评分理由: 论文提出ESPIRE基准，专注于评估视觉语言模型（VLMs）在具身空间推理任务中的表现。核心相关关键词包括：‘LLM Agents’（10分，论文直接涉及具身智能体在模拟世界中的任务执行）、‘World Models’（8分，论文创建模拟世界来物理化VLMs）、‘Chain of Thought’和’System 2 Thinking’（各8分，论文强调空间推理的分解和推理行动过程）。其他关键词如’Large Language Models’（8分，VLMs是大模型的一种）、‘Pre-training’、‘SFT’、‘Instruction Tuning’、‘Explainable AI’、‘In-context Learning’（各5分，论文涉及模型适应和诊断分析，但非核心）。其余关键词与论文内容无关，得0分。

视觉语言模型（VLMs）近期的一个趋势是增强其在具身领域中的空间认知能力。尽管已有进展，现有评估方法在范式和覆盖范围上均存在局限，阻碍了模型的快速迭代开发。为应对这些不足，我们提出了ESPIRE——一个用于具身空间推理的诊断性基准。ESPIRE提供了一个模拟世界，将VLMs置于物理环境中，并以空间推理为核心的机器人任务对其进行评估，从而缩小了评估与实际部署之间的差距。为使VLMs适应机器人任务，我们将每项任务分解为定位与执行两个阶段，并将二者均构建为生成式问题，这与当前主流的基于干扰项且忽略执行的判别式评估（例如通过视觉问答）形成鲜明对比。这种分解进一步支持了从被动空间推理到行动推理的细粒度分析。我们在指令层面和环境层面系统化地设计了ESPIRE，确保其广泛覆盖各类空间推理场景。我们利用ESPIRE对一系列前沿VLM进行诊断，并深入分析了它们的空间推理行为。

A recent trend in vision-language models (VLMs) has been to enhance their spatial cognition for embodied domains. Despite progress, existing evaluations have been limited both in paradigm and in coverage, hindering rapid, iterative model development. To address these limitations, we propose ESPIRE, a diagnostic benchmark for embodied spatial reasoning. ESPIRE offers a simulated world that physically grounds VLMs and evaluates them on spatial-reasoning-centric robotic tasks, thus narrowing the gap between evaluation and real-world deployment. To adapt VLMs to robotic tasks, we decompose each task into localization and execution, and frame both as generative problems, in stark contrast to predominant discriminative evaluations (e.g., via visual-question answering) that rely on distractors and discard execution. This decomposition further enables a fine-grained analysis beyond passive spatial reasoning toward reasoning to act. We systematically design ESPIRE both at the instruction level and at the environment level, ensuring broad coverage of spatial reasoning scenarios. We use ESPIRE to diagnose a range of frontier VLMs and provide in-depth analysis of their spatial reasoning behaviors.

评分理由: 论文NeuroLoRA的核心贡献是提出了一种基于混合专家（MoE）的LoRA框架，用于大语言模型（LLMs）的参数高效微调（PEFT）。因此，与"Large Language Models (LLMs)"、“Mixture of Experts (MoE)“和"PEFT/LoRA"高度相关（10分）。论文明确涉及多任务适应和顺序持续学习，与"Domain Adaptation"和"Model Merging"有一定关联（5分）。论文未直接讨论其他关键词，如SLMs、Scaling Laws、RAG、推理加速或特定科学AI应用，因此这些关键词得0分。

参数高效微调技术，特别是低秩自适应方法，已成为将大语言模型适配至下游任务的关键工具。尽管近期提出的FlyLoRA框架成功利用仿生稀疏随机投影来缓解参数干扰，但其依赖静态的、基于幅度的路由机制，无法感知输入上下文。本文受生物神经调节机制——即根据上下文动态调控神经元兴奋性的启发，提出一种基于混合专家模型的新型LoRA框架：NeuroLoRA。该框架在保持冻结随机投影计算效率的同时，引入了一个轻量级、可学习的神经调节门，能在专家选择前根据上下文对投影空间进行动态重缩放。我们进一步提出对比正交性损失，以显式增强专家子空间之间的分离性，从而提升任务解耦与持续学习能力。在MMLU、GSM8K和ScienceQA基准上的大量实验表明，NeuroLoRA在单任务适配、多任务模型融合及序列持续学习场景中，均持续优于FlyLoRA及其他强基线方法，同时保持了相当的参数效率。

Parameter-Efficient Fine-Tuning (PEFT) techniques, particularly Low-Rank Adaptation (LoRA), have become essential for adapting Large Language Models (LLMs) to downstream tasks. While the recent FlyLoRA framework successfully leverages bio-inspired sparse random projections to mitigate parameter interference, it relies on a static, magnitude-based routing mechanism that is agnostic to input context. In this paper, we propose NeuroLoRA, a novel Mixture-of-Experts (MoE) based LoRA framework inspired by biological neuromodulation – the dynamic regulation of neuronal excitability based on context. NeuroLoRA retains the computational efficiency of frozen random projections while introducing a lightweight, learnable neuromodulation gate that contextually rescales the projection space prior to expert selection. We further propose a Contrastive Orthogonality Loss to explicitly enforce separation between expert subspaces, enhancing both task decoupling and continual learning capacity. Extensive experiments on MMLU, GSM8K, and ScienceQA demonstrate that NeuroLoRA consistently outperforms FlyLoRA and other strong baselines across single-task adaptation, multi-task model merging, and sequential continual learning scenarios, while maintaining comparable parameter efficiency.

评分理由: 论文核心研究奖励模型（RMs）的去偏方法，属于大模型对齐（Alignment）和RLHF技术范畴。与"Instruction Tuning/Alignment/Value Alignment"和"RLHF/RLAIF/DPO"高度相关（10分），因为奖励模型是RLHF对齐流程的关键组件。与"Large Language Models/LLMs/Foundation Models"相关（8分），因为奖励模型通常基于大语言模型构建。与"Mechanistic Interpretability/Explainable AI"相关（8分），因为使用稀疏自编码器（SAE）进行特征解释和干预。与"Hallucination Mitigation/Factuality/Truthfulness"有一定关联（5分），因为去偏有助于提升模型对语义而非表面风格的偏好，间接提升事实性。其他关键词如MoE、SLMs、Scaling Laws、Pre-training、RAG、Context Window等与论文内容无直接关系，得0分。

奖励模型（Reward Models, RMs）是对齐流程中的关键组件，但它们表现出对表层风格线索的偏好偏差，倾向于选择呈现形式更佳而非语义更优的回应。现有的去偏方法通常需要重新训练或调整模型架构，而直接抑制激活会因表征纠缠导致性能下降。我们提出了SteerRM，这是首个基于稀疏自编码器（Sparse Autoencoder, SAE）干预、无需训练即可实现奖励模型去偏的方法。SteerRM利用对比配对回应分离风格效应，通过强度-稳定性准则识别与偏差相关的SAE特征，并在推理阶段抑制这些特征。在RM-Bench的六个奖励模型上，SteerRM平均将Hard-split准确率提升了7.3个百分点，同时保持了整体性能。基于Gemma的奖励模型实验及对非格式偏差的受控研究进一步表明，该方法可泛化至不同RM架构与偏差类型。我们还发现，与格式相关的特征集中分布于模型浅层，且在不同模型间可迁移，这揭示了架构层面共享的偏差编码模式。这些结果表明，基于SAE的干预能够在无需重新训练的情况下缓解奖励模型偏差，为对齐流程提供了一种实用且可解释的解决方案。

Reward models (RMs) are critical components of alignment pipelines, yet they exhibit biases toward superficial stylistic cues, preferring better-presented responses over semantically superior ones. Existing debiasing methods typically require retraining or architectural modifications, while direct activation suppression degrades performance due to representation entanglement. We propose SteerRM, the first training-free method for debiasing reward models using Sparse Autoencoder (SAE)-based interventions. SteerRM isolates stylistic effects using contrastive paired responses, identifies bias-related SAE features with a strength-stability criterion, and suppresses them at inference time. Across six reward models on RM-Bench, SteerRM improves Hard-split accuracy by 7.3 points on average while preserving overall performance. Results on a Gemma-based reward model and a controlled non-format bias further suggest generalization across RM architectures and bias types. We further find that format-related features are concentrated in shallow layers and transfer across models, revealing shared architecture-level bias encoding patterns. These results show that SAE-based interventions can mitigate reward-model biases without retraining, providing a practical and interpretable solution for alignment pipelines.

