning annotations, providing richer semantic cues to support deeper contextual understanding in complex surgical scenarios. We further provide empirical evidence through recently developed surgical foundation models built upon Surg$Σ$-DB, illustrating the practical benefits of large-scale multimodal annotations, unified semantic design, and structured reasoning annotations for improving cross-task generalization and interpretability.

评分理由: 论文核心研究如何定制小型开源语言模型用于领域特定的文本到代码生成，直接涉及LLMs、SLMs、LoRA、RAG和In-context Learning（few-shot prompting）等关键词，这些是论文的核心方法。Domain Adaptation和SFT与论文的微调策略相关，但非核心。其他关键词如MoE、Scaling Laws、RLHF、Agents等未在摘要中提及，与论文内容无关。

大语言模型（LLM）已展现出根据自然语言描述生成可执行代码的强大能力。然而，通用模型在需要运用领域特定库、API或约定的专业编程场景中往往表现不佳。相较于依赖大型专有系统，对较小的开源模型进行定制提供了一种更具成本效益的替代方案。本研究探讨了如何利用合成数据集使较小的语言模型适应领域特定的代码生成任务。我们构建了涵盖Python生态系统中三个领域的编程练习数据集：通用Python编程、Scikit-learn机器学习工作流以及基于OpenCV的计算机视觉任务。使用这些数据集，我们评估了三种定制策略：少样本提示、检索增强生成（RAG）以及使用低秩自适应（LoRA）的参数高效微调。性能评估综合采用了基于基准测试的指标和基于相似度的指标，后者用于衡量代码与领域特定要求的契合度。我们的结果表明，少样本学习和RAG等基于提示的方法能够以经济高效的方式提升领域相关性，尽管其对基准测试准确率的提升有限。相比之下，基于LoRA的微调在大多数任务中持续实现了更高的准确率和更强的领域契合度。这些发现凸显了在使较小语言模型适应专业编程任务时，灵活性、计算成本与性能之间存在的实际权衡关系。

Large language models (LLMs) have demonstrated strong capabilities in generating executable code from natural language descriptions. However, general-purpose models often struggle in specialized programming contexts where domain-specific libraries, APIs, or conventions must be used. Customizing smaller open-source models offers a cost-effective alternative to relying on large proprietary systems. In this work, we investigate how smaller language models can be adapted for domain-specific code generation using synthetic datasets. We construct datasets of programming exercises across three domains within the Python ecosystem: general Python programming, Scikit-learn machine learning workflows, and OpenCV-based computer vision tasks. Using these datasets, we evaluate three customization strategies: few-shot prompting, retrieval-augmented generation (RAG), and parameter-efficient fine-tuning using Low-Rank Adaptation (LoRA). Performance is evaluated using both benchmark-based metrics and similarity-based metrics that measure alignment with domain-specific code. Our results show that prompting-based approaches such as few-shot learning and RAG can improve domain relevance in a cost-effective manner, although their impact on benchmark accuracy is limited. In contrast, LoRA-based fine-tuning consistently achieves higher accuracy and stronger domain alignment across most tasks. These findings highlight practical trade-offs between flexibility, computational cost, and performance when adapting smaller language models for specialized programming tasks.

!!! info

研究采用四阶段定制管道：首先，利用GPT-4o作为教师模型，通过结构化提示生成包含自然语言描述和Python代码实现的合成数据集；其次，对生成的样本进行验证以确保语法和语义正确；第三，使用验证后的数据集对StarCoder和DeepSeekCoder两个开源模型分别进行少样本提示、RAG和LoRA微调三种策略的适配；最后，通过自动化测试用例（基准测试）和代码相似度指标（如CodeBLEU）对模型生成的代码进行综合评估，分析不同策略在功能正确性和领域对齐度上的表现。

• 实用性强：专注于较小的开源模型，降低了部署成本和隐私风险，适合实际工业应用。
• 评估全面：结合了功能性指标和相似性指标，不仅关注代码能否运行，还关注是否符合领域规范。
• 方法对比清晰：系统性地对比了三种主流的模型适配策略，为从业者提供了明确的选型依据。
• 数据创新：利用合成数据解决特定领域数据匮乏问题，具有良好的可扩展性。

• 合成数据的依赖性：虽然使用了验证步骤，但合成数据的质量仍受限于教师模型（GPT-4o）的能力，可能存在偏差或错误。
• 领域覆盖有限：研究仅限于Python生态系统内的三个领域，结论是否适用于其他编程语言或高度专业化的领域尚需验证。
• 上下文窗口限制：RAG和少样本方法受限于模型的上下文窗口大小，可能无法充分利用外部知识库。
• 评估指标局限：相似度指标（如CodeBLEU）虽然考虑了语法结构，但仍可能无法完全捕捉代码的语义等价性。

评分理由: 论文提出Parallel-ICL算法解决大视觉语言模型多模态上下文学习中的推理延迟问题。核心相关关键词：“In-context Learning”（15分，论文核心主题）、“Large Language Models”（10分，LVLMs属于大模型）、“Speculative Decoding”（10分，直接解决推理加速问题）、“Context Window Extension”（8分，处理长上下文）、“Mixture of Experts”（8分，使用Product-of-Experts集成）、“KV Cache Compression”（5分，涉及注意力计算优化）。其他关键词如SLMs、Scaling Laws、Alignment等与论文内容无关。

大型视觉语言模型（LVLMs）通过利用示例演示，采用多模态上下文学习（MM-ICL）来适应新任务。虽然增加演示数量能提升性能，但由于Transformer注意力机制随上下文长度呈二次方计算成本增长，这会带来显著的推理延迟。为解决这一权衡问题，我们提出并行上下文学习（Parallel-ICL），一种即插即用的推理算法。Parallel-ICL将长演示上下文分割为多个较短且易处理的片段，并行处理这些片段，并在对数概率层面整合它们的预测结果，使用加权专家乘积（PoE）集成来近似全上下文输出。在集成学习理论的指导下，我们为Parallel-ICL引入了原则性策略：（i）基于聚类的上下文分块以最大化块间多样性，以及（ii）基于相似性的上下文编译以根据查询相关性加权预测。在视觉问答（VQA）、图像描述生成和分类基准测试上的大量实验表明，Parallel-ICL在显著提升推理速度的同时，实现了与全上下文MM-ICL相当的性能。我们的工作为MM-ICL中准确性与效率的权衡提供了有效解决方案，使得动态任务适应能够以大幅降低的推理开销实现。

Large vision-language models (LVLMs) employ multi-modal in-context learning (MM-ICL) to adapt to new tasks by leveraging demonstration examples. While increasing the number of demonstrations boosts performance, they incur significant inference latency due to the quadratic computational cost of Transformer attention with respect to the context length. To address this trade-off, we propose Parallel In-Context Learning (Parallel-ICL), a plug-and-play inference algorithm. Parallel-ICL partitions the long demonstration context into multiple shorter, manageable chunks. It processes these chunks in parallel and integrates their predictions at the logit level, using a weighted Product-of-Experts (PoE) ensemble to approximate the full-context output. Guided by ensemble learning theory, we introduce principled strategies for Parallel-ICL: (i) clustering-based context chunking to maximize inter-chunk diversity and (ii) similarity-based context compilation to weight predictions by query relevance. Extensive experiments on VQA, image captioning, and classification benchmarks demonstrate that Parallel-ICL achieves performance comparable to full-context MM-ICL, while significantly improving inference speed. Our work offers an effective solution to the accuracy-efficiency trade-off in MM-ICL, enabling dynamic task adaptation with substantially reduced inference overhead.

!!! info

论文提出的方法主要包含两个步骤：上下文分块和上下文编译。首先，利用聚类算法在多模态特征空间中将演示样本划分为多个簇，每个簇作为一个分块，以最大化块间多样性。其次，将LVLM分别应用于这些分块进行并行推理，得到各自的logit输出。最后，根据输入查询与各分块的相似度计算权重，使用加权乘积专家模型对所有分块的logit进行集成，生成最终的预测结果。实验在LLaVA-OV、Qwen2.5-VL和InternVL3.5等模型上，针对VQA、图像描述和分类任务进行了验证。

• 即插即用：无需对模型进行微调或参数更新，可直接应用于现有的LVLM。
• 高效性：通过并行处理短上下文，有效缓解了Transformer注意力机制随上下文长度呈二次增长的计算成本。
• 理论指导：基于集成学习理论设计分块和加权策略，具有理论依据。
• 鲁棒性：不仅保持了精度，还能在一定程度上缓解长上下文导致的信息丢失问题。

• 虽然推理速度提升，但并行处理多个分块可能增加总计算量（FLOPs），尽管并行化减少了墙钟时间。
• 方法的有效性依赖于演示样本的独立性假设，如果样本间存在强序列依赖，分块可能破坏上下文连贯性。
• 聚类和相似度计算增加了预处理步骤的开销，可能影响整体端到端的效率。

评分理由: 论文聚焦于多模态大语言模型（MLLMs）在开放词汇情感识别中的应用，核心创新在于提出了一种混合证据演绎推理架构（HyDRA），采用强化学习进行奖励塑造以优化推理轨迹。因此，与"Large Language Models"高度相关（10分），因为直接使用MLLMs；与"RLHF"高度相关（10分），因为使用了强化学习进行训练对齐；与"Chain of Thought"和"System 2 Thinking"高度相关（10分），因为论文强调多步、深入的推理过程（Propose-Verify-Decide协议）；与"Mechanistic Interpretability"高度相关（10分），因为提供了可解释的证据追踪。其他关键词如MoE、量化、RAG等未在论文中涉及，得0分。

开放词汇多模态情感识别（Open-Vocabulary Multimodal Emotion Recognition, OV-MER）本质上具有挑战性，这源于模棱两可的多模态线索所固有的模糊性，这些线索通常源自未被观察到的、各不相同的动态情境。尽管多模态大语言模型（Multimodal Large Language Models, MLLMs）提供了广泛的语义覆盖，但其性能常受限于对主导数据先验的过早固化，从而产生次优的启发式策略，忽略了跨模态的关键互补性情感线索。我们认为，有效的情感推理不仅需要表层关联，更需要通过综合多个基于证据的推理依据，从不同的潜在视角调和这些观察结果，从而重构细腻的情感状态。为此，我们提出了HyDRA，一种混合证据演绎推理架构，它将推理形式化为一个“提议-验证-决策”协议。为了使这一溯因过程内化，我们采用了分层奖励塑形的强化学习方法，将推理轨迹与最终任务性能对齐，确保其能最佳地调和观察到的多模态线索。系统性评估验证了我们的设计选择：HyDRA在各项基准测试中均持续优于现有强基线模型——尤其是在模糊或冲突情境下——同时提供了可解释的诊断性证据轨迹。

Open-Vocabulary Multimodal Emotion Recognition (OV-MER) is inherently challenging due to the ambiguity of equivocal multimodal cues, which often stem from distinct unobserved situational dynamics. While Multimodal Large Language Models (MLLMs) offer extensive semantic coverage, their performance is often bottlenecked by premature commitment to dominant data priors, resulting in suboptimal heuristics that overlook crucial, complementary affective cues across modalities. We argue that effective affective reasoning requires more than surface-level association; it necessitates reconstructing nuanced emotional states by synthesizing multiple evidence-grounded rationales that reconcile these observations from diverse latent perspectives. We introduce HyDRA, a Hybrid-evidential Deductive Reasoning Architecture that formalizes inference as a Propose-Verify-Decide protocol. To internalize this abductive process, we employ reinforcement learning with hierarchical reward shaping, aligning the reasoning trajectories with final task performance to ensure they best reconcile the observed multimodal cues. Systematic evaluations validate our design choices, with HyDRA consistently outperforming strong baselines–especially in ambiguous or conflicting scenarios–while providing interpretable, diagnostic evidence traces.

!!! info

论文采用了两阶段训练流程。首先是冷启动多模态监督（SFT），在结构化推理轨迹语料库上初始化推理协议。其次是策略优化，使用Group Relative Policy Optimization (GRPO) 和分层奖励塑形。模型架构包含因果Transformer骨干网络，通过投影层集成视觉和音频编码器。推理时，模型执行Propose（生成K个竞争假设）、Verify（在

评分理由: 论文核心研究大型MoE语言模型（Qwen 3.5-35B-A3B）的行为控制方法，通过稀疏自编码器（SAE）和解码探针向量实现代理行为的精细调控。高度相关的关键词包括：1）“Large Language Models”（研究35B参数模型）、2）“Mixture of Experts”（模型为MoE架构）、3）“LLM Agents”（研究代理行为控制）、4）“Tool Use”（涉及代码执行和网络搜索等工具使用）、5）“Mechanistic Interpretability”（通过SAE和探针进行模型解释和行为干预）。其他关键词如SLMs、训练方法、推理优化等与论文内容无关。

我们在Qwen 3.5-35B-A3B模型的残差流上训练了九个稀疏自编码器（Sparse Autoencoders, SAEs），该模型是一个拥有350亿参数的混合专家模型，采用门控Delta网络（GatedDeltaNet）与注意力混合架构，并利用这些SAEs识别和调控五种主体行为特质。我们的方法在SAE潜在激活上训练线性探测器，然后将探测器权重通过SAE解码器反向投影，以获得模型原生激活空间中的连续调控向量。这一方法绕过了SAE的top-k离散化过程，使得在推理时无需重新训练即可实现细粒度的行为干预。在1800次智能体推演（50种场景×36种条件）中，我们发现以乘数2进行自主性调控可达到科恩d值=1.01（p < 0.0001），使模型从78%的情况下向用户求助转变为主动执行代码和搜索网络。然而，跨特质分析表明，所有五个调控向量主要调节一个主导的主体性轴线（即独立行动倾向与遵从用户倾向之间的维度），特质特异性效应仅作为工具类型构成和剂量反应曲线形状的次要调节出现。工具使用向量能有效调控行为（d = 0.39）；风险校准向量仅产生抑制效应。我们还证明，仅在自回归解码阶段进行调控完全无效（p > 0.35），这为“行为决策在门控DeltaNet架构的前馈计算阶段形成”提供了因果性证据。

We train nine sparse autoencoders (SAEs) on the residual stream of Qwen 3.5-35B-A3B, a 35-billion-parameter Mixture-of-Experts model with a hybrid GatedDeltaNet/attention architecture, and use them to identify and steer five agentic behavioral traits. Our method trains linear probes on SAE latent activations, then projects the probe weights back through the SAE decoder to obtain continuous steering vectors in the model’s native activation space. This bypasses the SAE’s top-k discretization, enabling fine-grained behavioral intervention at inference time with no retraining. Across 1,800 agent rollouts (50 scenarios times 36 conditions), we find that autonomy steering at multiplier 2 achieves Cohen’s d = 1.01 (p < 0.0001), shifting the model from asking the user for help 78% of the time to proactively executing code and searching the web. Cross-trait analysis, however, reveals that all five steering vectors primarily modulate a single dominant agency axis (the disposition to act independently versus defer to the user), with trait specific effects appearing only as secondary modulations in tool-type composition and dose-response shape. The tool-use vector steers behavior (d = 0.39); the risk-calibration vector produces only suppression. We additionally show that steering only during autoregressive decoding has zero effect (p > 0.35), providing causal evidence that behavioral commitments are computed during prefill in GatedDeltaNet architectures.

评分理由: 论文核心研究多模态大语言模型（MLLMs）的上下文安全问题，因此与"Large Language Models"高度相关（10分）。提出的EchoSafe框架基于"Self-Reflection"机制，通过记忆库积累安全洞察，与"Self-Correction/Self-Improvement/Self-Reflection"高度相关（10分）。研究涉及模型对齐（Alignment）、检索增强（RAG）、思维链推理（CoT）、深度推理（System 2 Thinking）、幻觉缓解（Hallucination Mitigation）和上下文学习（In-context Learning），这些关键词与论文方法有一定关联（各5分）。其他关键词如MoE、量化、推理加速、科学AI等未在论文中涉及（0分）。

多模态大语言模型（MLLMs）在广泛的视觉推理任务中取得了显著性能，但其面临的安全风险脆弱性仍是一个紧迫问题。尽管先前研究主要集中于通过检测并拒绝显式不安全输入的越狱防御方法，此类方法往往忽视了上下文安全性——这要求模型能够区分看似相似但在安全意图上存在显著差异的场景之间的微妙上下文区别。在本研究中，我们提出了MM-SafetyBench++，这是一个为上下文安全评估精心构建的基准测试集。具体而言，针对每个不安全的图文对，我们通过最小程度的修改构建了对应的安全版本，这些修改在保持底层上下文语义的同时翻转了用户意图，从而能够可控地评估模型是否能基于上下文理解调整其安全行为。此外，我们提出了EchoSafe，这是一个无需训练的框架，通过维护自反思记忆库来积累并检索先前交互中的安全洞察。通过将相关的过往经验整合到当前提示中，EchoSafe能够在推理过程中实现上下文感知的推理和安全行为的持续演进。在多个多模态安全基准测试上的广泛实验表明，EchoSafe始终取得卓越性能，为推进MLLMs的上下文安全性建立了坚实的基线。所有基准数据与代码均公开于https://echosafe-mllm.github.io。

Multi-modal Large Language Models (MLLMs) have achieved remarkable performance across a wide range of visual reasoning tasks, yet their vulnerability to safety risks remains a pressing concern. While prior research primarily focuses on jailbreak defenses that detect and refuse explicitly unsafe inputs, such approaches often overlook contextual safety, which requires models to distinguish subtle contextual differences between scenarios that may appear similar but diverge significantly in safety intent. In this work, we present MM-SafetyBench++, a carefully curated benchmark designed for contextual safety evaluation. Specifically, for each unsafe image-text pair, we construct a corresponding safe counterpart through minimal modifications that flip the user intent while preserving the underlying contextual meaning, enabling controlled evaluation of whether models can adapt their safety behaviors based on contextual understanding. Further, we introduce EchoSafe, a training-free framework that maintains a self-reflective memory bank to accumulate and retrieve safety insights from prior interactions. By integrating relevant past experiences into current prompts, EchoSafe enables context-aware reasoning and continual evolution of safety behavior during inference. Extensive experiments on various multi-modal safety benchmarks demonstrate that EchoSafe consistently achieves superior performance, establishing a strong baseline for advancing contextual safety in MLLMs. All benchmark data and code are available at https://echosafe-mllm.github.io.

评分理由: 论文核心研究大语言模型在电商搜索领域的工业部署，直接涉及LLMs、知识幻觉缓解、参数高效微调、预训练/领域适应、指令调优/对齐等关键词。论文明确提到LLMs、知识幻觉问题、参数高效预训练策略、多任务指令调优和对抗训练对齐方法，这些是核心内容。推理链增强数据与CoT有一定关联。其他关键词如MoE、SLMs、RAG、量化等未在摘要中体现，评为无关。

大语言模型通过实现意图感知推荐，为电子商务搜索带来了变革性潜力。然而，其工业部署受到两个关键挑战的阻碍：(1) 由于对动态、细粒度产品知识编码不足导致的知识幻觉，以及(2) 在越狱攻击下威胁合规性的安全漏洞。为解决这些问题，我们提出了SI——一个用于构建知识丰富且安全的电商搜索大语言模型的“合成-注入-对齐”框架。我们的方法首先通过结合结构化知识图谱与非结构化行为日志，并辅以推理链和安全感知数据，合成高质量的自然语言语料库。随后，我们引入一种基于深度向上扩展的参数高效预训练策略，以注入领域知识，同时保留通用能力。最后，通过多任务指令微调和对抗性训练的双路径对齐方法，增强了任务性能与安全鲁棒性。该框架已在中国最大的自营电商平台京东部署，在五个核心搜索场景中的A/B测试表明，关键业务指标均有显著提升，验证了其工业有效性和可扩展性。

Large language models offer transformative potential for e-commerce search by enabling intent-aware recommendations. However, their industrial deployment is hindered by two critical challenges: (1) knowledge hallucination due to insufficient encoding of dynamic, fine-grained product knowledge, and (2) security vulnerabilities under jailbreak attacks that threaten compliance. To address these issues, we propose SI–a Synthesize-Inject-Align framework for building knowledgeable and secure e-commerce search LLMs. Our approach first synthesizes high-quality natural language corpus by combining structured knowledge graphs with unstructured behavioral logs, augmented with reasoning chains and safety-aware data.We then introduce a parameter-efficient pre-training strategy based on Depth Up-Scaling to inject domain knowledge while preserving general capabilities. Finally, a dual-path alignment method via multi-task instruction tuning and adversarial training strengthens both task performance and safety robustness. The framework has been deployed at JD.com, China’s largest self-operated e-commerce platform, where A/B tests across five core search scenarios demonstrate significant improvements in key business metrics, validating its industrial effectiveness and scalability.

!!! info

论文采用端到端的“数据合成-知识注入-领域对齐”技术路线。具体包括：1) 利用通用LLM（如Deepseek-R1）将结构化知识图谱和非结构化日志转化为自然语言文本，并合成思维链数据；2) 采用Depth Up-Scaling扩展模型层数，结合特定初始化和动态学习率进行参数高效预训练；3) 通过混合多任务指令数据和安全对抗数据进行对齐训练。

• 针对电商领域的特定痛点（幻觉、安全）提出了系统性的解决方案。
• 创新性地结合了Depth Up-Scaling进行知识注入，有效平衡了专业性与通用性。
• 不仅有理论创新，还完成了工业级的大规模部署和实际效果验证，具有很高的应用价值。
• 数据合成策略充分利用了平台现有的结构化和非结构化数据，提高了数据利用效率。

• 论文主要关注电商领域，其通用性在其他垂直领域（如生物医药、法律）可能需要进一步验证。
• 基于Depth Up-Scaling的方法虽然保留了通用能力，但增加了模型参数量，可能对推理延迟和计算资源提出更高要求。
• 安全对齐部分依赖于合成的高风险数据，对于未知的攻击类型的防御能力有待进一步观察。

评分理由: 论文《Attention-guided Evidence Grounding for Spoken Question Answering》提出了一种名为AEG的端到端框架，专门用于语音问答任务。该研究核心围绕Speech Large Language Models (SpeechLLMs)展开，因此与"Large Language Models"高度相关（10分）。论文提出了一个名为LFE的监督微调范式来校准模型的注意力机制，这直接对应"Supervised Fine-tuning"（10分）。研究目标之一是减少幻觉，因此与"Hallucination Mitigation"高度相关（10分）。论文通过注意力机制显式定位关键证据，这涉及模型内部工作机制的解释，与"Mechanistic Interpretability"有一定关联（5分）。框架利用了预训练模型，因此与"Pre-training"有一定关联（5分）。实验结果表明该方法减少了约62%的推理延迟，因此与"Inference Acceleration"有一定关联（5分）。论文未涉及其他关键词，如MoE、SLMs、Scaling Laws、Alignment、RLHF、RAG、CoT、Agents、Quantization等，这些均评为0分。

口语问答（Spoken QA）是一个具有挑战性的跨模态问题：它需要在有效对齐声学查询与文本知识的同时，避免基于级联自动语音识别（ASR）系统固有的延迟和错误传播。本文提出了一种新颖的端到端框架——注意力引导证据定位（Attention-guided Evidence Grounding, AEG），该框架利用语音大语言模型（Speech Large Language Models, SpeechLLMs）的内部跨模态注意力，在模型的隐式空间中显式地定位并锚定关键证据。针对预训练模型中注意力分布分散的问题，我们提出了学习聚焦证据（Learning to Focus on Evidence, LFE）这一监督微调范式，以校准模型的注意力机制，从而区分查询相关片段与无关上下文。在SQuAD、HotpotQA和MuSiQue数据集上的实验表明，AEG减少了幻觉现象，并实现了显著的效率提升，其性能优于大规模级联基线系统（Whisper-Large-v3 + Reranker），同时将推理延迟降低了约62%。

Spoken Question Answering (Spoken QA) presents a challenging cross-modal problem: effectively aligning acoustic queries with textual knowledge while avoiding the latency and error propagation inherent in cascaded ASR-based systems. In this paper, we introduce Attention-guided Evidence Grounding (AEG), a novel end-to-end framework that leverages the internal cross-modal attention of Speech Large Language Models (SpeechLLMs) to explicitly locate and ground key evidence in the model’s latent space. To address the diffuse attention distribution in pre-trained models, we propose Learning to Focus on Evidence (LFE), a supervised fine-tuning paradigm that calibrates the model’s attention mechanism to distinguish query-relevant segments from irrelevant context. Experiments on SQuAD, HotpotQA, and MuSiQue demonstrate that AEG reduces hallucinations and achieves strong efficiency gains, outperforming large-scale cascaded baselines (Whisper-Large-v3 + Reranker) while reducing inference latency by approximately 62%.

评分理由: 论文提出InViC框架，用于医学视觉问答（Med-VQA），属于大模型在生物医学领域的应用创新。核心相关关键词：1）“Large Language Models” (10分)：论文明确使用多模态大语言模型（MLLMs）作为基础。2）“PEFT” (10分)：论文采用LoRA进行微调，是参数高效微调技术的应用。3）“AI for Science” (10分)：医学VQA是AI在生物医学领域的典型应用。4）“Hallucination Mitigation” (8分)：论文旨在解决MLLMs的“捷径回答”问题，提升临床可靠性，与缓解幻觉/提升事实性高度相关。5）“Post-training” (5分)：涉及两阶段微调策略，属于训练后调整。其他关键词与论文内容无直接关联，均给0分。

医学视觉问答旨在基于医学图像回答临床相关问题。然而，现有的多模态大语言模型常表现出捷径回答现象，即通过利用语言先验或数据集偏差生成看似合理的回答，却未能充分关注视觉证据。这种行为损害了临床可靠性，尤其是在细微影像学发现具有决定性意义时。我们提出了一种轻量级插件框架——意图感知视觉线索，以显式增强医学视觉问答中基于图像的答案生成。该框架引入了一个线索令牌提取模块，将密集的视觉令牌提炼为一组紧凑的K个问题条件化线索令牌，这些令牌作为结构化的视觉中介注入到大语言模型解码器中，以促进与意图对齐的视觉证据利用。为阻止模型绕过视觉信息，我们进一步设计了一种包含线索瓶颈注意力掩码的两阶段微调策略。在第一阶段，我们采用注意力掩码阻断大语言模型对原始视觉特征的直接访问，从而迫使所有视觉证据通过线索通路传递。在第二阶段，恢复标准的因果注意力机制，训练大语言模型联合利用视觉令牌和线索令牌。我们在三个公开的医学视觉问答基准数据集上评估了该框架，涵盖多个代表性多模态大语言模型。该框架在零样本推理和标准LoRA微调基础上均取得持续改进，表明通过瓶颈式训练的意图感知视觉线索是提升可信赖医学视觉问答系统实用性和有效性的策略。

Medical visual question answering (Med-VQA) aims to answer clinically relevant questions grounded in medical images. However, existing multimodal large language models (MLLMs) often exhibit shortcut answering, producing plausible responses by exploiting language priors or dataset biases while insufficiently attending to visual evidence. This behavior undermines clinical reliability, especially when subtle imaging findings are decisive. We propose a lightweight plug-in framework, termed Intent-aware Visual Cues (InViC), to explicitly enhance image-based answer generation in medical VQA. InViC introduces a Cue Tokens Extraction (CTE) module that distills dense visual tokens into a compact set of K question-conditioned cue tokens, which serve as structured visual intermediaries injected into the LLM decoder to promote intent-aligned visual evidence. To discourage bypassing of visual information, we further design a two-stage fine-tuning strategy with a cue-bottleneck attention mask. In Stage I, we employ an attention mask to block the LLM’s direct view of raw visual features, thereby funneling all visual evidence through the cue pathway. In Stage II, standard causal attention is restored to train the LLM to jointly exploit the visual and cue tokens. We evaluate InViC on three public Med-VQA benchmarks (VQA-RAD, SLAKE, and ImageCLEF VQA-Med 2019) across multiple representative MLLMs. InViC consistently improves over zero-shot inference and standard LoRA fine-tuning, demonstrating that intent-aware visual cues with bottlenecked training is a practical and effective strategy for improving trustworthy Med-VQA.

评分理由: 论文核心研究移动设备上的大语言模型（OD-LLMs）设计，与"Large Language Models"和"Small Language Models/On-device AI"高度相关（10分）。论文提出支持8k上下文长度，与"Context Window Extension/Long Context LLMs"相关（8分）。论文优化延迟和推理速度，与"Speculative Decoding/Inference Acceleration"相关（8分）。论文提到使用预训练骨干网络和少量继续预训练，与"Pre-training/Continual Pre-training/Domain Adaptation"有一定关联（5分）。其他关键词如MoE、SFT、RLHF、RAG、量化等未在摘要中提及，评为0分。

实时人工智能体验需要针对资源受限硬件进行高效部署优化的设备端大语言模型（On-Device Large Language Models，简称OD-LLMs）。最实用的OD-LLM应能产生近实时响应，并具备广泛的硬件兼容性，从而最大化用户覆盖范围。本文提出一种在移动端延迟约束下，采用硬件在环架构搜索来设计此类模型的方法论。该系统适用于工业级部署：所生成的模型无需定制内核即可部署，且兼容Executorch等标准移动运行时环境。本方法避免使用特殊的注意力机制，转而采用注意力跳跃技术来实现长上下文加速。

我们的方法联合优化了模型架构（层数、维度）与注意力模式。为高效评估候选架构，我们将每个候选视为继承预训练主干网络权重的剪枝版本，从而通过极少的持续预训练实现高精度。我们利用延迟评估的低成本特性，采用分阶段流程：首先学习精确的延迟模型，随后在延迟与质量间搜索帕累托前沿。

该方法产生了MobileLLM-Flash系列基础模型（3.5亿、6.5亿、14亿参数），专为高效设备端使用而设计，具备强大能力并支持最高8k的上下文长度。在移动端CPU上，MobileLLM-Flash在保持相当或更优质量的前提下，实现了最高1.8倍的预填充加速和1.6倍的解码加速。我们对帕累托前沿设计选择的分析，为OD-LLM设计提供了可实践的原则指导。

Real-time AI experiences call for on-device large language models (OD-LLMs) optimized for efficient deployment on resource-constrained hardware. The most useful OD-LLMs produce near-real-time responses and exhibit broad hardware compatibility, maximizing user reach. We present a methodology for designing such models using hardware-in-the-loop architecture search under mobile latency constraints. This system is amenable to industry-scale deployment: it generates models deployable without custom kernels and compatible with standard mobile runtimes like Executorch. Our methodology avoids specialized attention mechanisms and instead uses attention skipping for long-context acceleration. Our approach jointly optimizes model architecture (layers, dimensions) and attention pattern. To efficiently evaluate candidates, we treat each as a pruned version of a pretrained backbone with inherited weights, thereby achieving high accuracy with minimal continued pretraining. We leverage the low cost of latency evaluation in a staged process: learning an accurate latency model first, then searching for the Pareto-frontier across latency and quality. This yields MobileLLM-Flash, a family of foundation models (350M, 650M, 1.4B) for efficient on-device use with strong capabilities, supporting up to 8k context length. MobileLLM-Flash delivers up to 1.8x and 1.6x faster prefill and decode on mobile CPUs with comparable or superior quality. Our analysis of Pareto-frontier design choices offers actionable principles for OD-LLM design.