!!! info

论文采用三阶段技术路线：首先，合成格式控制的数据对，即生成语义内容相同但表面格式不同（如Markdown与纯文本）的响应对；其次，利用预训练的SAE分解奖励模型的隐藏状态，通过计算配对差异，结合强度和稳定性标准筛选出与格式相关的SAE特征；最后，在推理阶段通过干预机制抑制这些特定特征的激活，从而在不更新模型参数的情况下消除奖励模型的格式偏见。

• 免训练高效性：无需重新训练模型或修改架构，降低了计算成本和实施复杂度。
• 可解释性强：利用SAE将内部表示分解为可解释的特征，明确了偏见的来源。
• 性能保持：在去偏的同时，有效保留了模型对语义内容的评估能力，避免了性能退化。
• 泛化能力：不仅适用于特定模型（如LLaMA），还推广到Gemma及其他类型的风格偏见。

• 依赖预训练SAE：方法的有效性依赖于预训练的SAE字典的质量和可用性，如果SAE未能完美重建特征，去偏效果可能受限。
• 特征识别的准确性：基于强度-稳定性标准的特征筛选可能无法完全捕捉所有复杂的偏见特征，存在误判风险。
• 推理开销：虽然免训练，但在推理过程中需要运行SAE编码和解码，可能会增加一定的延迟和计算开销。
• 特定偏见类型：主要针对格式偏见，虽然声称可推广，但对其他深层语义偏见的处理能力尚需进一步验证。

评分理由: 论文研究大型视觉语言模型（LVLMs）在微调时易受后门攻击的问题，并提出了一种无需重新训练、仅需在测试时操作的防御方法CleanSight。该论文与以下关键词高度相关：1）“Large Language Models” OR “LLMs” OR “Foundation Models”（8分）：论文研究LVLMs，属于大模型范畴，但更侧重于视觉语言模型而非纯语言模型。2）“Post-training” OR “Supervised Fine-tuning” OR “SFT”（8分）：论文明确提到后门攻击发生在微调（fine-tuning）阶段，这是后训练的关键环节。3）“PEFT” OR “LoRA” OR “Parameter-efficient Fine-tuning”（10分）：论文直接提到LoRA模块作为后门攻击的可能目标，并指出现有防御方法涉及重新训练这些参数，这是论文的核心技术背景之一。4）“Mechanistic Interpretability” OR “Explainable AI”（10分）：论文的核心贡献是提供了对LVLMs中后门行为的机制性理解（mechanistic understanding），即通过跨模态注意力重分布（attention stealing）来解释后门激活机制，这直接属于机制可解释性研究。其他关键词与论文内容无关，因为论文专注于视觉语言模型的后门防御、注意力机制分析和测试时净化，不涉及MoE、小模型、缩放律、预训练、对齐、RLHF、RAG、上下文扩展、推理加速、智能体、量化等主题。

尽管大型视觉语言模型（LVLMs）具备强大的多模态性能，其在微调过程中仍易受到后门攻击的威胁——攻击者将嵌入触发器的样本注入训练数据，以植入可在测试阶段被恶意激活的行为。现有防御方法通常依赖于使用干净数据重新训练被植入后门的参数（例如适配器或LoRA模块），这种方式计算成本高昂且常导致模型性能下降。本研究对LVLMs中的后门行为提出了新的机制性解释：触发器并非通过底层视觉模式影响预测，而是通过异常的跨模态注意力重分配发挥作用——携带触发器的视觉令牌会从文本上下文中窃取注意力，我们将此现象称为注意力窃取。基于此发现，我们提出了CleanSight：一种无需训练、即插即用的纯测试阶段防御方案。CleanSight（i）通过选定跨模态融合层中的视觉-文本注意力相对比例来检测中毒输入，并（ii）通过选择性剪枝可疑的高注意力视觉令牌以净化输入，从而中和后门激活。大量实验表明，CleanSight在多种数据集和攻击类型下均显著优于现有的基于像素的净化防御方法，同时在干净样本与中毒样本上均能保持模型的原有效能。

Despite the strong multimodal performance, large vision-language models (LVLMs) are vulnerable during fine-tuning to backdoor attacks, where adversaries insert trigger-embedded samples into the training data to implant behaviors that can be maliciously activated at test time. Existing defenses typically rely on retraining backdoored parameters (e.g., adapters or LoRA modules) with clean data, which is computationally expensive and often degrades model performance. In this work, we provide a new mechanistic understanding of backdoor behaviors in LVLMs: the trigger does not influence prediction through low-level visual patterns, but through abnormal cross-modal attention redistribution, where trigger-bearing visual tokens steal attention away from the textual context - a phenomenon we term attention stealing. Motivated by this, we propose CleanSight, a training-free, plug-and-play defense that operates purely at test time. CleanSight (i) detects poisoned inputs based on the relative visual-text attention ratio in selected cross-modal fusion layers, and (ii) purifies the input by selectively pruning the suspicious high-attention visual tokens to neutralize the backdoor activation. Extensive experiments show that CleanSight significantly outperforms existing pixel-based purification defenses across diverse datasets and backdoor attack types, while preserving the model’s utility on both clean and poisoned samples.

评分理由: 论文是大型语言模型（LLMs）持续学习的综述，核心围绕LLMs的持续学习方法和挑战。因此，与"Large Language Models"和"Continual Pre-training"高度相关（10分）。论文提到持续微调和持续对齐，与"Supervised Fine-tuning"和"Alignment"有一定关联（5分）。论文讨论参数效率，与"Parameter-efficient Fine-tuning"相关（5分）。其他关键词如MoE、SLMs、RAG、推理加速、AI for Science等未在摘要中提及，因此评分为0分。

持续学习（Continual Learning, CL）已成为关键范式，旨在使大语言模型（Large Language Models, LLMs）能够动态适应不断演进的知识与序列任务，同时缓解灾难性遗忘——这一现代大语言模型固有的静态预训练范式的核心局限。本综述全面梳理了针对大语言模型设计的持续学习方法，围绕三个核心训练阶段展开：持续预训练、持续微调与持续对齐。在经典的基于复演、正则化与架构的方法分类之外，我们进一步依据其独特的遗忘缓解机制对每类方法进行细分，并对传统持续学习方法在大语言模型中的适应性与关键改进进行了严谨的比较分析。在此过程中，我们明确强调了大语言模型持续学习与传统机器学习之间的核心区别，特别是在模型规模、参数效率与涌现能力方面。我们的分析涵盖了核心评估指标，包括遗忘率与知识迁移效率，以及评估持续学习性能的新兴基准。本综述揭示，尽管现有方法在特定领域展现出有前景的结果，但在实现跨多样任务与时间尺度的无缝知识整合方面，仍存在根本性挑战。本系统性综述为不断增长的大语言模型适应研究提供了贡献，为研究者与实践者提供了一个结构化框架，以理解语言模型终身学习领域的当前成就与未来机遇。

Continual learning (CL) has emerged as a pivotal paradigm to enable large language models (LLMs) to dynamically adapt to evolving knowledge and sequential tasks while mitigating catastrophic forgetting-a critical limitation of the static pre-training paradigm inherent to modern LLMs. This survey presents a comprehensive overview of CL methodologies tailored for LLMs, structured around three core training stages: continual pre-training, continual fine-tuning, and continual alignment.Beyond the canonical taxonomy of rehearsal-, regularization-, and architecture-based methods, we further subdivide each category by its distinct forgetting mitigation mechanisms and conduct a rigorous comparative analysis of the adaptability and critical improvements of traditional CL methods for LLMs. In doing so, we explicitly highlight core distinctions between LLM CL and traditional machine learning, particularly with respect to scale, parameter efficiency, and emergent capabilities. Our analysis covers essential evaluation metrics, including forgetting rates and knowledge transfer efficiency, along with emerging benchmarks for assessing CL performance. This survey reveals that while current methods demonstrate promising results in specific domains, fundamental challenges persist in achieving seamless knowledge integration across diverse tasks and temporal scales. This systematic review contributes to the growing body of knowledge on LLM adaptation, providing researchers and practitioners with a structured framework for understanding current achievements and future opportunities in lifelong learning for language models.

评分理由: 论文聚焦于视觉-语言模型（VLMs）在拓扑异常检测中的应用，属于大模型在科学领域（医学图像分析）的应用研究。核心相关性体现在：1）使用VLMs（属于大模型范畴），得5分；2）明确采用监督微调（SFT）和强化学习（GRPO，属于RLHF/RLAIF范畴）的两阶段训练方法，这两项得10分；3）应用领域为生物医学图像分析（血管、神经纤维），属于AI for Science/Bioinformatics，得10分。其他关键词如MoE、量化、推理加速等与论文技术内容无关，得0分。

拓扑正确性对于血管、神经纤维及道路网络等管状结构至关重要。现有拓扑保持方法依赖于特定领域的标注真值，其获取成本高昂且难以跨领域迁移。当部署至缺乏标注的新领域时，一个关键问题随之产生：如何在无真值监督的情况下检测拓扑异常？我们将此问题重构为拓扑异常检测——一项结构化视觉推理任务，要求模型在预测的分割掩码中定位并分类拓扑错误。视觉-语言模型（VLMs）是天然的候选方案；然而，我们发现当前最先进的视觉-语言模型表现近乎随机，缺乏识别密集结构中稀疏连通性错误所需的细粒度拓扑感知能力。为弥补这一差距，我们开发了一套自动化数据构建流程，该流程能合成具有可验证标注的多样化拓扑异常，并设置渐进难度等级，从而构建了首个面向该任务的大规模、多领域基准数据集。随后，我们提出Topo-R1框架，通过两阶段训练赋予视觉-语言模型拓扑感知能力：首先进行监督微调，随后采用基于群组相对策略优化（GRPO）的强化学习。我们方法的核心在于设计了一种拓扑感知复合奖励机制，该机制整合了面向类型的匈牙利匹配算法以实现结构化错误分类、空间定位评分，以及直接惩罚连通性中断的中心线戴斯系数（clDice）奖励，从而共同激励语义精确性与结构保真度。大量实验表明，Topo-R1为无标注拓扑质量评估建立了新范式，在所有评估协议中均持续优于通用视觉-语言模型及有监督基线方法。

Topological correctness is crucial for tubular structures such as blood vessels, nerve fibers, and road networks. Existing topology-preserving methods rely on domain-specific ground truth, which is costly and rarely transfers across domains. When deployed to a new domain without annotations, a key question arises: how can we detect topological anomalies without ground-truth supervision? We reframe this as topological anomaly detection, a structured visual reasoning task requiring a model to locate and classify topological errors in predicted segmentation masks. Vision-Language Models (VLMs) are natural candidates; however, we find that state-of-the-art VLMs perform nearly at random, lacking the fine-grained, topology-aware perception needed to identify sparse connectivity errors in dense structures. To bridge this gap, we develop an automated data-curation pipeline that synthesizes diverse topological anomalies with verifiable annotations across progressively difficult levels, thereby constructing the first large-scale, multi-domain benchmark for this task. We then introduce Topo-R1, a framework that endows VLMs with topology-aware perception via two-stage training: supervised fine-tuning followed by reinforcement learning with Group Relative Policy Optimization (GRPO). Central to our approach is a topology-aware composite reward that integrates type-aware Hungarian matching for structured error classification, spatial localization scoring, and a centerline Dice (clDice) reward that directly penalizes connectivity disruptions, thereby jointly incentivizing semantic precision and structural fidelity. Extensive experiments demonstrate that Topo-R1 establishes a new paradigm for annotation-free topological quality assessment, consistently outperforming general-purpose VLMs and supervised baselines across all evaluation protocols.

评分理由: 论文研究视觉文档检索（VDR）中的模型蒸馏技术，将2B参数的视觉语言模型（VLM）蒸馏为69M参数的纯文本编码器。核心相关关键词：1）“Small Language Models” (8分)：论文明确开发69M参数的轻量级模型，属于小型语言模型范畴；2）“Retrieval-Augmented Generation” (8分)：论文专注于检索任务，是RAG系统的关键组件；3）“Quantization” (8分)：通过蒸馏实现模型压缩，大幅减少参数和延迟；4）“Large Language Models” (5分)：使用2B VLM作为教师模型，属于大模型范畴；5）“Speculative Decoding” (5分)：通过轻量化实现推理加速，与加速目标相关。其他关键词如MoE、对齐、推理方法等与论文内容无直接关联。

基于视觉语言模型（VLM）的检索器已将视觉文档检索（VDR）的质量提升至令人瞩目的水平。这类方法在文档索引和查询编码时均需使用相同的数十亿参数编码器，导致即使处理纯文本查询也存在高延迟和强GPU依赖的问题。我们观察到，这种设计存在不必要的对称性：文档在视觉上复杂，需要强大的视觉理解能力，而查询仅为短文本字符串。NanoVDR通过解耦两条编码路径来利用这种查询-文档不对称性：一个冻结的20亿参数VLM教师模型离线处理文档索引，而一个经蒸馏的、小至6900万参数的纯文本学生模型在推理时编码查询。其核心设计在于蒸馏目标的选择。通过对三种骨干网络和22个ViDoRe基准数据集上的六种目标进行系统比较，我们发现，在查询文本上进行逐点余弦对齐的方法持续优于基于排序和对比学习的替代方案，且训练时仅需预缓存的教师查询嵌入，无需处理文档。此外，我们指出跨语言迁移是主要性能瓶颈，并通过使用机器翻译的查询数据增强训练集，以低成本解决了该问题。最终得到的NanoVDR-S-Multi（基于DistilBERT，6900万参数）保留了教师模型95.1%的性能，在v2和v3版本上超越了DSE-Qwen2（20亿参数），同时参数量减少了32倍，CPU查询延迟降低了50倍，总训练成本低于13 GPU小时。

Vision-Language Model (VLM) based retrievers have advanced visual document retrieval (VDR) to impressive quality. They require the same multi-billion parameter encoder for both document indexing and query encoding, incurring high latency and GPU dependence even for plain-text queries. We observe that this design is unnecessarily symmetric: documents are visually complex and demand strong visual understanding, whereas queries are just short text strings. NanoVDR exploits this query–document asymmetry by decoupling the two encoding paths: a frozen 2B VLM teacher indexes documents offline, while a distilled text-only student as small as 69M parameters encodes queries at inference. The key design choice is the distillation objective. Through systematic comparison of six objectives across three backbones and 22 ViDoRe benchmark datasets, we find that pointwise cosine alignment on query text consistently outperforms ranking-based and contrastive alternatives, while requiring only pre-cached teacher query embeddings and no document processing during training. Furthermore, we identify cross-lingual transfer as the primary performance bottleneck, and resolve it cheaply by augmenting training data with machine-translated queries. The resulting NanoVDR-S-Multi (DistilBERT, 69M) retains 95.1% of teacher quality and outperforms DSE-Qwen2 (2B) on v2 and v3 with 32$\times$ fewer parameters and 50$\times$ lower CPU query latency, at a total training cost under 13 GPU-hours.

评分理由: 论文提出AVION框架，用于遥感图像视觉语言模型的适应，核心涉及大语言模型（LLMs）生成文本描述（相关度8），属于视觉语言模型的领域适应（相关度8），采用轻量级可学习提示进行参数高效微调（相关度8），并应用于遥感科学（AI for Science相关度8）。其他关键词如MoE、SLMs、SFT、RAG等未在摘要中体现，故评0分。

将视觉语言模型适配于遥感影像仍面临两大挑战：文本表征的语义覆盖范围有限，以及视觉特征适应性不足。这些问题在涉及多样化视觉外观与细粒度目标区分的航空场景中尤为显著。我们提出AVION——一个专为视觉语言模型遥感适配设计的知识蒸馏框架。教师模块通过收集大语言模型的描述文本并利用遥感影像特征验证有效性，构建语义丰富的文本原型。学生模块则在视觉与语言编码器中分别集成轻量级可学习提示向量，并在教师模块的指导下对齐嵌入表示及其跨模态关联。训练完成后，学生模块可在推理阶段独立运行。在六个光学遥感基准数据集上的实验表明，AVION在提升小样本分类与基类识别准确率的同时，未削弱对新类别的泛化能力。该框架还显著提高了跨模态检索的平均召回率，且仅需引入极少量可训练参数。

Adapting vision-language models to remote sensing imagery remains challenging due to two key factors: limited semantic coverage in textual representations and insufficient adaptability of visual features. These issues are particularly significant in aerial scenes, which involve various visual appearances and fine-grained object distinctions. We propose AVION, a knowledge distillation framework tailored for remote sensing adaptation of vision-language models. The teacher module constructs semantically rich textual prototypes by collecting descriptions from a large language model and verifying validity using remote sensing image features. The student module integrates lightweight and learnable prompts into both vision and language encoders, guided by the teacher to align embeddings and their cross-modal relationships. Once trained, the student operates independently during inference. Experiments on six optical remote sensing benchmarks show that AVION improves few-shot classification and base-class accuracy without degrading generalization to novel categories. It also enhances mean recall for cross-modal retrieval, with minimal additional trainable parameters.