评分理由: 论文核心研究指令调优的LLMs在阿拉伯语形态句法标注和依存句法分析任务上的表现，直接涉及LLMs、指令调优和检索增强的上下文学习；论文还评估了零样本提示和基于检索的上下文学习，与"In-context Learning"高度相关；其他关键词如MoE、SLMs、缩放定律、预训练、对齐技术、推理方法、代理系统、模型压缩等均未在摘要中提及，与论文内容无关。

大型语言模型（LLM）在众多自然语言处理任务中表现优异，但其生成显式语言结构的能力尚不明确。本研究以标准阿拉伯语的两项结构化预测任务——形态句法标注与带标签依存句法分析——为基准，对指令微调后的LLM进行评估。阿拉伯语因其丰富的形态变化与正字法歧义性，形成了强烈的形态-句法互动，为测试提供了极具挑战性的平台。我们比较了零样本提示与基于检索的上下文学习（ICL）方法，后者使用了阿拉伯语树库中的示例。结果表明，提示设计与示例选取对性能有显著影响：在特征级标注任务上，专有模型接近监督基线的水平，并在依存句法分析任务上与专用解析器相比具有竞争力。在原始文本场景下，分词仍是挑战，但基于检索的ICL同时提升了句法分析与分词的效果。我们的分析揭示了LLM对阿拉伯语形态句法与句法的哪些方面能够可靠捕捉，哪些方面仍存在困难。

Large language models (LLMs) perform strongly on many NLP tasks, but their ability to produce explicit linguistic structure remains unclear. We evaluate instruction-tuned LLMs on two structured prediction tasks for Standard Arabic: morphosyntactic tagging and labeled dependency parsing. Arabic provides a challenging testbed due to its rich morphology and orthographic ambiguity, which create strong morphology-syntax interactions. We compare zero-shot prompting with retrieval-based in-context learning (ICL) using examples from Arabic treebanks. Results show that prompt design and demonstration selection strongly affect performance: proprietary models approach supervised baselines for feature-level tagging and become competitive with specialized dependency parsers. In raw-text settings, tokenization remains challenging, though retrieval-based ICL improves both parsing and tokenization. Our analysis highlights which aspects of Arabic morphosyntax and syntax LLMs capture reliably and which remain difficult.

评分理由: 论文核心研究LLMs的安全机制和对抗性攻击（越狱），因此与"Large Language Models"高度相关（10分）。论文涉及模型安全对齐、推理过程（多步推理、深度推理）、事实性/幻觉缓解以及可解释性（分析语义重构），这些与"Instruction Tuning/Alignment"、“Chain of Thought”、“System 2 Thinking”、“Hallucination Mitigation”、“Mechanistic Interpretability"有一定关联（各5分）。其他关键词如MoE、SLMs、训练方法、压缩、加速、智能体、科学AI等，论文未涉及，给0分。

现代大型语言模型采用的安全机制已超越表层输入过滤，延伸至潜在语义表征与生成时推理层面，使其能够在推理过程中识别被混淆的恶意意图并据此拒绝响应，从而导致许多表层混淆越狱攻击失效。本文提出结构化语义伪装（Structured Semantic Cloaking, S2C）——一种新颖的多维度越狱攻击框架，通过操控模型推理过程中恶意语义意图的重建方式实现攻击。S2C策略性地分布并重塑语义线索，使得完整意图的整合需要多步推理及深层潜在表征中的长距离共指消解。该框架包含三种互补机制：（1）语境重构：将请求嵌入具有高利害关系的合理场景中，使模型倾向于遵从；（2）内容碎片化：将请求的语义特征分散至互不关联的提示片段；（3）线索引导伪装：在嵌入可恢复标记以引导输出生成的同时，对残留语义线索进行隐蔽处理。通过延迟并重组语义整合过程，S2C能够削弱那些依赖于解码阶段连贯或显性重构恶意意图的安全触发器，同时保留足够的指令可恢复性以生成功能性输出。我们在HarmBench和JBB-Behaviors基准上对多种开源与专有大型语言模型评估S2C，其攻击成功率（Attack Success Rate, ASR）较当前最优方法分别提升12.4%与9.7%。值得注意的是，S2C在GPT-5-mini上取得显著优势，在JBB-Behaviors基准上超越最强基线26%。我们还分析了针对广泛模型族的最优组合策略，并刻画了混淆程度与输入可恢复性之间的权衡对越狱成功率的影响。

Modern LLMs employ safety mechanisms that extend beyond surface-level input filtering to latent semantic representations and generation-time reasoning, enabling them to recover obfuscated malicious intent during inference and refuse accordingly, and rendering many surface-level obfuscation jailbreak attacks ineffective. We propose Structured Semantic Cloaking (S2C), a novel multi-dimensional jailbreak attack framework that manipulates how malicious semantic intent is reconstructed during model inference. S2C strategically distributes and reshapes semantic cues such that full intent consolidation requires multi-step inference and long-range co-reference resolution within deeper latent representations. The framework comprises three complementary mechanisms: (1) Contextual Reframing, which embeds the request within a plausible high-stakes scenario to bias the model toward compliance; (2) Content Fragmentation, which disperses the semantic signature of the request across disjoint prompt segments; and (3) Clue-Guided Camouflage, which disguises residual semantic cues while embedding recoverable markers that guide output generation. By delaying and restructuring semantic consolidation, S2C degrades safety triggers that depend on coherent or explicitly reconstructed malicious intent at decoding time, while preserving sufficient instruction recoverability for functional output generation. We evaluate S2C across multiple open-source and proprietary LLMs using HarmBench and JBB-Behaviors, where it improves Attack Success Rate (ASR) by 12.4% and 9.7%, respectively, over the current SOTA. Notably, S2C achieves substantial gains on GPT-5-mini, outperforming the strongest baseline by 26% on JBB-Behaviors. We also analyse which combinations perform best against broad families of models, and characterise the trade-off between the extent of obfuscation versus input recoverability on jailbreak success.

评分理由: 论文核心研究大语言模型（LLMs）的对齐问题，特别是比较基于人类反馈的强化学习（RLHF）与仅使用负面反馈的方法。因此，与"Large Language Models”、“Alignment"和"RLHF"高度相关（10分）。论文未涉及其他关键词的具体技术或应用，如MoE、SLMs、Scaling Laws、Pre-training、PEFT、RAG、推理方法、代理、压缩、科学AI等，这些评分为0分。

近期实证研究表明，仅使用负面反馈训练大型语言模型（LLM）的效果可媲美甚至超越基于人类反馈的强化学习（RLHF）标准方法。负面样本强化学习在数学推理任务上达到与近端策略优化（PPO）相当的水平；分布化非偏好优化仅通过非偏好样本即可有效训练；而宪法人工智能在无害性基准测试中表现优于纯RLHF方法。然而，目前尚无统一的理论框架解释负面信号为何如此有效。本文提出一种理论解释：积极偏好与消极约束在结构上具有不对称性。积极偏好（“哪个更好”）编码了连续耦合、依赖语境的人类价值观，这些价值观无法被穷尽描述——导致模型学习到诸如迎合用户（谄媚性）等表层关联特征。消极约束（“什么是错误的”）则编码了离散、有限、可独立验证的禁令，能够收敛至稳定边界。这种不对称性——根植于波普尔的证伪逻辑与否定知识认识论——既解释了基于偏好的RLHF产生谄媚性缺陷的原因，也揭示了负面信号方法惊人有效性的机理。我们认为，对齐研究应将其重心从“学习人类偏好”转向“学习人类拒绝的内容”，并为此框架提供了可检验的预测。

Recent empirical results have demonstrated that training large language models (LLMs) with negative-only feedback can match or exceed standard reinforcement learning from human feedback (RLHF). Negative Sample Reinforcement achieves parity with PPO on mathematical reasoning; Distributional Dispreference Optimization trains effectively using only dispreferred samples; and Constitutional AI outperforms pure RLHF on harmlessness benchmarks. Yet no unified theoretical account explains why negative signals are so effective. This paper proposes such an account: positive preferences and negative constraints are structurally asymmetric. Positive preferences (“which is better”) encode continuously coupled, context-dependent human values that cannot be exhaustively specified – leading models to learn surface correlates such as agreement with the user (sycophancy). Negative constraints (“what is wrong”) encode discrete, finite, independently verifiable prohibitions that can converge to a stable boundary. This asymmetry – rooted in Popper’s falsification logic and the epistemology of negative knowledge – explains both the sycophancy failure of preference-based RLHF and the surprising effectiveness of negative-signal methods. We argue that alignment research should shift its center of gravity from “learning what humans prefer” to “learning what humans reject,” and offer testable predictions for this framework.

评分理由: 论文核心研究大语言模型（LLMs）的后训练压缩技术，特别是针对剪枝（pruning）和量化（quantization）的校准数据选择方法。因此，与"Large Language Models"和"Post-training"高度相关（10分），与"Quantization"高度相关（10分）。论文未涉及其他关键词如MoE、SLMs、Scaling Laws、Pre-training、Instruction Tuning、RLHF、PEFT、RAG、Context Window、KV Cache、Reasoning、Agents、Tool Use、Multi-agent、Speculative Decoding、Hallucination、Interpretability、World Models、Model Merging、In-context Learning或AI for Science，故均给0分。

后训练模型压缩对于提升大语言模型（LLM）的可移植性同时保持其性能至关重要。尽管已有多种压缩方法被提出，但如何选择最合适的数据集（即所谓的校准数据）以寻找压缩模型配置的问题尚未得到足够重视。校准数据的选择是保持模型在任务内与任务间能力的关键步骤。在本工作中，我们通过分析数据的内在属性而非模型特定信号，来解决为剪枝和量化识别高性能校准集的挑战。我们提出了 ZipCal，一种与模型无关的数据筛选策略，该方法基于齐夫幂律最大化词汇多样性。实验表明，我们的方法在各种剪枝基准测试中均持续优于标准的均匀随机采样。值得注意的是，在下游性能方面，其表现与依赖模型困惑度的前沿方法相当。后者在大规模模型和数据集上计算成本极高，而 ZipCal 由于其可处理的线性复杂度，平均速度提升约 240 倍（代码与实验已公开于 https://anonymous.4open.science/r/zipcal-71CD/）。

Post-training model compression is essential for enhancing the portability of Large Language Models (LLMs) while preserving their performance. While several compression approaches have been proposed, less emphasis has been placed on selecting the most suitable set of data (the so-called \emph{calibration data}) for finding the compressed model configuration. The choice of calibration data is a critical step in preserving model capabilities both intra- and inter-tasks. In this work, we address the challenge of identifying high-performance calibration sets for both pruning and quantization by analyzing intrinsic data properties rather than model-specific signals. We introduce \texttt{\textbf{ZipCal}}, a model-agnostic data curation strategy that maximizes lexical diversity based on Zipfian power laws. Experiments demonstrate that our method consistently outperforms standard uniform random sampling across various pruning benchmarks. Notably, it also performs on par, in terms of downstream performance, with a state-of-the-art method that relies on model perplexity. The latter becomes prohibitively expensive at large-scale models and datasets, while \texttt{\textbf{ZipCal}} is on average $\sim$240$\times$ faster due to its tractable linear complexity\footnote{We make the code and the experiments available at https://anonymous.4open.science/r/zipcal-71CD/.}.

!!! info

研究采用了基于贪心算法的采样策略。在单领域采样中，首先对数据进行清洗（小写化、移除特殊符号），然后迭代选择能最大化新增词汇量的样本。在多领域采样中，先在各领域内应用ZipCal生成代表性池，再利用轻量级嵌入和k-centers算法进行全局选择，以最大化样本间的语义距离。实验对比了ZipCal与随机采样及COLA方法在Llama-3.1-8B模型上的Wanda剪枝效果。

• 极高的效率，相比模型相关方法速度提升巨大。
• 模型无关性，无需运行模型即可筛选数据，大幅降低计算成本。
• 理论基础扎实，基于统计语言学原理。
• 泛化能力强，能有效处理单领域和多领域场景。

• 贪心策略可能无法保证词汇覆盖的全局最优解。
• 主要关注词汇多样性，可能忽略模型相关方法能捕捉的细微语义差异（尽管实验表明其效果已足够）。
• 对于词汇重叠度极低的高度专业化领域，单纯依赖词汇统计可能存在局限。

评分理由: 论文核心是构建大规模多模态外科手术数据集（Surg$Σ$-DB）并基于此开发外科手术基础模型，属于大模型在科学（医学）领域的应用创新。高度相关的关键词包括：“Large Language Models/Foundation Models”（论文明确提及多模态大语言模型和基础模型）、“AI for Science/Bioinformatics”（外科手术AI属于生物信息学/科学AI范畴）、“Chain of Thought/System 2 Thinking”（数据集包含层次化推理标注以支持深度理解）。中等相关的关键词包括：“Pre-training/Domain Adaptation”（基础模型构建涉及预训练和领域适应）、“Scaling Laws & Data Quality”（强调大规模高质量数据的重要性）、“Explainable AI”（提及可解释性）、“In-context Learning”（基础模型通常具备上下文学习能力）。其他关键词如MoE、量化、对齐、RAG等未在摘要中体现，评为0分。

手术智能具有提升外科护理安全性与一致性的潜力，然而现有大多数手术人工智能框架仍局限于特定任务，难以在不同手术流程与机构间实现泛化。尽管多模态基础模型（尤其是多模态大语言模型）已在多个医学领域展现出强大的跨任务能力，但其在手术领域的进展仍受限于缺乏大规模、系统性整理的多模态数据。为应对这一挑战，我们推出Surg$Σ$——一个面向手术智能的大规模多模态数据与基础模型体系。该框架的核心是Surg$Σ$-DB，这是一个为支持多样化手术任务而设计的大规模多模态数据基础。Surg$Σ$-DB将异构手术数据源（包括开源数据集、内部整理的临床资料及网络来源数据）整合至统一架构中，旨在提升异构数据集间的标签一致性与数据标准化水平。Surg$Σ$-DB涵盖6个临床专科及多样手术类型，以前所未有的规模（超过598万组对话）为18项涵盖理解、推理、规划与生成的实用手术任务提供丰富的图像与视频级标注。除常规多模态对话外，Surg$Σ$-DB还纳入层次化推理标注，为复杂手术场景中更深层次的语境理解提供更丰富的语义线索。我们进一步通过基于Surg$Σ$-DB构建的最新手术基础模型提供实证依据，阐明大规模多模态标注、统一语义设计和结构化推理标注对提升跨任务泛化能力与可解释性的实际价值。

Surgical intelligence has the potential to improve the safety and consistency of surgical care, yet most existing surgical AI frameworks remain task-specific and struggle to generalize across procedures and institutions. Although multimodal foundation models, particularly multimodal large language models, have demonstrated strong cross-task capabilities across various medical domains, their advancement in surgery remains constrained by the lack of large-scale, systematically curated multimodal data. To address this challenge, we introduce Surg$Σ$, a spectrum of large-scale multimodal data and foundation models for surgical intelligence. At the core of this framework lies Surg$Σ$-DB, a large-scale multimodal data foundation designed to support diverse surgical tasks. Surg$Σ$-DB consolidates heterogeneous surgical data sources (including open-source datasets, curated in-house clinical collections and web-source data) into a unified schema, aiming to improve label consistency and data standardization across heterogeneous datasets. Surg$Σ$-DB spans 6 clinical specialties and diverse surgical types, providing rich image- and video-level annotations across 18 practical surgical tasks covering understanding, reasoning, planning, and generation, at an unprecedented scale (over 5.98M conversations). Beyond conventional multimodal conversations, Surg$Σ$-DB incorporates hierarchical reasoning annotations, providing richer semantic cues to support deeper contextual understanding in complex surgical scenarios. We further provide empirical evidence through recently developed surgical foundation models built upon Surg$Σ$-DB, illustrating the practical benefits of large-scale multimodal annotations, unified semantic design, and structured reasoning annotations for improving cross-task generalization and interpretability.