评分理由: 论文核心研究Parameter-Efficient Fine-Tuning (PEFT)和Mixture-of-Experts (MoE)技术，提出Expert Pyramid Tuning (EPT)方法，直接与关键词"PEFT/LoRA"和"MoE"高度相关（10分）。论文明确针对LLMs部署，与"Large Language Models"高度相关（10分）。其他关键词如SLMs、Scaling Laws、RAG、RLHF等未在摘要中提及或相关，均给0分。

参数高效微调（Parameter-Efficient Fine-Tuning, PEFT）因其极高的参数效率，已成为在多任务场景中部署大语言模型（LLMs）的主导范式。尽管基于专家混合（Mixture-of-Experts, MoE）的LoRA变体通过将令牌动态路由至不同的低秩专家取得了有前景的结果，但它们大多忽视了任务复杂性的层级本质。现有方法通常采用架构统一的专家，限制了其捕捉不同任务所需多样化特征粒度的能力——有些任务需要高层语义抽象，而另一些则需要细粒度的句法操控。为弥补这一差距，我们提出了专家金字塔调优（Expert Pyramid Tuning, EPT），这是一种新颖的架构，它将计算机视觉中的多尺度特征金字塔概念融入了PEFT领域。与标准LoRA不同，EPT将任务适应分解为两个阶段：（1）一个共享的元知识子空间，用于在低维度编码通用语言模式；（2）一个金字塔投影机制，利用可学习的向上投影算子在不同尺度上重建高维特征。随后，一个任务感知路由器动态选择这些多尺度特征的最优组合。在多个多任务基准上的广泛实验表明，EPT显著优于当前最先进的MoE-LoRA变体。关键的是，得益于我们设计的重参数化能力，EPT在实现这一性能提升的同时，还减少了训练参数的数量。

Parameter-Efficient Fine-Tuning (PEFT) has become a dominant paradigm for deploying LLMs in multi-task scenarios due to its extreme parameter efficiency. While Mixture-of-Experts (MoE) based LoRA variants have achieved promising results by dynamically routing tokens to different low-rank experts, they largely overlook the hierarchical nature of task complexity. Existing methods typically employ experts with uniform architectures, limiting their ability to capture diverse feature granularities required by distinct tasks–where some tasks demand high-level semantic abstraction while others require fine-grained syntactic manipulation. To bridge this gap, we propose Expert Pyramid Tuning (EPT), a novel architecture that integrates the multi-scale feature pyramid concept from computer vision into the realm of PEFT. Unlike standard LoRA, EPT decomposes task adaptation into two stages: (1) A shared meta-knowledge Subspace that encodes universal linguistic patterns in low dimensions; (2) A Pyramid Projection Mechanism that utilizes learnable up-projection operators to reconstruct high-dimensional features at varying scales. A task-aware router then dynamically selects the optimal combination of these multi-scale features. Extensive experiments across multiple multi-task benchmarks demonstrate that EPT significantly outperforms SOTA MoE-LoRA variants. Crucially, thanks to the re-parameterization capability of our design, EPT achieves this performance improvement while simultaneously reducing the number of training parameters.

!!! info

论文首先构建一个共享的元知识子空间，使用高斯随机初始化的低秩矩阵编码通用语言模式。随后，利用多个具有不同核大小的反卷积层作为专家，将元知识投影到不同尺度的特征空间，形成参数金字塔。为了兼容冻结的预训练权重，引入Adaptive LoRA Pruner进行维度对齐。最后，采用Top-k路由机制结合对比学习生成的任务嵌入，动态选择并组合多尺度专家的输出，实现高效的多任务微调。

• 创新性地将计算机视觉中的特征金字塔思想迁移到NLP的PEFT领域，解决了传统MoE-LoRA专家架构单一的问题。
• 通过共享元知识子空间减少了参数冗余，在保持高性能的同时提升了参数效率。
• 结合对比学习优化路由策略，提高了多任务场景下专家选择的准确性和模型的鲁棒性。

• 引入反卷积和额外的路由机制可能会增加推理时的计算复杂度，尽管训练参数减少了，但操作步骤可能更为繁琐。
• 模型包含多个超参数（如反卷积核大小、Top-k值、温度参数等），需要进行细致的调优才能达到最佳效果。
• 论文主要在通用NLP任务上进行验证，在特定科学领域（如生物医药）的大模型应用效果尚需进一步实验验证。

评分理由: 论文提出ESPIRE基准，专注于评估视觉语言模型（VLMs）在具身空间推理任务中的表现。核心相关关键词包括：‘LLM Agents’（10分，论文直接涉及具身智能体在模拟世界中的任务执行）、‘World Models’（8分，论文创建模拟世界来物理化VLMs）、‘Chain of Thought’和’System 2 Thinking’（各8分，论文强调空间推理的分解和推理行动过程）。其他关键词如’Large Language Models’（8分，VLMs是大模型的一种）、‘Pre-training’、‘SFT’、‘Instruction Tuning’、‘Explainable AI’、‘In-context Learning’（各5分，论文涉及模型适应和诊断分析，但非核心）。其余关键词与论文内容无关，得0分。

视觉语言模型（VLMs）近期的一个趋势是增强其在具身领域中的空间认知能力。尽管已有进展，现有评估方法在范式和覆盖范围上均存在局限，阻碍了模型的快速迭代开发。为应对这些不足，我们提出了ESPIRE——一个用于具身空间推理的诊断性基准。ESPIRE提供了一个模拟世界，将VLMs置于物理环境中，并以空间推理为核心的机器人任务对其进行评估，从而缩小了评估与实际部署之间的差距。为使VLMs适应机器人任务，我们将每项任务分解为定位与执行两个阶段，并将二者均构建为生成式问题，这与当前主流的基于干扰项且忽略执行的判别式评估（例如通过视觉问答）形成鲜明对比。这种分解进一步支持了从被动空间推理到行动推理的细粒度分析。我们在指令层面和环境层面系统化地设计了ESPIRE，确保其广泛覆盖各类空间推理场景。我们利用ESPIRE对一系列前沿VLM进行诊断，并深入分析了它们的空间推理行为。

A recent trend in vision-language models (VLMs) has been to enhance their spatial cognition for embodied domains. Despite progress, existing evaluations have been limited both in paradigm and in coverage, hindering rapid, iterative model development. To address these limitations, we propose ESPIRE, a diagnostic benchmark for embodied spatial reasoning. ESPIRE offers a simulated world that physically grounds VLMs and evaluates them on spatial-reasoning-centric robotic tasks, thus narrowing the gap between evaluation and real-world deployment. To adapt VLMs to robotic tasks, we decompose each task into localization and execution, and frame both as generative problems, in stark contrast to predominant discriminative evaluations (e.g., via visual-question answering) that rely on distractors and discard execution. This decomposition further enables a fine-grained analysis beyond passive spatial reasoning toward reasoning to act. We systematically design ESPIRE both at the instruction level and at the environment level, ensuring broad coverage of spatial reasoning scenarios. We use ESPIRE to diagnose a range of frontier VLMs and provide in-depth analysis of their spatial reasoning behaviors.