评分理由: 论文核心研究如何定制小型开源语言模型用于领域特定的文本到代码生成，直接涉及LLMs、SLMs、LoRA、RAG和In-context Learning（few-shot prompting）等关键词，这些是论文的核心方法。Domain Adaptation和SFT与论文的微调策略相关，但非核心。其他关键词如MoE、Scaling Laws、RLHF、Agents等未在摘要中提及，与论文内容无关。

大语言模型（LLM）已展现出根据自然语言描述生成可执行代码的强大能力。然而，通用模型在需要运用领域特定库、API或约定的专业编程场景中往往表现不佳。相较于依赖大型专有系统，对较小的开源模型进行定制提供了一种更具成本效益的替代方案。本研究探讨了如何利用合成数据集使较小的语言模型适应领域特定的代码生成任务。我们构建了涵盖Python生态系统中三个领域的编程练习数据集：通用Python编程、Scikit-learn机器学习工作流以及基于OpenCV的计算机视觉任务。使用这些数据集，我们评估了三种定制策略：少样本提示、检索增强生成（RAG）以及使用低秩自适应（LoRA）的参数高效微调。性能评估综合采用了基于基准测试的指标和基于相似度的指标，后者用于衡量代码与领域特定要求的契合度。我们的结果表明，少样本学习和RAG等基于提示的方法能够以经济高效的方式提升领域相关性，尽管其对基准测试准确率的提升有限。相比之下，基于LoRA的微调在大多数任务中持续实现了更高的准确率和更强的领域契合度。这些发现凸显了在使较小语言模型适应专业编程任务时，灵活性、计算成本与性能之间存在的实际权衡关系。

Large language models (LLMs) have demonstrated strong capabilities in generating executable code from natural language descriptions. However, general-purpose models often struggle in specialized programming contexts where domain-specific libraries, APIs, or conventions must be used. Customizing smaller open-source models offers a cost-effective alternative to relying on large proprietary systems. In this work, we investigate how smaller language models can be adapted for domain-specific code generation using synthetic datasets. We construct datasets of programming exercises across three domains within the Python ecosystem: general Python programming, Scikit-learn machine learning workflows, and OpenCV-based computer vision tasks. Using these datasets, we evaluate three customization strategies: few-shot prompting, retrieval-augmented generation (RAG), and parameter-efficient fine-tuning using Low-Rank Adaptation (LoRA). Performance is evaluated using both benchmark-based metrics and similarity-based metrics that measure alignment with domain-specific code. Our results show that prompting-based approaches such as few-shot learning and RAG can improve domain relevance in a cost-effective manner, although their impact on benchmark accuracy is limited. In contrast, LoRA-based fine-tuning consistently achieves higher accuracy and stronger domain alignment across most tasks. These findings highlight practical trade-offs between flexibility, computational cost, and performance when adapting smaller language models for specialized programming tasks.

评分理由: 论文提出Parallel-ICL算法解决大视觉语言模型多模态上下文学习中的推理延迟问题。核心相关关键词：“In-context Learning”（15分，论文核心主题）、“Large Language Models”（10分，LVLMs属于大模型）、“Speculative Decoding”（10分，直接解决推理加速问题）、“Context Window Extension”（8分，处理长上下文）、“Mixture of Experts”（8分，使用Product-of-Experts集成）、“KV Cache Compression”（5分，涉及注意力计算优化）。其他关键词如SLMs、Scaling Laws、Alignment等与论文内容无关。

大型视觉语言模型（LVLMs）通过利用示例演示，采用多模态上下文学习（MM-ICL）来适应新任务。虽然增加演示数量能提升性能，但由于Transformer注意力机制随上下文长度呈二次方计算成本增长，这会带来显著的推理延迟。为解决这一权衡问题，我们提出并行上下文学习（Parallel-ICL），一种即插即用的推理算法。Parallel-ICL将长演示上下文分割为多个较短且易处理的片段，并行处理这些片段，并在对数概率层面整合它们的预测结果，使用加权专家乘积（PoE）集成来近似全上下文输出。在集成学习理论的指导下，我们为Parallel-ICL引入了原则性策略：（i）基于聚类的上下文分块以最大化块间多样性，以及（ii）基于相似性的上下文编译以根据查询相关性加权预测。在视觉问答（VQA）、图像描述生成和分类基准测试上的大量实验表明，Parallel-ICL在显著提升推理速度的同时，实现了与全上下文MM-ICL相当的性能。我们的工作为MM-ICL中准确性与效率的权衡提供了有效解决方案，使得动态任务适应能够以大幅降低的推理开销实现。

Large vision-language models (LVLMs) employ multi-modal in-context learning (MM-ICL) to adapt to new tasks by leveraging demonstration examples. While increasing the number of demonstrations boosts performance, they incur significant inference latency due to the quadratic computational cost of Transformer attention with respect to the context length. To address this trade-off, we propose Parallel In-Context Learning (Parallel-ICL), a plug-and-play inference algorithm. Parallel-ICL partitions the long demonstration context into multiple shorter, manageable chunks. It processes these chunks in parallel and integrates their predictions at the logit level, using a weighted Product-of-Experts (PoE) ensemble to approximate the full-context output. Guided by ensemble learning theory, we introduce principled strategies for Parallel-ICL: (i) clustering-based context chunking to maximize inter-chunk diversity and (ii) similarity-based context compilation to weight predictions by query relevance. Extensive experiments on VQA, image captioning, and classification benchmarks demonstrate that Parallel-ICL achieves performance comparable to full-context MM-ICL, while significantly improving inference speed. Our work offers an effective solution to the accuracy-efficiency trade-off in MM-ICL, enabling dynamic task adaptation with substantially reduced inference overhead.

评分理由: 论文聚焦于多模态大语言模型（MLLMs）在开放词汇情感识别中的应用，核心创新在于提出了一种混合证据演绎推理架构（HyDRA），采用强化学习进行奖励塑造以优化推理轨迹。因此，与"Large Language Models"高度相关（10分），因为直接使用MLLMs；与"RLHF"高度相关（10分），因为使用了强化学习进行训练对齐；与"Chain of Thought"和"System 2 Thinking"高度相关（10分），因为论文强调多步、深入的推理过程（Propose-Verify-Decide协议）；与"Mechanistic Interpretability"高度相关（10分），因为提供了可解释的证据追踪。其他关键词如MoE、量化、RAG等未在论文中涉及，得0分。

开放词汇多模态情感识别（Open-Vocabulary Multimodal Emotion Recognition, OV-MER）本质上具有挑战性，这源于模棱两可的多模态线索所固有的模糊性，这些线索通常源自未被观察到的、各不相同的动态情境。尽管多模态大语言模型（Multimodal Large Language Models, MLLMs）提供了广泛的语义覆盖，但其性能常受限于对主导数据先验的过早固化，从而产生次优的启发式策略，忽略了跨模态的关键互补性情感线索。我们认为，有效的情感推理不仅需要表层关联，更需要通过综合多个基于证据的推理依据，从不同的潜在视角调和这些观察结果，从而重构细腻的情感状态。为此，我们提出了HyDRA，一种混合证据演绎推理架构，它将推理形式化为一个“提议-验证-决策”协议。为了使这一溯因过程内化，我们采用了分层奖励塑形的强化学习方法，将推理轨迹与最终任务性能对齐，确保其能最佳地调和观察到的多模态线索。系统性评估验证了我们的设计选择：HyDRA在各项基准测试中均持续优于现有强基线模型——尤其是在模糊或冲突情境下——同时提供了可解释的诊断性证据轨迹。

Open-Vocabulary Multimodal Emotion Recognition (OV-MER) is inherently challenging due to the ambiguity of equivocal multimodal cues, which often stem from distinct unobserved situational dynamics. While Multimodal Large Language Models (MLLMs) offer extensive semantic coverage, their performance is often bottlenecked by premature commitment to dominant data priors, resulting in suboptimal heuristics that overlook crucial, complementary affective cues across modalities. We argue that effective affective reasoning requires more than surface-level association; it necessitates reconstructing nuanced emotional states by synthesizing multiple evidence-grounded rationales that reconcile these observations from diverse latent perspectives. We introduce HyDRA, a Hybrid-evidential Deductive Reasoning Architecture that formalizes inference as a Propose-Verify-Decide protocol. To internalize this abductive process, we employ reinforcement learning with hierarchical reward shaping, aligning the reasoning trajectories with final task performance to ensure they best reconcile the observed multimodal cues. Systematic evaluations validate our design choices, with HyDRA consistently outperforming strong baselines–especially in ambiguous or conflicting scenarios–while providing interpretable, diagnostic evidence traces.

评分理由: 论文核心研究大型MoE语言模型（Qwen 3.5-35B-A3B）的行为控制方法，通过稀疏自编码器（SAE）和解码探针向量实现代理行为的精细调控。高度相关的关键词包括：1）“Large Language Models”（研究35B参数模型）、2）“Mixture of Experts”（模型为MoE架构）、3）“LLM Agents”（研究代理行为控制）、4）“Tool Use”（涉及代码执行和网络搜索等工具使用）、5）“Mechanistic Interpretability”（通过SAE和探针进行模型解释和行为干预）。其他关键词如SLMs、训练方法、推理优化等与论文内容无关。

我们在Qwen 3.5-35B-A3B模型的残差流上训练了九个稀疏自编码器（Sparse Autoencoders, SAEs），该模型是一个拥有350亿参数的混合专家模型，采用门控Delta网络（GatedDeltaNet）与注意力混合架构，并利用这些SAEs识别和调控五种主体行为特质。我们的方法在SAE潜在激活上训练线性探测器，然后将探测器权重通过SAE解码器反向投影，以获得模型原生激活空间中的连续调控向量。这一方法绕过了SAE的top-k离散化过程，使得在推理时无需重新训练即可实现细粒度的行为干预。在1800次智能体推演（50种场景×36种条件）中，我们发现以乘数2进行自主性调控可达到科恩d值=1.01（p < 0.0001），使模型从78%的情况下向用户求助转变为主动执行代码和搜索网络。然而，跨特质分析表明，所有五个调控向量主要调节一个主导的主体性轴线（即独立行动倾向与遵从用户倾向之间的维度），特质特异性效应仅作为工具类型构成和剂量反应曲线形状的次要调节出现。工具使用向量能有效调控行为（d = 0.39）；风险校准向量仅产生抑制效应。我们还证明，仅在自回归解码阶段进行调控完全无效（p > 0.35），这为“行为决策在门控DeltaNet架构的前馈计算阶段形成”提供了因果性证据。

We train nine sparse autoencoders (SAEs) on the residual stream of Qwen 3.5-35B-A3B, a 35-billion-parameter Mixture-of-Experts model with a hybrid GatedDeltaNet/attention architecture, and use them to identify and steer five agentic behavioral traits. Our method trains linear probes on SAE latent activations, then projects the probe weights back through the SAE decoder to obtain continuous steering vectors in the model’s native activation space. This bypasses the SAE’s top-k discretization, enabling fine-grained behavioral intervention at inference time with no retraining. Across 1,800 agent rollouts (50 scenarios times 36 conditions), we find that autonomy steering at multiplier 2 achieves Cohen’s d = 1.01 (p < 0.0001), shifting the model from asking the user for help 78% of the time to proactively executing code and searching the web. Cross-trait analysis, however, reveals that all five steering vectors primarily modulate a single dominant agency axis (the disposition to act independently versus defer to the user), with trait specific effects appearing only as secondary modulations in tool-type composition and dose-response shape. The tool-use vector steers behavior (d = 0.39); the risk-calibration vector produces only suppression. We additionally show that steering only during autoregressive decoding has zero effect (p > 0.35), providing causal evidence that behavioral commitments are computed during prefill in GatedDeltaNet architectures.

评分理由: 论文核心研究多模态大语言模型（MLLMs）的上下文安全问题，因此与"Large Language Models"高度相关（10分）。提出的EchoSafe框架基于"Self-Reflection"机制，通过记忆库积累安全洞察，与"Self-Correction/Self-Improvement/Self-Reflection"高度相关（10分）。研究涉及模型对齐（Alignment）、检索增强（RAG）、思维链推理（CoT）、深度推理（System 2 Thinking）、幻觉缓解（Hallucination Mitigation）和上下文学习（In-context Learning），这些关键词与论文方法有一定关联（各5分）。其他关键词如MoE、量化、推理加速、科学AI等未在论文中涉及（0分）。

多模态大语言模型（MLLMs）在广泛的视觉推理任务中取得了显著性能，但其面临的安全风险脆弱性仍是一个紧迫问题。尽管先前研究主要集中于通过检测并拒绝显式不安全输入的越狱防御方法，此类方法往往忽视了上下文安全性——这要求模型能够区分看似相似但在安全意图上存在显著差异的场景之间的微妙上下文区别。在本研究中，我们提出了MM-SafetyBench++，这是一个为上下文安全评估精心构建的基准测试集。具体而言，针对每个不安全的图文对，我们通过最小程度的修改构建了对应的安全版本，这些修改在保持底层上下文语义的同时翻转了用户意图，从而能够可控地评估模型是否能基于上下文理解调整其安全行为。此外，我们提出了EchoSafe，这是一个无需训练的框架，通过维护自反思记忆库来积累并检索先前交互中的安全洞察。通过将相关的过往经验整合到当前提示中，EchoSafe能够在推理过程中实现上下文感知的推理和安全行为的持续演进。在多个多模态安全基准测试上的广泛实验表明，EchoSafe始终取得卓越性能，为推进MLLMs的上下文安全性建立了坚实的基线。所有基准数据与代码均公开于https://echosafe-mllm.github.io。

Multi-modal Large Language Models (MLLMs) have achieved remarkable performance across a wide range of visual reasoning tasks, yet their vulnerability to safety risks remains a pressing concern. While prior research primarily focuses on jailbreak defenses that detect and refuse explicitly unsafe inputs, such approaches often overlook contextual safety, which requires models to distinguish subtle contextual differences between scenarios that may appear similar but diverge significantly in safety intent. In this work, we present MM-SafetyBench++, a carefully curated benchmark designed for contextual safety evaluation. Specifically, for each unsafe image-text pair, we construct a corresponding safe counterpart through minimal modifications that flip the user intent while preserving the underlying contextual meaning, enabling controlled evaluation of whether models can adapt their safety behaviors based on contextual understanding. Further, we introduce EchoSafe, a training-free framework that maintains a self-reflective memory bank to accumulate and retrieve safety insights from prior interactions. By integrating relevant past experiences into current prompts, EchoSafe enables context-aware reasoning and continual evolution of safety behavior during inference. Extensive experiments on various multi-modal safety benchmarks demonstrate that EchoSafe consistently achieves superior performance, establishing a strong baseline for advancing contextual safety in MLLMs. All benchmark data and code are available at https://echosafe-mllm.github.io.