评分理由: 论文NeuroLoRA的核心贡献是提出了一种基于混合专家（MoE）的LoRA框架，用于大语言模型（LLMs）的参数高效微调（PEFT）。因此，与"Large Language Models (LLMs)"、“Mixture of Experts (MoE)“和"PEFT/LoRA"高度相关（10分）。论文明确涉及多任务适应和顺序持续学习，与"Domain Adaptation"和"Model Merging"有一定关联（5分）。论文未直接讨论其他关键词，如SLMs、Scaling Laws、RAG、推理加速或特定科学AI应用，因此这些关键词得0分。

参数高效微调技术，特别是低秩自适应方法，已成为将大语言模型适配至下游任务的关键工具。尽管近期提出的FlyLoRA框架成功利用仿生稀疏随机投影来缓解参数干扰，但其依赖静态的、基于幅度的路由机制，无法感知输入上下文。本文受生物神经调节机制——即根据上下文动态调控神经元兴奋性的启发，提出一种基于混合专家模型的新型LoRA框架：NeuroLoRA。该框架在保持冻结随机投影计算效率的同时，引入了一个轻量级、可学习的神经调节门，能在专家选择前根据上下文对投影空间进行动态重缩放。我们进一步提出对比正交性损失，以显式增强专家子空间之间的分离性，从而提升任务解耦与持续学习能力。在MMLU、GSM8K和ScienceQA基准上的大量实验表明，NeuroLoRA在单任务适配、多任务模型融合及序列持续学习场景中，均持续优于FlyLoRA及其他强基线方法，同时保持了相当的参数效率。

Parameter-Efficient Fine-Tuning (PEFT) techniques, particularly Low-Rank Adaptation (LoRA), have become essential for adapting Large Language Models (LLMs) to downstream tasks. While the recent FlyLoRA framework successfully leverages bio-inspired sparse random projections to mitigate parameter interference, it relies on a static, magnitude-based routing mechanism that is agnostic to input context. In this paper, we propose NeuroLoRA, a novel Mixture-of-Experts (MoE) based LoRA framework inspired by biological neuromodulation – the dynamic regulation of neuronal excitability based on context. NeuroLoRA retains the computational efficiency of frozen random projections while introducing a lightweight, learnable neuromodulation gate that contextually rescales the projection space prior to expert selection. We further propose a Contrastive Orthogonality Loss to explicitly enforce separation between expert subspaces, enhancing both task decoupling and continual learning capacity. Extensive experiments on MMLU, GSM8K, and ScienceQA demonstrate that NeuroLoRA consistently outperforms FlyLoRA and other strong baselines across single-task adaptation, multi-task model merging, and sequential continual learning scenarios, while maintaining comparable parameter efficiency.

评分理由: 论文核心研究奖励模型（RMs）的去偏方法，属于大模型对齐（Alignment）和RLHF技术范畴。与"Instruction Tuning/Alignment/Value Alignment"和"RLHF/RLAIF/DPO"高度相关（10分），因为奖励模型是RLHF对齐流程的关键组件。与"Large Language Models/LLMs/Foundation Models"相关（8分），因为奖励模型通常基于大语言模型构建。与"Mechanistic Interpretability/Explainable AI"相关（8分），因为使用稀疏自编码器（SAE）进行特征解释和干预。与"Hallucination Mitigation/Factuality/Truthfulness"有一定关联（5分），因为去偏有助于提升模型对语义而非表面风格的偏好，间接提升事实性。其他关键词如MoE、SLMs、Scaling Laws、Pre-training、RAG、Context Window等与论文内容无直接关系，得0分。

奖励模型（Reward Models, RMs）是对齐流程中的关键组件，但它们表现出对表层风格线索的偏好偏差，倾向于选择呈现形式更佳而非语义更优的回应。现有的去偏方法通常需要重新训练或调整模型架构，而直接抑制激活会因表征纠缠导致性能下降。我们提出了SteerRM，这是首个基于稀疏自编码器（Sparse Autoencoder, SAE）干预、无需训练即可实现奖励模型去偏的方法。SteerRM利用对比配对回应分离风格效应，通过强度-稳定性准则识别与偏差相关的SAE特征，并在推理阶段抑制这些特征。在RM-Bench的六个奖励模型上，SteerRM平均将Hard-split准确率提升了7.3个百分点，同时保持了整体性能。基于Gemma的奖励模型实验及对非格式偏差的受控研究进一步表明，该方法可泛化至不同RM架构与偏差类型。我们还发现，与格式相关的特征集中分布于模型浅层，且在不同模型间可迁移，这揭示了架构层面共享的偏差编码模式。这些结果表明，基于SAE的干预能够在无需重新训练的情况下缓解奖励模型偏差，为对齐流程提供了一种实用且可解释的解决方案。

Reward models (RMs) are critical components of alignment pipelines, yet they exhibit biases toward superficial stylistic cues, preferring better-presented responses over semantically superior ones. Existing debiasing methods typically require retraining or architectural modifications, while direct activation suppression degrades performance due to representation entanglement. We propose SteerRM, the first training-free method for debiasing reward models using Sparse Autoencoder (SAE)-based interventions. SteerRM isolates stylistic effects using contrastive paired responses, identifies bias-related SAE features with a strength-stability criterion, and suppresses them at inference time. Across six reward models on RM-Bench, SteerRM improves Hard-split accuracy by 7.3 points on average while preserving overall performance. Results on a Gemma-based reward model and a controlled non-format bias further suggest generalization across RM architectures and bias types. We further find that format-related features are concentrated in shallow layers and transfer across models, revealing shared architecture-level bias encoding patterns. These results show that SAE-based interventions can mitigate reward-model biases without retraining, providing a practical and interpretable solution for alignment pipelines.

评分理由: 论文研究大型视觉语言模型（LVLMs）在微调时易受后门攻击的问题，并提出了一种无需重新训练、仅需在测试时操作的防御方法CleanSight。该论文与以下关键词高度相关：1）“Large Language Models” OR “LLMs” OR “Foundation Models”（8分）：论文研究LVLMs，属于大模型范畴，但更侧重于视觉语言模型而非纯语言模型。2）“Post-training” OR “Supervised Fine-tuning” OR “SFT”（8分）：论文明确提到后门攻击发生在微调（fine-tuning）阶段，这是后训练的关键环节。3）“PEFT” OR “LoRA” OR “Parameter-efficient Fine-tuning”（10分）：论文直接提到LoRA模块作为后门攻击的可能目标，并指出现有防御方法涉及重新训练这些参数，这是论文的核心技术背景之一。4）“Mechanistic Interpretability” OR “Explainable AI”（10分）：论文的核心贡献是提供了对LVLMs中后门行为的机制性理解（mechanistic understanding），即通过跨模态注意力重分布（attention stealing）来解释后门激活机制，这直接属于机制可解释性研究。其他关键词与论文内容无关，因为论文专注于视觉语言模型的后门防御、注意力机制分析和测试时净化，不涉及MoE、小模型、缩放律、预训练、对齐、RLHF、RAG、上下文扩展、推理加速、智能体、量化等主题。

尽管大型视觉语言模型（LVLMs）具备强大的多模态性能，其在微调过程中仍易受到后门攻击的威胁——攻击者将嵌入触发器的样本注入训练数据，以植入可在测试阶段被恶意激活的行为。现有防御方法通常依赖于使用干净数据重新训练被植入后门的参数（例如适配器或LoRA模块），这种方式计算成本高昂且常导致模型性能下降。本研究对LVLMs中的后门行为提出了新的机制性解释：触发器并非通过底层视觉模式影响预测，而是通过异常的跨模态注意力重分配发挥作用——携带触发器的视觉令牌会从文本上下文中窃取注意力，我们将此现象称为注意力窃取。基于此发现，我们提出了CleanSight：一种无需训练、即插即用的纯测试阶段防御方案。CleanSight（i）通过选定跨模态融合层中的视觉-文本注意力相对比例来检测中毒输入，并（ii）通过选择性剪枝可疑的高注意力视觉令牌以净化输入，从而中和后门激活。大量实验表明，CleanSight在多种数据集和攻击类型下均显著优于现有的基于像素的净化防御方法，同时在干净样本与中毒样本上均能保持模型的原有效能。

Despite the strong multimodal performance, large vision-language models (LVLMs) are vulnerable during fine-tuning to backdoor attacks, where adversaries insert trigger-embedded samples into the training data to implant behaviors that can be maliciously activated at test time. Existing defenses typically rely on retraining backdoored parameters (e.g., adapters or LoRA modules) with clean data, which is computationally expensive and often degrades model performance. In this work, we provide a new mechanistic understanding of backdoor behaviors in LVLMs: the trigger does not influence prediction through low-level visual patterns, but through abnormal cross-modal attention redistribution, where trigger-bearing visual tokens steal attention away from the textual context - a phenomenon we term attention stealing. Motivated by this, we propose CleanSight, a training-free, plug-and-play defense that operates purely at test time. CleanSight (i) detects poisoned inputs based on the relative visual-text attention ratio in selected cross-modal fusion layers, and (ii) purifies the input by selectively pruning the suspicious high-attention visual tokens to neutralize the backdoor activation. Extensive experiments show that CleanSight significantly outperforms existing pixel-based purification defenses across diverse datasets and backdoor attack types, while preserving the model’s utility on both clean and poisoned samples.