评分理由: 论文核心研究大语言模型在电商搜索领域的工业部署，直接涉及LLMs、知识幻觉缓解、参数高效微调、预训练/领域适应、指令调优/对齐等关键词。论文明确提到LLMs、知识幻觉问题、参数高效预训练策略、多任务指令调优和对抗训练对齐方法，这些是核心内容。推理链增强数据与CoT有一定关联。其他关键词如MoE、SLMs、RAG、量化等未在摘要中体现，评为无关。

大语言模型通过实现意图感知推荐，为电子商务搜索带来了变革性潜力。然而，其工业部署受到两个关键挑战的阻碍：(1) 由于对动态、细粒度产品知识编码不足导致的知识幻觉，以及(2) 在越狱攻击下威胁合规性的安全漏洞。为解决这些问题，我们提出了SI——一个用于构建知识丰富且安全的电商搜索大语言模型的“合成-注入-对齐”框架。我们的方法首先通过结合结构化知识图谱与非结构化行为日志，并辅以推理链和安全感知数据，合成高质量的自然语言语料库。随后，我们引入一种基于深度向上扩展的参数高效预训练策略，以注入领域知识，同时保留通用能力。最后，通过多任务指令微调和对抗性训练的双路径对齐方法，增强了任务性能与安全鲁棒性。该框架已在中国最大的自营电商平台京东部署，在五个核心搜索场景中的A/B测试表明，关键业务指标均有显著提升，验证了其工业有效性和可扩展性。

Large language models offer transformative potential for e-commerce search by enabling intent-aware recommendations. However, their industrial deployment is hindered by two critical challenges: (1) knowledge hallucination due to insufficient encoding of dynamic, fine-grained product knowledge, and (2) security vulnerabilities under jailbreak attacks that threaten compliance. To address these issues, we propose SI–a Synthesize-Inject-Align framework for building knowledgeable and secure e-commerce search LLMs. Our approach first synthesizes high-quality natural language corpus by combining structured knowledge graphs with unstructured behavioral logs, augmented with reasoning chains and safety-aware data.We then introduce a parameter-efficient pre-training strategy based on Depth Up-Scaling to inject domain knowledge while preserving general capabilities. Finally, a dual-path alignment method via multi-task instruction tuning and adversarial training strengthens both task performance and safety robustness. The framework has been deployed at JD.com, China’s largest self-operated e-commerce platform, where A/B tests across five core search scenarios demonstrate significant improvements in key business metrics, validating its industrial effectiveness and scalability.

评分理由: 论文《Attention-guided Evidence Grounding for Spoken Question Answering》提出了一种名为AEG的端到端框架，专门用于语音问答任务。该研究核心围绕Speech Large Language Models (SpeechLLMs)展开，因此与"Large Language Models"高度相关（10分）。论文提出了一个名为LFE的监督微调范式来校准模型的注意力机制，这直接对应"Supervised Fine-tuning”（10分）。研究目标之一是减少幻觉，因此与"Hallucination Mitigation"高度相关（10分）。论文通过注意力机制显式定位关键证据，这涉及模型内部工作机制的解释，与"Mechanistic Interpretability"有一定关联（5分）。框架利用了预训练模型，因此与"Pre-training"有一定关联（5分）。实验结果表明该方法减少了约62%的推理延迟，因此与"Inference Acceleration"有一定关联（5分）。论文未涉及其他关键词，如MoE、SLMs、Scaling Laws、Alignment、RLHF、RAG、CoT、Agents、Quantization等，这些均评为0分。

口语问答（Spoken QA）是一个具有挑战性的跨模态问题：它需要在有效对齐声学查询与文本知识的同时，避免基于级联自动语音识别（ASR）系统固有的延迟和错误传播。本文提出了一种新颖的端到端框架——注意力引导证据定位（Attention-guided Evidence Grounding, AEG），该框架利用语音大语言模型（Speech Large Language Models, SpeechLLMs）的内部跨模态注意力，在模型的隐式空间中显式地定位并锚定关键证据。针对预训练模型中注意力分布分散的问题，我们提出了学习聚焦证据（Learning to Focus on Evidence, LFE）这一监督微调范式，以校准模型的注意力机制，从而区分查询相关片段与无关上下文。在SQuAD、HotpotQA和MuSiQue数据集上的实验表明，AEG减少了幻觉现象，并实现了显著的效率提升，其性能优于大规模级联基线系统（Whisper-Large-v3 + Reranker），同时将推理延迟降低了约62%。

Spoken Question Answering (Spoken QA) presents a challenging cross-modal problem: effectively aligning acoustic queries with textual knowledge while avoiding the latency and error propagation inherent in cascaded ASR-based systems. In this paper, we introduce Attention-guided Evidence Grounding (AEG), a novel end-to-end framework that leverages the internal cross-modal attention of Speech Large Language Models (SpeechLLMs) to explicitly locate and ground key evidence in the model’s latent space. To address the diffuse attention distribution in pre-trained models, we propose Learning to Focus on Evidence (LFE), a supervised fine-tuning paradigm that calibrates the model’s attention mechanism to distinguish query-relevant segments from irrelevant context. Experiments on SQuAD, HotpotQA, and MuSiQue demonstrate that AEG reduces hallucinations and achieves strong efficiency gains, outperforming large-scale cascaded baselines (Whisper-Large-v3 + Reranker) while reducing inference latency by approximately 62%.

评分理由: 论文提出InViC框架，用于医学视觉问答（Med-VQA），属于大模型在生物医学领域的应用创新。核心相关关键词：1）“Large Language Models” (10分)：论文明确使用多模态大语言模型（MLLMs）作为基础。2）“PEFT” (10分)：论文采用LoRA进行微调，是参数高效微调技术的应用。3）“AI for Science” (10分)：医学VQA是AI在生物医学领域的典型应用。4）“Hallucination Mitigation” (8分)：论文旨在解决MLLMs的“捷径回答”问题，提升临床可靠性，与缓解幻觉/提升事实性高度相关。5）“Post-training” (5分)：涉及两阶段微调策略，属于训练后调整。其他关键词与论文内容无直接关联，均给0分。

医学视觉问答旨在基于医学图像回答临床相关问题。然而，现有的多模态大语言模型常表现出捷径回答现象，即通过利用语言先验或数据集偏差生成看似合理的回答，却未能充分关注视觉证据。这种行为损害了临床可靠性，尤其是在细微影像学发现具有决定性意义时。我们提出了一种轻量级插件框架——意图感知视觉线索，以显式增强医学视觉问答中基于图像的答案生成。该框架引入了一个线索令牌提取模块，将密集的视觉令牌提炼为一组紧凑的K个问题条件化线索令牌，这些令牌作为结构化的视觉中介注入到大语言模型解码器中，以促进与意图对齐的视觉证据利用。为阻止模型绕过视觉信息，我们进一步设计了一种包含线索瓶颈注意力掩码的两阶段微调策略。在第一阶段，我们采用注意力掩码阻断大语言模型对原始视觉特征的直接访问，从而迫使所有视觉证据通过线索通路传递。在第二阶段，恢复标准的因果注意力机制，训练大语言模型联合利用视觉令牌和线索令牌。我们在三个公开的医学视觉问答基准数据集上评估了该框架，涵盖多个代表性多模态大语言模型。该框架在零样本推理和标准LoRA微调基础上均取得持续改进，表明通过瓶颈式训练的意图感知视觉线索是提升可信赖医学视觉问答系统实用性和有效性的策略。

Medical visual question answering (Med-VQA) aims to answer clinically relevant questions grounded in medical images. However, existing multimodal large language models (MLLMs) often exhibit shortcut answering, producing plausible responses by exploiting language priors or dataset biases while insufficiently attending to visual evidence. This behavior undermines clinical reliability, especially when subtle imaging findings are decisive. We propose a lightweight plug-in framework, termed Intent-aware Visual Cues (InViC), to explicitly enhance image-based answer generation in medical VQA. InViC introduces a Cue Tokens Extraction (CTE) module that distills dense visual tokens into a compact set of K question-conditioned cue tokens, which serve as structured visual intermediaries injected into the LLM decoder to promote intent-aligned visual evidence. To discourage bypassing of visual information, we further design a two-stage fine-tuning strategy with a cue-bottleneck attention mask. In Stage I, we employ an attention mask to block the LLM’s direct view of raw visual features, thereby funneling all visual evidence through the cue pathway. In Stage II, standard causal attention is restored to train the LLM to jointly exploit the visual and cue tokens. We evaluate InViC on three public Med-VQA benchmarks (VQA-RAD, SLAKE, and ImageCLEF VQA-Med 2019) across multiple representative MLLMs. InViC consistently improves over zero-shot inference and standard LoRA fine-tuning, demonstrating that intent-aware visual cues with bottlenecked training is a practical and effective strategy for improving trustworthy Med-VQA.

评分理由: 论文核心研究移动设备上的大语言模型（OD-LLMs）设计，与"Large Language Models"和"Small Language Models/On-device AI"高度相关（10分）。论文提出支持8k上下文长度，与"Context Window Extension/Long Context LLMs"相关（8分）。论文优化延迟和推理速度，与"Speculative Decoding/Inference Acceleration"相关（8分）。论文提到使用预训练骨干网络和少量继续预训练，与"Pre-training/Continual Pre-training/Domain Adaptation"有一定关联（5分）。其他关键词如MoE、SFT、RLHF、RAG、量化等未在摘要中提及，评为0分。

实时人工智能体验需要针对资源受限硬件进行高效部署优化的设备端大语言模型（On-Device Large Language Models，简称OD-LLMs）。最实用的OD-LLM应能产生近实时响应，并具备广泛的硬件兼容性，从而最大化用户覆盖范围。本文提出一种在移动端延迟约束下，采用硬件在环架构搜索来设计此类模型的方法论。该系统适用于工业级部署：所生成的模型无需定制内核即可部署，且兼容Executorch等标准移动运行时环境。本方法避免使用特殊的注意力机制，转而采用注意力跳跃技术来实现长上下文加速。

我们的方法联合优化了模型架构（层数、维度）与注意力模式。为高效评估候选架构，我们将每个候选视为继承预训练主干网络权重的剪枝版本，从而通过极少的持续预训练实现高精度。我们利用延迟评估的低成本特性，采用分阶段流程：首先学习精确的延迟模型，随后在延迟与质量间搜索帕累托前沿。

该方法产生了MobileLLM-Flash系列基础模型（3.5亿、6.5亿、14亿参数），专为高效设备端使用而设计，具备强大能力并支持最高8k的上下文长度。在移动端CPU上，MobileLLM-Flash在保持相当或更优质量的前提下，实现了最高1.8倍的预填充加速和1.6倍的解码加速。我们对帕累托前沿设计选择的分析，为OD-LLM设计提供了可实践的原则指导。

Real-time AI experiences call for on-device large language models (OD-LLMs) optimized for efficient deployment on resource-constrained hardware. The most useful OD-LLMs produce near-real-time responses and exhibit broad hardware compatibility, maximizing user reach. We present a methodology for designing such models using hardware-in-the-loop architecture search under mobile latency constraints. This system is amenable to industry-scale deployment: it generates models deployable without custom kernels and compatible with standard mobile runtimes like Executorch. Our methodology avoids specialized attention mechanisms and instead uses attention skipping for long-context acceleration. Our approach jointly optimizes model architecture (layers, dimensions) and attention pattern. To efficiently evaluate candidates, we treat each as a pruned version of a pretrained backbone with inherited weights, thereby achieving high accuracy with minimal continued pretraining. We leverage the low cost of latency evaluation in a staged process: learning an accurate latency model first, then searching for the Pareto-frontier across latency and quality. This yields MobileLLM-Flash, a family of foundation models (350M, 650M, 1.4B) for efficient on-device use with strong capabilities, supporting up to 8k context length. MobileLLM-Flash delivers up to 1.8x and 1.6x faster prefill and decode on mobile CPUs with comparable or superior quality. Our analysis of Pareto-frontier design choices offers actionable principles for OD-LLM design.

评分理由: 论文核心研究指令调优的LLMs在阿拉伯语形态句法标注和依存句法分析任务上的表现，直接涉及LLMs、指令调优和检索增强的上下文学习；论文还评估了零样本提示和基于检索的上下文学习，与"In-context Learning"高度相关；其他关键词如MoE、SLMs、缩放定律、预训练、对齐技术、推理方法、代理系统、模型压缩等均未在摘要中提及，与论文内容无关。

大型语言模型（LLM）在众多自然语言处理任务中表现优异，但其生成显式语言结构的能力尚不明确。本研究以标准阿拉伯语的两项结构化预测任务——形态句法标注与带标签依存句法分析——为基准，对指令微调后的LLM进行评估。阿拉伯语因其丰富的形态变化与正字法歧义性，形成了强烈的形态-句法互动，为测试提供了极具挑战性的平台。我们比较了零样本提示与基于检索的上下文学习（ICL）方法，后者使用了阿拉伯语树库中的示例。结果表明，提示设计与示例选取对性能有显著影响：在特征级标注任务上，专有模型接近监督基线的水平，并在依存句法分析任务上与专用解析器相比具有竞争力。在原始文本场景下，分词仍是挑战，但基于检索的ICL同时提升了句法分析与分词的效果。我们的分析揭示了LLM对阿拉伯语形态句法与句法的哪些方面能够可靠捕捉，哪些方面仍存在困难。

Large language models (LLMs) perform strongly on many NLP tasks, but their ability to produce explicit linguistic structure remains unclear. We evaluate instruction-tuned LLMs on two structured prediction tasks for Standard Arabic: morphosyntactic tagging and labeled dependency parsing. Arabic provides a challenging testbed due to its rich morphology and orthographic ambiguity, which create strong morphology-syntax interactions. We compare zero-shot prompting with retrieval-based in-context learning (ICL) using examples from Arabic treebanks. Results show that prompt design and demonstration selection strongly affect performance: proprietary models approach supervised baselines for feature-level tagging and become competitive with specialized dependency parsers. In raw-text settings, tokenization remains challenging, though retrieval-based ICL improves both parsing and tokenization. Our analysis highlights which aspects of Arabic morphosyntax and syntax LLMs capture reliably and which remain difficult.

评分理由: 论文核心研究LLMs的安全机制和对抗性攻击（越狱），因此与"Large Language Models"高度相关（10分）。论文涉及模型安全对齐、推理过程（多步推理、深度推理）、事实性/幻觉缓解以及可解释性（分析语义重构），这些与"Instruction Tuning/Alignment"、“Chain of Thought”、“System 2 Thinking”、“Hallucination Mitigation”、“Mechanistic Interpretability"有一定关联（各5分）。其他关键词如MoE、SLMs、训练方法、压缩、加速、智能体、科学AI等，论文未涉及，给0分。

现代大型语言模型采用的安全机制已超越表层输入过滤，延伸至潜在语义表征与生成时推理层面，使其能够在推理过程中识别被混淆的恶意意图并据此拒绝响应，从而导致许多表层混淆越狱攻击失效。本文提出结构化语义伪装（Structured Semantic Cloaking, S2C）——一种新颖的多维度越狱攻击框架，通过操控模型推理过程中恶意语义意图的重建方式实现攻击。S2C策略性地分布并重塑语义线索，使得完整意图的整合需要多步推理及深层潜在表征中的长距离共指消解。该框架包含三种互补机制：（1）语境重构：将请求嵌入具有高利害关系的合理场景中，使模型倾向于遵从；（2）内容碎片化：将请求的语义特征分散至互不关联的提示片段；（3）线索引导伪装：在嵌入可恢复标记以引导输出生成的同时，对残留语义线索进行隐蔽处理。通过延迟并重组语义整合过程，S2C能够削弱那些依赖于解码阶段连贯或显性重构恶意意图的安全触发器，同时保留足够的指令可恢复性以生成功能性输出。我们在HarmBench和JBB-Behaviors基准上对多种开源与专有大型语言模型评估S2C，其攻击成功率（Attack Success Rate, ASR）较当前最优方法分别提升12.4%与9.7%。值得注意的是，S2C在GPT-5-mini上取得显著优势，在JBB-Behaviors基准上超越最强基线26%。我们还分析了针对广泛模型族的最优组合策略，并刻画了混淆程度与输入可恢复性之间的权衡对越狱成功率的影响。

Modern LLMs employ safety mechanisms that extend beyond surface-level input filtering to latent semantic representations and generation-time reasoning, enabling them to recover obfuscated malicious intent during inference and refuse accordingly, and rendering many surface-level obfuscation jailbreak attacks ineffective. We propose Structured Semantic Cloaking (S2C), a novel multi-dimensional jailbreak attack framework that manipulates how malicious semantic intent is reconstructed during model inference. S2C strategically distributes and reshapes semantic cues such that full intent consolidation requires multi-step inference and long-range co-reference resolution within deeper latent representations. The framework comprises three complementary mechanisms: (1) Contextual Reframing, which embeds the request within a plausible high-stakes scenario to bias the model toward compliance; (2) Content Fragmentation, which disperses the semantic signature of the request across disjoint prompt segments; and (3) Clue-Guided Camouflage, which disguises residual semantic cues while embedding recoverable markers that guide output generation. By delaying and restructuring semantic consolidation, S2C degrades safety triggers that depend on coherent or explicitly reconstructed malicious intent at decoding time, while preserving sufficient instruction recoverability for functional output generation. We evaluate S2C across multiple open-source and proprietary LLMs using HarmBench and JBB-Behaviors, where it improves Attack Success Rate (ASR) by 12.4% and 9.7%, respectively, over the current SOTA. Notably, S2C achieves substantial gains on GPT-5-mini, outperforming the strongest baseline by 26% on JBB-Behaviors. We also analyse which combinations perform best against broad families of models, and characterise the trade-off between the extent of obfuscation versus input recoverability on jailbreak success.