评分理由: 论文是大型语言模型（LLMs）持续学习的综述，核心围绕LLMs的持续学习方法和挑战。因此，与"Large Language Models"和"Continual Pre-training"高度相关（10分）。论文提到持续微调和持续对齐，与"Supervised Fine-tuning"和"Alignment"有一定关联（5分）。论文讨论参数效率，与"Parameter-efficient Fine-tuning"相关（5分）。其他关键词如MoE、SLMs、RAG、推理加速、AI for Science等未在摘要中提及，因此评分为0分。

持续学习（Continual Learning, CL）已成为关键范式，旨在使大语言模型（Large Language Models, LLMs）能够动态适应不断演进的知识与序列任务，同时缓解灾难性遗忘——这一现代大语言模型固有的静态预训练范式的核心局限。本综述全面梳理了针对大语言模型设计的持续学习方法，围绕三个核心训练阶段展开：持续预训练、持续微调与持续对齐。在经典的基于复演、正则化与架构的方法分类之外，我们进一步依据其独特的遗忘缓解机制对每类方法进行细分，并对传统持续学习方法在大语言模型中的适应性与关键改进进行了严谨的比较分析。在此过程中，我们明确强调了大语言模型持续学习与传统机器学习之间的核心区别，特别是在模型规模、参数效率与涌现能力方面。我们的分析涵盖了核心评估指标，包括遗忘率与知识迁移效率，以及评估持续学习性能的新兴基准。本综述揭示，尽管现有方法在特定领域展现出有前景的结果，但在实现跨多样任务与时间尺度的无缝知识整合方面，仍存在根本性挑战。本系统性综述为不断增长的大语言模型适应研究提供了贡献，为研究者与实践者提供了一个结构化框架，以理解语言模型终身学习领域的当前成就与未来机遇。

Continual learning (CL) has emerged as a pivotal paradigm to enable large language models (LLMs) to dynamically adapt to evolving knowledge and sequential tasks while mitigating catastrophic forgetting-a critical limitation of the static pre-training paradigm inherent to modern LLMs. This survey presents a comprehensive overview of CL methodologies tailored for LLMs, structured around three core training stages: continual pre-training, continual fine-tuning, and continual alignment.Beyond the canonical taxonomy of rehearsal-, regularization-, and architecture-based methods, we further subdivide each category by its distinct forgetting mitigation mechanisms and conduct a rigorous comparative analysis of the adaptability and critical improvements of traditional CL methods for LLMs. In doing so, we explicitly highlight core distinctions between LLM CL and traditional machine learning, particularly with respect to scale, parameter efficiency, and emergent capabilities. Our analysis covers essential evaluation metrics, including forgetting rates and knowledge transfer efficiency, along with emerging benchmarks for assessing CL performance. This survey reveals that while current methods demonstrate promising results in specific domains, fundamental challenges persist in achieving seamless knowledge integration across diverse tasks and temporal scales. This systematic review contributes to the growing body of knowledge on LLM adaptation, providing researchers and practitioners with a structured framework for understanding current achievements and future opportunities in lifelong learning for language models.

评分理由: 论文聚焦于视觉-语言模型（VLMs）在拓扑异常检测中的应用，属于大模型在科学领域（医学图像分析）的应用研究。核心相关性体现在：1）使用VLMs（属于大模型范畴），得5分；2）明确采用监督微调（SFT）和强化学习（GRPO，属于RLHF/RLAIF范畴）的两阶段训练方法，这两项得10分；3）应用领域为生物医学图像分析（血管、神经纤维），属于AI for Science/Bioinformatics，得10分。其他关键词如MoE、量化、推理加速等与论文技术内容无关，得0分。

拓扑正确性对于血管、神经纤维及道路网络等管状结构至关重要。现有拓扑保持方法依赖于特定领域的标注真值，其获取成本高昂且难以跨领域迁移。当部署至缺乏标注的新领域时，一个关键问题随之产生：如何在无真值监督的情况下检测拓扑异常？我们将此问题重构为拓扑异常检测——一项结构化视觉推理任务，要求模型在预测的分割掩码中定位并分类拓扑错误。视觉-语言模型（VLMs）是天然的候选方案；然而，我们发现当前最先进的视觉-语言模型表现近乎随机，缺乏识别密集结构中稀疏连通性错误所需的细粒度拓扑感知能力。为弥补这一差距，我们开发了一套自动化数据构建流程，该流程能合成具有可验证标注的多样化拓扑异常，并设置渐进难度等级，从而构建了首个面向该任务的大规模、多领域基准数据集。随后，我们提出Topo-R1框架，通过两阶段训练赋予视觉-语言模型拓扑感知能力：首先进行监督微调，随后采用基于群组相对策略优化（GRPO）的强化学习。我们方法的核心在于设计了一种拓扑感知复合奖励机制，该机制整合了面向类型的匈牙利匹配算法以实现结构化错误分类、空间定位评分，以及直接惩罚连通性中断的中心线戴斯系数（clDice）奖励，从而共同激励语义精确性与结构保真度。大量实验表明，Topo-R1为无标注拓扑质量评估建立了新范式，在所有评估协议中均持续优于通用视觉-语言模型及有监督基线方法。

Topological correctness is crucial for tubular structures such as blood vessels, nerve fibers, and road networks. Existing topology-preserving methods rely on domain-specific ground truth, which is costly and rarely transfers across domains. When deployed to a new domain without annotations, a key question arises: how can we detect topological anomalies without ground-truth supervision? We reframe this as topological anomaly detection, a structured visual reasoning task requiring a model to locate and classify topological errors in predicted segmentation masks. Vision-Language Models (VLMs) are natural candidates; however, we find that state-of-the-art VLMs perform nearly at random, lacking the fine-grained, topology-aware perception needed to identify sparse connectivity errors in dense structures. To bridge this gap, we develop an automated data-curation pipeline that synthesizes diverse topological anomalies with verifiable annotations across progressively difficult levels, thereby constructing the first large-scale, multi-domain benchmark for this task. We then introduce Topo-R1, a framework that endows VLMs with topology-aware perception via two-stage training: supervised fine-tuning followed by reinforcement learning with Group Relative Policy Optimization (GRPO). Central to our approach is a topology-aware composite reward that integrates type-aware Hungarian matching for structured error classification, spatial localization scoring, and a centerline Dice (clDice) reward that directly penalizes connectivity disruptions, thereby jointly incentivizing semantic precision and structural fidelity. Extensive experiments demonstrate that Topo-R1 establishes a new paradigm for annotation-free topological quality assessment, consistently outperforming general-purpose VLMs and supervised baselines across all evaluation protocols.

评分理由: 论文研究视觉文档检索（VDR）中的模型蒸馏技术，将2B参数的视觉语言模型（VLM）蒸馏为69M参数的纯文本编码器。核心相关关键词：1）“Small Language Models” (8分)：论文明确开发69M参数的轻量级模型，属于小型语言模型范畴；2）“Retrieval-Augmented Generation” (8分)：论文专注于检索任务，是RAG系统的关键组件；3）“Quantization” (8分)：通过蒸馏实现模型压缩，大幅减少参数和延迟；4）“Large Language Models” (5分)：使用2B VLM作为教师模型，属于大模型范畴；5）“Speculative Decoding” (5分)：通过轻量化实现推理加速，与加速目标相关。其他关键词如MoE、对齐、推理方法等与论文内容无直接关联。

基于视觉语言模型（VLM）的检索器已将视觉文档检索（VDR）的质量提升至令人瞩目的水平。这类方法在文档索引和查询编码时均需使用相同的数十亿参数编码器，导致即使处理纯文本查询也存在高延迟和强GPU依赖的问题。我们观察到，这种设计存在不必要的对称性：文档在视觉上复杂，需要强大的视觉理解能力，而查询仅为短文本字符串。NanoVDR通过解耦两条编码路径来利用这种查询-文档不对称性：一个冻结的20亿参数VLM教师模型离线处理文档索引，而一个经蒸馏的、小至6900万参数的纯文本学生模型在推理时编码查询。其核心设计在于蒸馏目标的选择。通过对三种骨干网络和22个ViDoRe基准数据集上的六种目标进行系统比较，我们发现，在查询文本上进行逐点余弦对齐的方法持续优于基于排序和对比学习的替代方案，且训练时仅需预缓存的教师查询嵌入，无需处理文档。此外，我们指出跨语言迁移是主要性能瓶颈，并通过使用机器翻译的查询数据增强训练集，以低成本解决了该问题。最终得到的NanoVDR-S-Multi（基于DistilBERT，6900万参数）保留了教师模型95.1%的性能，在v2和v3版本上超越了DSE-Qwen2（20亿参数），同时参数量减少了32倍，CPU查询延迟降低了50倍，总训练成本低于13 GPU小时。

Vision-Language Model (VLM) based retrievers have advanced visual document retrieval (VDR) to impressive quality. They require the same multi-billion parameter encoder for both document indexing and query encoding, incurring high latency and GPU dependence even for plain-text queries. We observe that this design is unnecessarily symmetric: documents are visually complex and demand strong visual understanding, whereas queries are just short text strings. NanoVDR exploits this query–document asymmetry by decoupling the two encoding paths: a frozen 2B VLM teacher indexes documents offline, while a distilled text-only student as small as 69M parameters encodes queries at inference. The key design choice is the distillation objective. Through systematic comparison of six objectives across three backbones and 22 ViDoRe benchmark datasets, we find that pointwise cosine alignment on query text consistently outperforms ranking-based and contrastive alternatives, while requiring only pre-cached teacher query embeddings and no document processing during training. Furthermore, we identify cross-lingual transfer as the primary performance bottleneck, and resolve it cheaply by augmenting training data with machine-translated queries. The resulting NanoVDR-S-Multi (DistilBERT, 69M) retains 95.1% of teacher quality and outperforms DSE-Qwen2 (2B) on v2 and v3 with 32$\times$ fewer parameters and 50$\times$ lower CPU query latency, at a total training cost under 13 GPU-hours.

评分理由: 论文提出AVION框架，用于遥感图像视觉语言模型的适应，核心涉及大语言模型（LLMs）生成文本描述（相关度8），属于视觉语言模型的领域适应（相关度8），采用轻量级可学习提示进行参数高效微调（相关度8），并应用于遥感科学（AI for Science相关度8）。其他关键词如MoE、SLMs、SFT、RAG等未在摘要中体现，故评0分。

将视觉语言模型适配于遥感影像仍面临两大挑战：文本表征的语义覆盖范围有限，以及视觉特征适应性不足。这些问题在涉及多样化视觉外观与细粒度目标区分的航空场景中尤为显著。我们提出AVION——一个专为视觉语言模型遥感适配设计的知识蒸馏框架。教师模块通过收集大语言模型的描述文本并利用遥感影像特征验证有效性，构建语义丰富的文本原型。学生模块则在视觉与语言编码器中分别集成轻量级可学习提示向量，并在教师模块的指导下对齐嵌入表示及其跨模态关联。训练完成后，学生模块可在推理阶段独立运行。在六个光学遥感基准数据集上的实验表明，AVION在提升小样本分类与基类识别准确率的同时，未削弱对新类别的泛化能力。该框架还显著提高了跨模态检索的平均召回率，且仅需引入极少量可训练参数。

Adapting vision-language models to remote sensing imagery remains challenging due to two key factors: limited semantic coverage in textual representations and insufficient adaptability of visual features. These issues are particularly significant in aerial scenes, which involve various visual appearances and fine-grained object distinctions. We propose AVION, a knowledge distillation framework tailored for remote sensing adaptation of vision-language models. The teacher module constructs semantically rich textual prototypes by collecting descriptions from a large language model and verifying validity using remote sensing image features. The student module integrates lightweight and learnable prompts into both vision and language encoders, guided by the teacher to align embeddings and their cross-modal relationships. Once trained, the student operates independently during inference. Experiments on six optical remote sensing benchmarks show that AVION improves few-shot classification and base-class accuracy without degrading generalization to novel categories. It also enhances mean recall for cross-modal retrieval, with minimal additional trainable parameters.

评分理由: 论文核心研究Parameter-Efficient Fine-Tuning (PEFT)和Mixture-of-Experts (MoE)技术，提出Expert Pyramid Tuning (EPT)方法，直接与关键词"PEFT/LoRA"和"MoE"高度相关（10分）。论文明确针对LLMs部署，与"Large Language Models"高度相关（10分）。其他关键词如SLMs、Scaling Laws、RAG、RLHF等未在摘要中提及或相关，均给0分。

参数高效微调（Parameter-Efficient Fine-Tuning, PEFT）因其极高的参数效率，已成为在多任务场景中部署大语言模型（LLMs）的主导范式。尽管基于专家混合（Mixture-of-Experts, MoE）的LoRA变体通过将令牌动态路由至不同的低秩专家取得了有前景的结果，但它们大多忽视了任务复杂性的层级本质。现有方法通常采用架构统一的专家，限制了其捕捉不同任务所需多样化特征粒度的能力——有些任务需要高层语义抽象，而另一些则需要细粒度的句法操控。为弥补这一差距，我们提出了专家金字塔调优（Expert Pyramid Tuning, EPT），这是一种新颖的架构，它将计算机视觉中的多尺度特征金字塔概念融入了PEFT领域。与标准LoRA不同，EPT将任务适应分解为两个阶段：（1）一个共享的元知识子空间，用于在低维度编码通用语言模式；（2）一个金字塔投影机制，利用可学习的向上投影算子在不同尺度上重建高维特征。随后，一个任务感知路由器动态选择这些多尺度特征的最优组合。在多个多任务基准上的广泛实验表明，EPT显著优于当前最先进的MoE-LoRA变体。关键的是，得益于我们设计的重参数化能力，EPT在实现这一性能提升的同时，还减少了训练参数的数量。

Parameter-Efficient Fine-Tuning (PEFT) has become a dominant paradigm for deploying LLMs in multi-task scenarios due to its extreme parameter efficiency. While Mixture-of-Experts (MoE) based LoRA variants have achieved promising results by dynamically routing tokens to different low-rank experts, they largely overlook the hierarchical nature of task complexity. Existing methods typically employ experts with uniform architectures, limiting their ability to capture diverse feature granularities required by distinct tasks–where some tasks demand high-level semantic abstraction while others require fine-grained syntactic manipulation. To bridge this gap, we propose Expert Pyramid Tuning (EPT), a novel architecture that integrates the multi-scale feature pyramid concept from computer vision into the realm of PEFT. Unlike standard LoRA, EPT decomposes task adaptation into two stages: (1) A shared meta-knowledge Subspace that encodes universal linguistic patterns in low dimensions; (2) A Pyramid Projection Mechanism that utilizes learnable up-projection operators to reconstruct high-dimensional features at varying scales. A task-aware router then dynamically selects the optimal combination of these multi-scale features. Extensive experiments across multiple multi-task benchmarks demonstrate that EPT significantly outperforms SOTA MoE-LoRA variants. Crucially, thanks to the re-parameterization capability of our design, EPT achieves this performance improvement while simultaneously reducing the number of training parameters.