评分理由: 论文核心研究大语言模型（LLMs）的对齐问题，特别是比较基于人类反馈的强化学习（RLHF）与仅使用负面反馈的方法。因此，与"Large Language Models”、“Alignment"和"RLHF"高度相关（10分）。论文未涉及其他关键词的具体技术或应用，如MoE、SLMs、Scaling Laws、Pre-training、PEFT、RAG、推理方法、代理、压缩、科学AI等，这些评分为0分。

近期实证研究表明，仅使用负面反馈训练大型语言模型（LLM）的效果可媲美甚至超越基于人类反馈的强化学习（RLHF）标准方法。负面样本强化学习在数学推理任务上达到与近端策略优化（PPO）相当的水平；分布化非偏好优化仅通过非偏好样本即可有效训练；而宪法人工智能在无害性基准测试中表现优于纯RLHF方法。然而，目前尚无统一的理论框架解释负面信号为何如此有效。本文提出一种理论解释：积极偏好与消极约束在结构上具有不对称性。积极偏好（“哪个更好”）编码了连续耦合、依赖语境的人类价值观，这些价值观无法被穷尽描述——导致模型学习到诸如迎合用户（谄媚性）等表层关联特征。消极约束（“什么是错误的”）则编码了离散、有限、可独立验证的禁令，能够收敛至稳定边界。这种不对称性——根植于波普尔的证伪逻辑与否定知识认识论——既解释了基于偏好的RLHF产生谄媚性缺陷的原因，也揭示了负面信号方法惊人有效性的机理。我们认为，对齐研究应将其重心从“学习人类偏好”转向“学习人类拒绝的内容”，并为此框架提供了可检验的预测。

Recent empirical results have demonstrated that training large language models (LLMs) with negative-only feedback can match or exceed standard reinforcement learning from human feedback (RLHF). Negative Sample Reinforcement achieves parity with PPO on mathematical reasoning; Distributional Dispreference Optimization trains effectively using only dispreferred samples; and Constitutional AI outperforms pure RLHF on harmlessness benchmarks. Yet no unified theoretical account explains why negative signals are so effective. This paper proposes such an account: positive preferences and negative constraints are structurally asymmetric. Positive preferences (“which is better”) encode continuously coupled, context-dependent human values that cannot be exhaustively specified – leading models to learn surface correlates such as agreement with the user (sycophancy). Negative constraints (“what is wrong”) encode discrete, finite, independently verifiable prohibitions that can converge to a stable boundary. This asymmetry – rooted in Popper’s falsification logic and the epistemology of negative knowledge – explains both the sycophancy failure of preference-based RLHF and the surprising effectiveness of negative-signal methods. We argue that alignment research should shift its center of gravity from “learning what humans prefer” to “learning what humans reject,” and offer testable predictions for this framework.

评分理由: 论文核心研究大语言模型（LLMs）的后训练压缩技术，特别是针对剪枝（pruning）和量化（quantization）的校准数据选择方法。因此，与"Large Language Models"和"Post-training"高度相关（10分），与"Quantization"高度相关（10分）。论文未涉及其他关键词如MoE、SLMs、Scaling Laws、Pre-training、Instruction Tuning、RLHF、PEFT、RAG、Context Window、KV Cache、Reasoning、Agents、Tool Use、Multi-agent、Speculative Decoding、Hallucination、Interpretability、World Models、Model Merging、In-context Learning或AI for Science，故均给0分。

后训练模型压缩对于提升大语言模型（LLM）的可移植性同时保持其性能至关重要。尽管已有多种压缩方法被提出，但如何选择最合适的数据集（即所谓的校准数据）以寻找压缩模型配置的问题尚未得到足够重视。校准数据的选择是保持模型在任务内与任务间能力的关键步骤。在本工作中，我们通过分析数据的内在属性而非模型特定信号，来解决为剪枝和量化识别高性能校准集的挑战。我们提出了 ZipCal，一种与模型无关的数据筛选策略，该方法基于齐夫幂律最大化词汇多样性。实验表明，我们的方法在各种剪枝基准测试中均持续优于标准的均匀随机采样。值得注意的是，在下游性能方面，其表现与依赖模型困惑度的前沿方法相当。后者在大规模模型和数据集上计算成本极高，而 ZipCal 由于其可处理的线性复杂度，平均速度提升约 240 倍（代码与实验已公开于 https://anonymous.4open.science/r/zipcal-71CD/）。

Post-training model compression is essential for enhancing the portability of Large Language Models (LLMs) while preserving their performance. While several compression approaches have been proposed, less emphasis has been placed on selecting the most suitable set of data (the so-called \emph{calibration data}) for finding the compressed model configuration. The choice of calibration data is a critical step in preserving model capabilities both intra- and inter-tasks. In this work, we address the challenge of identifying high-performance calibration sets for both pruning and quantization by analyzing intrinsic data properties rather than model-specific signals. We introduce \texttt{\textbf{ZipCal}}, a model-agnostic data curation strategy that maximizes lexical diversity based on Zipfian power laws. Experiments demonstrate that our method consistently outperforms standard uniform random sampling across various pruning benchmarks. Notably, it also performs on par, in terms of downstream performance, with a state-of-the-art method that relies on model perplexity. The latter becomes prohibitively expensive at large-scale models and datasets, while \texttt{\textbf{ZipCal}} is on average $\sim$240$\times$ faster due to its tractable linear complexity\footnote{We make the code and the experiments available at https://anonymous.4open.science/r/zipcal-71CD/.}.

评分理由: 论文核心研究LLM作为合成代理（LLM agents）在公共意见模拟中的应用，直接涉及"Large Language Models"和"LLM Agents"关键词，给予10分。论文提到价值取向（value orientations）的注入，与"Value Alignment"有一定关联，给予5分。其他关键词如MoE、SFT、RAG等未在摘要中提及，与论文内容无关，给予0分。

大型语言模型（LLM）近期被采纳为合成代理用于公众意见模拟，为成本高昂且耗时的人工调查提供了一种前景广阔的替代方案。尽管具备可扩展性，当前基于LLM的模拟方法仍无法捕捉社会多样性，导致群体间差异被扁平化，且人口统计群体内部的回应过于同质化。我们将这一局限识别为LLM隐藏表征中的“多样性坍缩”现象，即不同社会身份在模型各层间逐渐变得难以区分。基于此观察，我们提出了参数化社会身份注入（Parametric Social Identity Injection，PSII）这一通用框架，该框架将人口属性与价值取向的显式参数化表征直接注入LLM的中间隐藏状态。与基于提示的人物设定方法不同，PSII能够在表征层面实现细粒度且可控的身份调制。基于世界价值观调查（World Values Survey）并使用多种开源LLM进行的大量实验表明，PSII显著提升了分布保真度与多样性，在增强整体多样性的同时，降低了模拟结果与真实世界调查数据之间的KL散度。这项工作为LLM代理的表征层面控制提供了新见解，并推动了可扩展、具备多样性意识的公众意见模拟的发展。代码与数据可在 https://github.com/halsayxi/PSII 获取。

Large language models (LLMs) have recently been adopted as synthetic agents for public opinion simulation, offering a promising alternative to costly and slow human surveys. Despite their scalability, current LLM-based simulation methods fail to capture social diversity, producing flattened inter-group differences and overly homogeneous responses within demographic groups. We identify this limitation as a Diversity Collapse phenomenon in LLM hidden representations, where distinct social identities become increasingly indistinguishable across layers. Motivated by this observation, we propose Parametric Social Identity Injection (PSII), a general framework that injects explicit, parametric representations of demographic attributes and value orientations directly into intermediate hidden states of LLMs. Unlike prompt-based persona conditioning, PSII enables fine-grained and controllable identity modulation at the representation level. Extensive experiments on the World Values Survey using multiple open-source LLMs show that PSII significantly improves distributional fidelity and diversity, reducing KL divergence to real-world survey data while enhancing overall diversity. This work provides new insights into representation-level control of LLM agents and advances scalable, diversity-aware public opinion simulation. Code and data are available at https://github.com/halsayxi/PSII.

评分理由: 论文提出CABTO框架，利用预训练大模型（LMs）自动化构建机器人行为树系统，属于大模型在机器人领域的应用研究。与"Large Language Models"高度相关（8分），因为摘要明确提到"leverages pre-trained Large Models (LMs)"。与"LLM Agents"高度相关（8分），因为该框架本质上是一个利用大模型进行决策和规划的自主代理系统。与"Pre-training"有一定关联（5分），因为使用了预训练模型。其他关键词如MoE、SFT、RAG等均未在论文中涉及，得0分。

行为树（Behavior Trees，BTs）为设计模块化、反应式的机器人控制器提供了一种强大的范式。行为树规划作为一个新兴领域，为自动化生成可靠的行为树提供了理论保证。然而，行为树规划通常假设一个精心设计的行为树系统已经具备基础——即包含高层动作模型和底层控制策略——这往往需要大量的专家知识与人工投入。本文正式定义了行为树基础构建问题：即自动化构建一个完整且一致的行为树系统。我们分析了该问题的复杂性，并提出了CABTO（上下文感知行为树基础构建框架），这是首个能高效解决此挑战的框架。CABTO利用预训练大模型（Large Models，LMs），在行为树规划器的上下文反馈与环境观测的引导下，启发式地搜索动作模型与控制策略的空间。在三种不同机器人操作场景下的七组任务实验中，CABTO在生成完整且一致的行为树系统方面展现出卓越的有效性与效率。

Behavior Trees (BTs) offer a powerful paradigm for designing modular and reactive robot controllers. BT planning, an emerging field, provides theoretical guarantees for the automated generation of reliable BTs. However, BT planning typically assumes that a well-designed BT system is already grounded – comprising high-level action models and low-level control policies – which often requires extensive expert knowledge and manual effort. In this paper, we formalize the BT Grounding problem: the automated construction of a complete and consistent BT system. We analyze its complexity and introduce CABTO (Context-Aware Behavior Tree grOunding), the first framework to efficiently solve this challenge. CABTO leverages pre-trained Large Models (LMs) to heuristically search the space of action models and control policies, guided by contextual feedback from BT planners and environmental observations. Experiments spanning seven task sets across three distinct robotic manipulation scenarios demonstrate CABTO’s effectiveness and efficiency in generating complete and consistent behavior tree systems.

评分理由: 论文核心研究LLMs在低资源机器翻译中的应用，通过结合语言相关的枢轴语言和少样本上下文示例进行推理时适配，不涉及参数更新。因此，与"Large Language Models"和"In-context Learning"高度相关（10分），因为论文明确研究LLMs和少样本上下文学习。其他关键词如MoE、SFT、RAG、量化等均未在摘要中提及或相关，故得0分。

大语言模型（LLMs）已在众多下游任务中展现出强大性能，但其在极低资源机器翻译场景下的有效性仍显不足。标准的适应技术通常依赖于大规模平行语料或大量微调，这对于众多资源匮乏的语种而言并不可行。本研究探讨了一个更具约束性的问题：在数据稀缺的情况下，语言相似的枢轴语言和少量示例能在多大程度上为大语言模型的即时适应提供有效指导？我们设计了一种数据高效的实验方案，将语言相关的枢轴语言与少量上下文示例相结合，且不进行任何参数更新，并在受控条件下评估翻译表现。分析表明，尽管基于枢轴语言的提示方法在某些配置下能带来改进——尤其是在目标语言在模型词汇表中表征不足的情况下——但其提升幅度通常有限，且对少量示例的构建方式较为敏感。对于高度相关或表征较好的语言变体，我们观察到其增益会递减或不稳定。本研究结果提供了实证依据，说明了在低资源翻译场景中，如何以及何时可将推理时提示与基于枢轴语言的示例作为微调的一种轻量级替代方案。

Large Language Models (LLMs) have achieved strong performance across many downstream tasks, yet their effectiveness in extremely low-resource machine translation remains limited. Standard adaptation techniques typically rely on large-scale parallel data or extensive fine-tuning, which are infeasible for the long tail of underrepresented languages. In this work, we investigate a more constrained question: in data-scarce settings, to what extent can linguistically similar pivot languages and few-shot demonstrations provide useful guidance for on-the-fly adaptation in LLMs? We study a data-efficient experimental setup that combines linguistically related pivot languages with few-shot in-context examples, without any parameter updates, and evaluate translation behavior under controlled conditions. Our analysis shows that while pivot-based prompting can yield improvements in certain configurations, particularly in settings where the target language is less well represented in the model’s vocabulary, the gains are often modest and sensitive to few shot example construction. For closely related or better represented varieties, we observe diminishing or inconsistent gains. Our findings provide empirical guidance on how and when inference-time prompting and pivot-based examples can be used as a lightweight alternative to fine-tuning in low-resource translation settings.

评分理由: 论文核心研究LLMs在金融领域（期权策略生成）的应用，属于大模型在不同领域的研究应用。高度相关（10分）的关键词只有"Large Language Models”，因为论文明确使用LLMs作为核心方法。“Chain of Thought"和"System 2 Thinking"得5分，因为论文提到期权设计需要推理（reasoning）多维数据和约束，涉及多步逻辑思考，但并非论文主要技术焦点。其他关键词如MoE、SFT、RAG等均未在摘要中提及，完全无关（0分）。

大型语言模型（LLM）在通用代码生成方面表现出色，但将自然语言交易意图转化为正确的期权策略仍具挑战性。现实中的期权设计需要对海量、多维度的期权链数据进行严格约束下的推理，这往往使直接生成方法难以应对。我们引入期权查询语言（Option Query Language, OQL），这是一种领域特定的中间表示，在语法规则下将期权市场抽象为高层级原语，使LLM能够充当可靠的语义解析器而非自由形式的编程器。OQL查询随后由引擎进行确定性验证和执行，以实例化可执行策略。我们还为此任务提出了一个新的数据集，并证明我们的神经符号混合流程相较于直接基线方法，在执行准确性与逻辑一致性方面均有显著提升。

Large Language Models (LLMs) excel at general code generation, yet translating natural-language trading intents into correct option strategies remains challenging. Real-world option design requires reasoning over massive, multi-dimensional option chain data with strict constraints, which often overwhelms direct generation methods. We introduce the Option Query Language (OQL), a domain-specific intermediate representation that abstracts option markets into high-level primitives under grammatical rules, enabling LLMs to function as reliable semantic parsers rather than free-form programmers. OQL queries are then validated and executed deterministically by an engine to instantiate executable strategies. We also present a new dataset for this task and demonstrate that our neuro-symbolic pipeline significantly improves execution accuracy and logical consistency over direct baselines.

评分理由: 论文研究在冻结的编码器-解码器大语言模型（Flan-T5-XL）中实现持久记忆的方法，属于大模型技术原理的创新。核心涉及大语言模型（LLMs）和参数高效微调（PEFT），因为论文使用冻结主干和可训练的小型适配器来实现记忆功能。其他关键词如MoE、SLMs、Scaling Laws、RAG、推理加速等与论文内容无关，论文未涉及这些技术。

冻结的编码器-解码器语言模型是无状态的：每次前向传播后潜在表示即被丢弃，因此跨会话间不存在信息留存。本文提出一项概念验证性先导研究，证明在冻结大语言模型的连续潜在空间中实现持久记忆是可行的——即使在严苛的资源限制下（单一冻结的Flan-T5-XL主干网络、小型可训练适配器、单一数据集）。我们实现了涵盖三个注入点和四种写入机制的六种架构方法；与文本级记忆系统不同，每次写入和读取都是对稠密向量的可微分操作。仅训练适配器后，记忆库在无需梯度的推理阶段持续积累，实现了对话式学习。在LoCoMo数据集上采用遗忘曲线评估两种容量规模（1倍与10倍）时，无状态基线得分恰好为零；在10倍容量下，所有六个训练后的适配器均产生正向记忆召回曲线；在1倍容量下三种方法失效，表明容量是关键设计参数。由于记忆库是紧凑的数值阵列，其容量可扩展至任意规模而无需改动主干网络。我们认为，采用更大模型、更庞大数据集及数量级更大记忆库的端到端完整训练将产生显著更强的结果；本先导研究确立了此类工作所需的可行性基线及设计空间分类体系。

Frozen encoder–decoder language models are stateless: the latent representation is discarded after every forward pass, so no information persists across sessions. This paper presents a \textbf{proof-of-concept pilot study} showing that persistent memory in the \emph{continuous latent space} of a frozen LLM is feasible – even under severe resource constraints (a single frozen Flan-T5-XL backbone, small trainable adapters, a single dataset). We implement six architectural methods spanning three injection points and four write mechanisms; unlike text-level memory systems, every write and read is a differentiable operation on dense vectors. After training only the adapter, the memory bank continues to accumulate at inference time without gradients, enabling \emph{conversational learning}. Under a forgetting-curve evaluation on LoCoMo at two capacity scales (1$\times$ and 10$\times$), the stateless baseline scores exactly zero; at 10$\times$ all six trained adapters produce positive memory-recall curves; at 1$\times$ three methods collapse, revealing capacity as a critical design parameter. Because the memory bank is a compact numerical array, it can be scaled to arbitrarily large capacity without altering the backbone. We argue that full end-to-end training with larger models, larger data, and orders-of-magnitude larger memory will yield substantially stronger results; this pilot study establishes the feasibility baseline and design-space taxonomy that such efforts require.

评分理由: 论文专注于大语言模型（LLMs）的梯度反演攻击和隐私风险，核心是提出SOMP方法来解决大规模批处理和长序列下的文本重建问题。因此，仅与关键词"Large Language Models” OR “LLMs” OR “Foundation Models"高度相关（10分），因为论文明确研究LLMs的隐私漏洞和攻击方法。其他关键词涉及模型架构、训练技术、推理优化、应用领域等，论文未直接涉及，故均评0分。

梯度反演攻击表明，共享的梯度可能被用于重建私有训练文本，这对大语言模型（LLM）构成了隐私风险。尽管现有方法在小批量设置下表现良好，但由于严重的信号混合、高昂的计算成本以及保真度下降，将其扩展到更大的批处理规模和更长的序列仍然具有挑战性。我们提出了SOMP（子空间引导的正交匹配追踪），一种可扩展的梯度反演框架，它将从聚合梯度中恢复文本的问题转化为稀疏信号恢复问题。我们的核心见解是，聚合后的Transformer梯度保留了可利用的头部几何结构以及样本级别的稀疏性。SOMP利用这些特性逐步缩小搜索空间并解耦混合信号，而无需进行穷举搜索。在多个LLM系列、不同模型规模和五种语言上的实验表明，在聚合梯度场景下，SOMP始终优于现有方法。在批处理大小B=16的长序列设置中，SOMP实现了比强基线方法显著更高的重建保真度，同时保持了计算效率上的竞争力。即使在极端聚合条件下（高达B=128），SOMP仍能恢复有意义的文本，这表明在现有攻击方法效果大幅下降的场景下，隐私泄露风险可能依然存在。

Gradient inversion attacks reveal that private training text can be reconstructed from shared gradients, posing a privacy risk to large language models (LLMs). While prior methods perform well in small-batch settings, scaling to larger batch sizes and longer sequences remains challenging due to severe signal mixing, high computational cost, and degraded fidelity. We present SOMP (Subspace-Guided Orthogonal Matching Pursuit), a scalable gradient inversion framework that casts text recovery from aggregated gradients as a sparse signal recovery problem. Our key insight is that aggregated transformer gradients retain exploitable head-wise geometric structure together with sample-level sparsity. SOMP leverages these properties to progressively narrow the search space and disentangle mixed signals without exhaustive search. Experiments across multiple LLM families, model scales, and five languages show that SOMP consistently outperforms prior methods in the aggregated-gradient regime.For long sequences at batch size B=16, SOMP achieves substantially higher reconstruction fidelity than strong baselines, while remaining computationally competitive. Even under extreme aggregation (up to B=128), SOMP still recovers meaningful text, suggesting that privacy leakage can persist in regimes where prior attacks become much less effective.

评分理由: 论文《Mixture of Style Experts for Diverse Image Stylization》专注于计算机视觉领域的图像风格化任务，采用基于扩散模型和混合专家（MoE）的框架。其核心创新在于将MoE架构应用于风格化，以处理从浅层纹理到深层语义的多样化风格。因此，仅与关键词“Mixture of Experts” OR “MoE” OR “Sparse Models”高度相关（评分为10分），因为MoE是其核心方法。其他关键词均涉及大语言模型（LLM）或特定NLP/大模型技术（如RLHF、RAG、量化等），而本文研究的是图像处理，未涉及任何语言模型或相关技术，故其余关键词评分为0分。加权总分计算为10.0（10.0 × 1.0）。

基于扩散的风格化方法已取得显著进展，但现有方法多局限于色彩驱动的转换，忽略了复杂的语义与材质细节。本文提出StyleExpert，一种基于专家混合模型（Mixture of Experts, MoE）的语义感知框架。该框架采用一个统一的风格编码器，通过在我们构建的大规模内容-风格-风格化三元组数据集上进行训练，将多样化的风格嵌入到一致的潜在空间中。该嵌入表示随后用于驱动一个相似性感知的门控机制，动态地将不同风格路由至MoE架构中的特定专家模块。借助这一MoE架构，我们的方法能够灵活处理涵盖多个语义层次的多样化风格，从浅层纹理到深层语义。大量实验表明，StyleExpert在保持语义与材质细节方面优于现有方法，并能有效泛化至未见过的风格。我们的代码与收集的图像已在项目页面公开：https://hh-lg.github.io/StyleExpert-Page/。

Diffusion-based stylization has advanced significantly, yet existing methods are limited to color-driven transformations, neglecting complex semantics and material details.We introduce StyleExpert, a semantic-aware framework based on the Mixture of Experts (MoE). Our framework employs a unified style encoder, trained on our large-scale dataset of content-style-stylized triplets, to embed diverse styles into a consistent latent space. This embedding is then used to condition a similarity-aware gating mechanism, which dynamically routes styles to specialized experts within the MoE architecture. Leveraging this MoE architecture, our method adeptly handles diverse styles spanning multiple semantic levels, from shallow textures to deep semantics. Extensive experiments show that StyleExpert outperforms existing approaches in preserving semantics and material details, while generalizing to unseen styles. Our code and collected images are available at the project page: https://hh-lg.github.io/StyleExpert-Page/.

评分理由: 论文S-VAM专注于机器人学习中的视频动作模型（VAM），提出了一种通过自蒸馏实现单步推理的高效模型。其核心是视觉基础模型（VFM）和扩散模型的应用，属于计算机视觉和机器人领域，而非大语言模型（LLM）的直接研究。因此，仅与关键词"Large Language Models” OR “LLMs” OR “Foundation Models"有间接关联（因提及"vision foundation model”，属于基础模型范畴），给予5分；其他关键词均与论文内容无关，给予0分。

视频动作模型（Video Action Models, VAMs）因其对复杂操作任务具有强大的视觉预见能力，已成为机器人学习领域一种前景广阔的范式。然而，现有的视频动作模型通常依赖于缓慢的多步视频生成或噪声较多的一步特征提取，无法同时保证实时推理与高保真度的预见效果。为克服这一局限，我们提出了S-VAM，一种捷径视频动作模型，能够通过单次前向传播预见连贯的几何与语义表征。这些预见表征作为稳定的蓝图，显著简化了动作预测过程。为实现这一高效捷径，我们引入了一种新颖的自蒸馏策略，将多步去噪的结构化生成先验压缩至一步推理中。具体而言，我们从扩散模型自身多步生成的视频中提取视觉基础模型（Vision Foundation Model, VFM）表征作为教师目标，而轻量级解耦器作为学生模型，学习直接将含噪声的一步特征映射至这些目标。在仿真与真实环境中的大量实验表明，我们的S-VAM优于现有先进方法，能够在复杂环境中实现高效且精确的操作。项目页面为 https://haodong-yan.github.io/S-VAM/。

Video action models (VAMs) have emerged as a promising paradigm for robot learning, owing to their powerful visual foresight for complex manipulation tasks. However, current VAMs, typically relying on either slow multi-step video generation or noisy one-step feature extraction, cannot simultaneously guarantee real-time inference and high-fidelity foresight. To address this limitation, we propose S-VAM, a shortcut video-action model that foresees coherent geometric and semantic representations via a single forward pass. Serving as a stable blueprint, these foreseen representations significantly simplify the action prediction. To enable this efficient shortcut, we introduce a novel self-distillation strategy that condenses structured generative priors of multi-step denoising into one-step inference. Specifically, vision foundation model (VFM) representations extracted from the diffusion model’s own multi-step generated videos provide teacher targets. Lightweight decouplers, as students, learn to directly map noisy one-step features to these targets. Extensive experiments in simulation and the real world demonstrate that our S-VAM outperforms state-of-the-art methods, enabling efficient and precise manipulation in complex environments. Our project page is https://haodong-yan.github.io/S-VAM/

评分理由: 论文《GIST: Gauge-Invariant Spectral Transformers for Scalable Graph Neural Operators》专注于图神经网络（GNNs）和神经算子领域，特别是针对图结构数据和网格的Transformer架构创新。其核心贡献是提出一种新的图Transformer（GIST），通过随机投影实现O(N)复杂度并保持规范不变性，以解决现有方法在计算效率和泛化性上的问题。论文应用在科学计算领域（如空气动力学预测），因此与关键词“AI for Science”有一定关联（5分）。然而，论文完全不涉及大语言模型（LLMs）、深度学习技术原理（如MoE、Scaling Laws、各种训练对齐技术、推理优化、智能体等）或生物信息学/化学信息学的具体应用，因此其他所有关键词均得0分。

将Transformer位置编码适配于网格和图结构数据面临显著的计算挑战：精确谱方法需要立方复杂度的特征分解，且可能因数值求解器伪影无意间破坏规范不变性；而高效的近似方法则从设计上牺牲了规范对称性。这两种失效模式都会导致归纳学习中的灾难性泛化问题——即使用一组数值选择训练的模型，在遇到相似图的不同谱分解或同一网格的不同离散化时完全失效。我们提出GIST（规范不变谱Transformer），一种新的图Transformer架构，通过随机投影实现端到端$\mathcal{O}(N)$复杂度，同时在算法层面利用投影嵌入的基于内积的注意力机制保持规范不变性，从而解决了这一难题。我们证明GIST能以有界失配误差实现离散化不变学习，使得神经算子应用中的参数能够跨任意网格分辨率迁移。实证研究表明，GIST在标准图基准测试中达到最先进水平（如在PPI数据集上实现99.50%的微平均F1分数），同时能独特地扩展到高达75万节点的基于网格的神经算子基准，在极具挑战性的DrivAerNet和DrivAerNet++数据集上实现了最先进的空气动力学预测性能。

Adapting transformer positional encoding to meshes and graph-structured data presents significant computational challenges: exact spectral methods require cubic-complexity eigendecomposition and can inadvertently break gauge invariance through numerical solver artifacts, while efficient approximate methods sacrifice gauge symmetry by design. Both failure modes cause catastrophic generalization in inductive learning, where models trained with one set of numerical choices fail when encountering different spectral decompositions of similar graphs or discretizations of the same mesh. We propose GIST (Gauge-Invariant Spectral Transformers), a new graph transformer architecture that resolves this challenge by achieving end-to-end $\mathcal{O}(N)$ complexity through random projections while algorithmically preserving gauge invariance via inner-product-based attention on the projected embeddings. We prove GIST achieves discretization-invariant learning with bounded mismatch error, enabling parameter transfer across arbitrary mesh resolutions for neural operator applications. Empirically, GIST matches state-of-the-art on standard graph benchmarks (e.g., achieving 99.50% micro-F1 on PPI) while uniquely scaling to mesh-based Neural Operator benchmarks with up to 750K nodes, achieving state-of-the-art aerodynamic prediction on the challenging DrivAerNet and DrivAerNet++ datasets.

评分理由: 论文ManiTwin专注于机器人操作模拟的数据生成，提出自动化管道从单图像生成3D数字资产并构建大规模数据集。所有评分关键词均涉及大模型/深度学习技术原理（如LLM、MoE、训练方法、推理优化等）或特定科学AI应用（如生物信息学），而本文核心是3D重建、数据集构建和机器人模拟，未涉及任何大模型技术或深度学习创新，也未在生物/化学等科学领域应用AI，因此所有关键词相关度为0。

仿真学习为扩展机器人操作能力提供了重要基础。然而，该范式通常在数据生成就绪的数字资产方面存在规模与多样性的不足。本研究提出ManiTwin，一个自动化且高效的流程，用于生成数据生成就绪的数字物体孪生体。我们的流程将单张图像转化为仿真就绪且带有语义标注的3D资产，从而支持大规模的机器人操作数据生成。利用该流程，我们构建了ManiTwin-100K数据集，包含10万个高质量标注的3D资产。每个资产均配备物理属性、语言描述、功能标注以及经过验证的操作建议。实验表明，ManiTwin提供了高效的资产合成与标注工作流，且ManiTwin-100K为操作数据生成、随机场景合成以及视觉问答（VQA）数据生成提供了高质量、多样化的资产，为可扩展的仿真数据合成与策略学习奠定了坚实基础。项目网页地址：https://manitwin.github.io/。

Learning in simulation provides a useful foundation for scaling robotic manipulation capabilities. However, this paradigm often suffers from a lack of data-generation-ready digital assets, in both scale and diversity. In this work, we present ManiTwin, an automated and efficient pipeline for generating data-generation-ready digital object twins. Our pipeline transforms a single image into simulation-ready and semantically annotated 3D asset, enabling large-scale robotic manipulation data generation. Using this pipeline, we construct ManiTwin-100K, a dataset containing 100K high-quality annotated 3D assets. Each asset is equipped with physical properties, language descriptions, functional annotations, and verified manipulation proposals. Experiments demonstrate that ManiTwin provides an efficient asset synthesis and annotation workflow, and that ManiTwin-100K offers high-quality and diverse assets for manipulation data generation, random scene synthesis, and VQA data generation, establishing a strong foundation for scalable simulation data synthesis and policy learning. Our webpage is available at https://manitwin.github.io/.

评分理由: 该论文专注于计算机视觉领域的3D场景重建，特别是单目图像中的物体级重建和接触关系建模，使用了现代神经网络架构（如SAM 3D）和大规模数据集，但未涉及任何大语言模型（LLM）、深度学习技术原理创新或AI for Science的具体应用，所有关键词均与大模型、深度学习技术原理或科学AI应用无关，因此所有关键词相关度均为0分。

单目三维场景重建近期取得显著进展。借助现代神经架构与大规模数据，现有方法在单图像深度估计方面已实现高性能表现。然而，由于物体种类繁多、遮挡频繁以及物体间关系复杂，将常见场景重建并分解为独立三维物体仍面临严峻挑战。值得注意的是，除单个物体的形状与姿态估计外，机器人学与动画等应用场景需要物理合理的三维重建，即物体必须遵循非穿透与真实接触的物理规律。本研究从两个方向推进物体级场景重建：首先，我们提出MessyKitchens数据集，该数据集包含真实世界杂乱环境场景，并以三维物体形状、姿态及精确物体接触关系的形式提供高保真物体级真值标注。其次，我们在近期单物体重建方法SAM 3D的基础上，引入多物体解码器（Multi-Object Decoder, MOD）以实现联合物体级场景重建。为验证贡献，我们证明MessyKitchens在配准精度与物体间穿透误差方面显著优于现有数据集。同时，我们在三个数据集上对比多物体重建方法，结果表明MOD相较现有最优方法取得持续且显著的改进。我们的新基准数据集、代码与预训练模型将在项目网站公开：https://messykitchens.github.io/。

Monocular 3D scene reconstruction has recently seen significant progress. Powered by the modern neural architectures and large-scale data, recent methods achieve high performance in depth estimation from a single image. Meanwhile, reconstructing and decomposing common scenes into individual 3D objects remains a hard challenge due to the large variety of objects, frequent occlusions and complex object relations. Notably, beyond shape and pose estimation of individual objects, applications in robotics and animation require physically-plausible scene reconstruction where objects obey physical principles of non-penetration and realistic contacts. In this work we advance object-level scene reconstruction along two directions. First, we introduceMessyKitchens, a new dataset with real-world scenes featuring cluttered environments and providing high-fidelity object-level ground truth in terms of 3D object shapes, poses and accurate object contacts. Second, we build on the recent SAM 3D approach for single-object reconstruction and extend it with Multi-Object Decoder (MOD) for joint object-level scene reconstruction. To validate our contributions, we demonstrate MessyKitchens to significantly improve previous datasets in registration accuracy and inter-object penetration. We also compare our multi-object reconstruction approach on three datasets and demonstrate consistent and significant improvements of MOD over the state of the art. Our new benchmark, code and pre-trained models will become publicly available on our project website: https://messykitchens.github.io/.

评分理由: 该论文研究视频生成模型中的推理机制，与大多数大语言模型（LLM）特定技术关键词无关。相关关键词包括：1）‘Chain of Thought OR CoT Reasoning OR Multi-step Reasoning’（8分）：论文挑战了Chain-of-Frames机制，提出了Chain-of-Steps（CoS）机制，涉及多步推理过程；2）‘System 2 Thinking OR Slow Thinking OR In-depth Reasoning’（7分）：论文分析了模型在去噪步骤中的深度推理行为；3）‘Self-Correction OR Self-Improvement OR Self-Reflection’（9分）：论文明确识别了自我纠正和增强作为关键推理行为；4）‘Mechanistic Interpretability OR Explainable AI’（8分）：论文通过定性分析和探测实验系统理解推理机制，属于可解释AI范畴。其他关键词与视频生成模型研究无直接关联。

近期视频生成领域的研究揭示了一个意外现象：基于扩散的视频模型展现出显著的推理能力。先前研究将此归因于帧间链式机制，即假设推理过程在视频帧间顺序展开。本研究挑战了这一假设，并揭示了一种根本不同的机制。我们证明视频模型中的推理主要沿着扩散去噪步骤产生。通过定性分析和针对性探测实验，我们发现模型在早期去噪步骤中探索多个候选解，并逐步收敛至最终答案，这一过程被我们称为步骤链式机制。除核心机制外，我们还识别出几种对模型性能至关重要的涌现推理行为：（1）工作记忆，支持持续性参照；（2）自我校正与增强，允许从错误中间解中恢复；（3）先感知后操作，即早期步骤建立语义基础，后期步骤执行结构化处理。在单个扩散步骤中，我们进一步发现扩散Transformer内部存在自演进的功能分化：早期层编码密集感知结构，中间层执行推理，后期层整合潜在表征。基于这些发现，我们提出一种简单的免训练策略作为概念验证，通过集成相同模型在不同随机种子下的潜在轨迹，展示了如何提升推理能力。总体而言，本研究系统阐释了视频生成模型中推理能力的涌现机制，为未来研究如何更好地利用视频模型固有的推理动态作为智能新载体奠定了理论基础。

Recent advances in video generation have revealed an unexpected phenomenon: diffusion-based video models exhibit non-trivial reasoning capabilities. Prior work attributes this to a Chain-of-Frames (CoF) mechanism, where reasoning is assumed to unfold sequentially across video frames. In this work, we challenge this assumption and uncover a fundamentally different mechanism. We show that reasoning in video models instead primarily emerges along the diffusion denoising steps. Through qualitative analysis and targeted probing experiments, we find that models explore multiple candidate solutions in early denoising steps and progressively converge to a final answer, a process we term Chain-of-Steps (CoS). Beyond this core mechanism, we identify several emergent reasoning behaviors critical to model performance: (1) working memory, enabling persistent reference; (2) self-correction and enhancement, allowing recovery from incorrect intermediate solutions; and (3) perception before action, where early steps establish semantic grounding and later steps perform structured manipulation. During a diffusion step, we further uncover self-evolved functional specialization within Diffusion Transformers, where early layers encode dense perceptual structure, middle layers execute reasoning, and later layers consolidate latent representations. Motivated by these insights, we present a simple training-free strategy as a proof-of-concept, demonstrating how reasoning can be improved by ensembling latent trajectories from identical models with different random seeds. Overall, our work provides a systematic understanding of how reasoning emerges in video generation models, offering a foundation to guide future research in better exploiting the inherent reasoning dynamics of video models as a new substrate for intelligence.

评分理由: 论文研究Omni-modal large language models (OLMs)的社交交互能力评估，与’Large Language Models OR LLMs OR Foundation Models’高度相关（10分），因为OLMs是LLMs的多模态扩展。论文聚焦于评估基准创建和模型能力分析，不涉及其他关键词的具体技术细节（如MoE、SFT、RAG等）或特定应用领域（如生物信息学），因此其他关键词评分为0。

全模态大语言模型通过原生整合音频、视觉与文本，重新定义了人机交互。然而，现有全模态大语言模型评测基准仍局限于静态的、以准确性为中心的任务，在评估社交互动性这一自然对话中处理动态线索的核心能力方面存在关键空白。为此，我们提出SocialOmni——一个综合性评测基准，从三个核心维度对对话互动性进行可操作化评估：（一）说话人分离与识别（谁在说话），（二）打断时机控制（何时插话），（三）自然打断生成（如何组织打断语句）。SocialOmni包含2000个感知样本，以及一个经过质量控制的诊断集（含209个具有严格时空与上下文约束的交互生成实例），并辅以受控的视听不一致场景以测试模型鲁棒性。我们对12个领先的全模态大语言模型进行了基准测试，结果揭示了不同模型在社交互动能力上存在显著差异。进一步分析表明，模型的感知准确性与其生成上下文适宜打断语句的能力之间存在明显脱钩，这说明仅依靠以理解为中心的度量指标不足以刻画对话社交能力。更令人鼓舞的是，SocialOmni提供的诊断结果为未来全模态大语言模型弥合感知与交互之间的鸿沟产生了可操作的改进信号。

Omni-modal large language models (OLMs) redefine human-machine interaction by natively integrating audio, vision, and text. However, existing OLM benchmarks remain anchored to static, accuracy-centric tasks, leaving a critical gap in assessing social interactivity, the fundamental capacity to navigate dynamic cues in natural dialogues. To this end, we propose SocialOmni, a comprehensive benchmark that operationalizes the evaluation of this conversational interactivity across three core dimensions: (i) speaker separation and identification (who is speaking), (ii) interruption timing control (when to interject), and (iii) natural interruption generation (how to phrase the interruption). SocialOmni features 2,000 perception samples and a quality-controlled diagnostic set of 209 interaction-generation instances with strict temporal and contextual constraints, complemented by controlled audio-visual inconsistency scenarios to test model robustness. We benchmarked 12 leading OLMs, which uncovers significant variance in their social-interaction capabilities across models. Furthermore, our analysis reveals a pronounced decoupling between a model’s perceptual accuracy and its ability to generate contextually appropriate interruptions, indicating that understanding-centric metrics alone are insufficient to characterize conversational social competence. More encouragingly, these diagnostics from SocialOmni yield actionable signals for bridging the perception-interaction divide in future OLMs.

评分理由: 论文专注于视频超分辨率（VSR）的交互式框架SparkVSR，通过稀疏关键帧传播实现可控的视频增强。研究内容属于计算机视觉和视频处理领域，核心是视频恢复技术、关键帧条件化训练和交互式控制机制。所有评分关键词均围绕大模型、深度学习技术原理及其在科学领域的应用，但本文未涉及任何大模型（LLM/SLM）、MoE、缩放定律、预训练/后训练、对齐、RLHF、PEFT、RAG、注意力优化、推理技术、智能体、量化、幻觉缓解、可解释性、世界模型、模型合并、上下文学习或AI for Science等主题。论文的创新在于视频处理框架设计，而非大模型技术或其在科学领域的应用，因此所有关键词相关度均为0分。

视频超分辨率（Video Super-Resolution，VSR）旨在从低分辨率（Low-Resolution，LR）估计中恢复高质量的视频帧，然而现有的大多数VSR方法在推理时如同黑盒：用户无法可靠地修正意外的伪影，而只能接受模型生成的任何结果。本文提出了一种新颖的交互式VSR框架，称为SparkVSR，它将稀疏关键帧作为一种简洁而富有表现力的控制信号。具体而言，用户可以首先使用任何现成的图像超分辨率（Image Super-Resolution，ISR）模型对少量关键帧进行超分辨率处理（或选择性处理），随后SparkVSR将关键帧先验传播至整个视频序列，同时保持以原始低分辨率视频运动为基础。我们设计了一种关键帧条件化的潜在-像素两阶段训练流程，该流程将低分辨率视频潜在特征与稀疏编码的高分辨率关键帧潜在特征相融合，以学习鲁棒的跨空间传播并优化感知细节。在推理阶段，SparkVSR支持灵活的关键帧选择方式（手动指定、编解码器I帧提取或随机采样），并采用一种无参考的引导机制，持续平衡对关键帧的遵循与盲恢复，确保即使在参考关键帧缺失或不完美时也能实现鲁棒性能。在多个VSR基准测试上的实验表明，该方法提升了时间一致性并实现了强大的恢复质量，在CLIP-IQA、DOVER和MUSIQ指标上分别超越基线高达24.6%、21.8%和5.6%，实现了可控的、关键帧驱动的视频超分辨率。此外，我们证明SparkVSR是一种通用的交互式、关键帧条件化视频处理框架，因为它可以直接应用于未见过的任务，如老电影修复和视频风格迁移。我们的项目页面位于：https://sparkvsr.github.io/

Video Super-Resolution (VSR) aims to restore high-quality video frames from low-resolution (LR) estimates, yet most existing VSR approaches behave like black boxes at inference time: users cannot reliably correct unexpected artifacts, but instead can only accept whatever the model produces. In this paper, we propose a novel interactive VSR framework dubbed SparkVSR that makes sparse keyframes a simple and expressive control signal. Specifically, users can first super-resolve or optionally a small set of keyframes using any off-the-shelf image super-resolution (ISR) model, then SparkVSR propagates the keyframe priors to the entire video sequence while remaining grounded by the original LR video motion. Concretely, we introduce a keyframe-conditioned latent-pixel two-stage training pipeline that fuses LR video latents with sparsely encoded HR keyframe latents to learn robust cross-space propagation and refine perceptual details. At inference time, SparkVSR supports flexible keyframe selection (manual specification, codec I-frame extraction, or random sampling) and a reference-free guidance mechanism that continuously balances keyframe adherence and blind restoration, ensuring robust performance even when reference keyframes are absent or imperfect. Experiments on multiple VSR benchmarks demonstrate improved temporal consistency and strong restoration quality, surpassing baselines by up to 24.6%, 21.8%, and 5.6% on CLIP-IQA, DOVER, and MUSIQ, respectively, enabling controllable, keyframe-driven video super-resolution. Moreover, we demonstrate that SparkVSR is a generic interactive, keyframe-conditioned video processing framework as it can be applied out of the box to unseen tasks such as old-film restoration and video style transfer. Our project page is available at: https://sparkvsr.github.io/

评分理由: 论文《SOMA: Unifying Parametric Human Body Models》专注于计算机图形学中的人体建模技术，提出了一种统一不同参数化人体模型（如SMPL、SMPL-X等）的框架。所有评分关键词均与大模型、深度学习技术原理或AI在科学领域的应用直接相关，而本文研究的是特定领域的3D人体模型统一方法，不涉及大模型、深度学习技术或AI在生物医药等科学领域的应用，因此与所有关键词完全无关，均得0分。

参数化人体模型是人体重建、动画与仿真的基础，然而现有模型之间仍存在互不兼容的问题：SMPL、SMPL-X、MHR、Anny 及相关模型在网格拓扑、骨骼结构、形状参数化与单位约定上各有差异，导致难以在单一流程中综合利用它们的互补优势。本文提出 SOMA，一种通过三层抽象机制桥接这些异构表示的统一人体层。网格拓扑抽象层可将任何源模型的身份特征以每顶点恒定时间映射至共享规范网格；骨骼抽象层通过单次闭式计算（无需迭代优化或针对各模型的专门训练），从任意身体形状（无论是静止姿态还是任意姿态配置）恢复出一整套适应身份信息的关节变换；姿态抽象层通过逆向蒙皮流程，直接从任何支持模型的姿态顶点中恢复出统一的骨骼旋转，使得异构运动数据集无需定制重定向即可使用。这些抽象层共同将原本 $O(M^2)$ 的逐对适配问题简化为 $O(M)$ 的单后端连接器，让使用者能够在推理阶段自由混合不同身份来源与姿态数据。整个流程通过 NVIDIA-Warp 实现完全端到端可微分且 GPU 加速。

Parametric human body models are foundational to human reconstruction, animation, and simulation, yet they remain mutually incompatible: SMPL, SMPL-X, MHR, Anny, and related models each diverge in mesh topology, skeletal structure, shape parameterization, and unit convention, making it impractical to exploit their complementary strengths within a single pipeline. We present SOMA, a unified body layer that bridges these heterogeneous representations through three abstraction layers. Mesh topology abstraction maps any source model’s identity to a shared canonical mesh in constant time per vertex. Skeletal abstraction recovers a full set of identity-adapted joint transforms from any body shape, whether in rest pose or an arbitrary posed configuration, in a single closed-form pass, with no iterative optimization or per-model training. Pose abstraction inverts the skinning pipeline to recover unified skeleton rotations directly from posed vertices of any supported model, enabling heterogeneous motion datasets to be consumed without custom retargeting. Together, these layers reduce the $O(M^2)$ per-pair adapter problem to $O(M)$ single-backend connectors, letting practitioners freely mix identity sources and pose data at inference time. The entire pipeline is fully differentiable end-to-end and GPU-accelerated via NVIDIA-Warp.