评分理由: 论文核心研究基于LLMs的In-Context Learning（ICL）在CAD代码生成中的应用，因此与"Large Language Models"和"In-context Learning"高度相关（10分）。论文涉及CAD（计算机辅助设计），属于AI在工程/科学领域的应用，与"AI for Science"有一定关联（5分）。其他关键词如MoE、SFT、RAG、量化等均未在摘要中提及或与论文主题无关，故评0分。

大型语言模型（LLM）在代码生成方面展现出卓越能力，但由于训练数据稀缺，其在计算机辅助设计（CAD）代码生成等特定领域任务中表现欠佳。上下文学习（In-Context Learning, ICL）通过提供任务相关的示例样本，提供了一种免训练的替代方案。然而，现有的示例选择策略通常优先考虑相似性或点状多样性，往往产生冗余选择，无法满足复杂CAD设计规范中组合式需求的要求。在本研究中，我们提出将知识充分性作为示例选择的原则性目标，旨在最大程度地满足设计规范中的所有需求。为实现这一目标，我们引入了设计规范平铺（Design-Specification Tiling, DST）方法，该方法通过提取多粒度设计组件并计算所选示例覆盖查询组件的比例，以代理平铺率量化知识充分性。我们证明了最大化该目标等价于子模最大化问题，并提出了一种具有（1-1/e）近似保证的多项式时间贪心算法。大量实验表明，DST显著提升了CAD代码生成质量，在ICL中持续优于现有的示例选择策略。

Large language models (LLMs) have demonstrated remarkable capabilities in code generation, yet they underperform on domain-specific tasks such as Computer-Aided Design (CAD) code generation due to scarce training data. In-Context Learning (ICL) offers a training-free alternative through task-specific exemplars. However, existing selection strategies prioritize similarity or point-wise diversity, often producing redundant selections that fail to satisfy the compositional requirements of complex CAD design specifications. In this work, we propose knowledge sufficiency as a principled objective for exemplar selection that aims to maximally satisfy all requirements within design specifications. To realize this objective, we introduce Design-Specification Tiling (DST), which quantifies knowledge sufficiency through a surrogate tiling ratio by extracting multi-granular design components and measuring the proportion of query components covered by selected exemplars. We demonstrate that maximizing this objective constitutes submodular maximization and provide a polynomial-time greedy algorithm with a (1-1/e)-approximation guarantee. Extensive experiments demonstrate that DST substantially improves CAD code generation quality, consistently outperforming existing exemplar selection strategies in ICL.

评分理由: 论文研究的是图神经网络中的上下文学习（in-context learning）在时空预测中的应用，与"In-context Learning"关键词高度相关（10分），因为这是论文的核心方法。论文应用于空气质量预测，属于科学领域的AI应用，与"AI for Science"有一定关联（8分）。其他所有关键词都涉及大语言模型（LLM）的特定技术、训练方法、推理优化、对齐、代理系统等，而本文研究的是图神经网络和算子学习，完全不涉及LLM技术，因此相关度为0分。

上下文算子学习使神经网络能够在不更新权重的情况下，从上下文示例中推断解算子。尽管先前研究已证明该范式在利用大规模数据集方面的有效性，但尚未出现使用相同训练数据与单算子学习进行系统比较的工作。我们通过控制实验填补了这一空白：在相同训练步数和数据集下，比较上下文算子学习与经典算子学习（即不使用上下文示例训练的单算子模型）。为了在现实世界时空系统上开展此项研究，我们提出了GICON（图上下文算子网络），该网络结合了用于几何泛化的图消息传递机制与用于基数泛化的示例感知位置编码。在两个中国区域的空气质量预测实验表明，对于复杂任务，上下文算子学习优于经典算子学习，能够实现跨空间域的泛化，并在推理时从少量训练示例稳健地扩展至100个示例。

In-context operator learning enables neural networks to infer solution operators from contextual examples without weight updates. While prior work has demonstrated the effectiveness of this paradigm in leveraging vast datasets, a systematic comparison against single-operator learning using identical training data has been absent. We address this gap through controlled experiments comparing in-context operator learning against classical operator learning (single-operator models trained without contextual examples), under the same training steps and dataset. To enable this investigation on real-world spatiotemporal systems, we propose GICON (Graph In-Context Operator Network), combining graph message passing for geometric generalization with example-aware positional encoding for cardinality generalization. Experiments on air quality prediction across two Chinese regions show that in-context operator learning outperforms classical operator learning on complex tasks, generalizing across spatial domains and scaling robustly from few training examples to 100 at inference.

评分理由: 论文主要研究利用人机协作系统SCILIRE从科学文献中创建和整理数据集，核心是LLM在科学领域的应用（AI for Science），因此与"Large Language Models"和"AI for Science"等关键词相关。其他关键词涉及具体技术原理（如MoE、SFT、RAG等）或特定应用场景（如生物信息学），论文未明确涉及，故评分为0。

科学文献的快速增长使得人工提取结构化知识日益不切实际。为应对这一挑战，我们引入了SCILIRE系统，该系统用于从科学文献中创建数据集。SCILIRE的设计基于以数据验证与整理工作流程为核心的人机协同原则。它支持一种迭代式工作流程，研究人员可在其中审查并修正人工智能的输出结果。此外，这种交互被用作反馈信号，以改进未来基于大语言模型（LLM）的推理性能。我们通过结合内在基准测试结果与跨多个领域的实际案例研究来评估该设计。结果表明，SCILIRE能够提升信息提取的准确性，并促进高效的数据集构建。

The rapid growth of scientific literature has made manual extraction of structured knowledge increasingly impractical. To address this challenge, we introduce SCILIRE, a system for creating datasets from scientific literature. SCILIRE has been designed around Human-AI teaming principles centred on workflows for verifying and curating data. It facilitates an iterative workflow in which researchers can review and correct AI outputs. Furthermore, this interaction is used as a feedback signal to improve future LLM-based inference. We evaluate our design using a combination of intrinsic benchmarking outcomes together with real-world case studies across multiple domains. The results demonstrate that SCILIRE improves extraction fidelity and facilitates efficient dataset creation.

评分理由: 论文研究单层物理信息神经网络（PINNs）在非线性偏微分方程上的经验缩放定律，属于AI for Science（科学AI）领域，因此与"AI for Science"关键词高度相关（8分）。论文核心是建立缩放定律，与"Scaling Laws"关键词直接相关（8分），但论文未涉及数据质量，因此未完全匹配"Scaling Laws AND Data Quality”。其他关键词均涉及大语言模型（LLMs）或深度学习技术原理，而本文专注于PINNs这一特定神经网络架构在科学计算中的应用，未涉及LLMs、MoE、训练方法、推理优化、代理系统等主题，因此相关度为0分。

我们针对典型非线性偏微分方程，建立了单层物理信息神经网络的实证缩放规律。我们发现了一种双重优化失效现象：(i) 基础性病理现象：即使在固定非线性度的情况下，解误差也未能随网络宽度增加而降低，未能达到理论近似界限；(ii) 复合性病理现象：非线性因素加剧了这种失效。我们提供了定量证据，表明简单的可分离幂律关系不足以描述该现象，其缩放行为受更复杂的非可分离关系支配。这种失效与谱偏差的概念一致，即神经网络难以学习随非线性增强而加剧的高频解分量。我们证明主要瓶颈在于优化过程而非近似能力，并提出了一种实证测量这些复杂缩放效应的方法论。

We establish empirical scaling laws for Single-Layer Physics-Informed Neural Networks on canonical nonlinear PDEs. We identify a dual optimization failure: (i) a baseline pathology, where the solution error fails to decrease with network width, even at fixed nonlinearity, falling short of theoretical approximation bounds, and (ii) a compounding pathology, where this failure is exacerbated by nonlinearity. We provide quantitative evidence that a simple separable power law is insufficient, and that the scaling behavior is governed by a more complex, non-separable relationship. This failure is consistent with the concept of spectral bias, where networks struggle to learn the high-frequency solution components that intensify with nonlinearity. We show that optimization, not approximation capacity, is the primary bottleneck, and propose a methodology to empirically measure these complex scaling effects.

评分理由: 论文研究语音匿名化攻击，属于语音处理和隐私保护领域，与大多数大模型/深度学习技术关键词无关。仅与"Pre-training"和"Post-training"有一定关联（5分），因为论文采用三阶段训练策略，涉及基础训练和轻量级适应，但并非大模型背景下的预训练或微调。其他关键词均未涉及，得0分。

语音匿名化技术旨在掩蔽说话人声纹特征的同时保留语言内容，但其输出仍可能泄露说话人特有的模式。为评估并强化隐私保护效果，我们提出一种双流攻击模型，该模型通过并行编码器融合声谱特征与自监督学习特征，并采用三阶段训练策略。第一阶段建立基础说话人判别表征。第二阶段利用语音转换与匿名化共有的身份转换特性，使模型接触多样化的转换后语音以构建跨系统鲁棒性。第三阶段对目标匿名化数据进行轻量级自适应。在VoicePrivacy攻击者挑战赛（VPAC）数据集上的实验表明，第二阶段是泛化能力的主要驱动力，使其在未见过的匿名化数据集上表现出强大的攻击性能。结合第三阶段后，仅需对目标匿名化数据集中10%的数据进行微调，即可在等错误率（EER）指标上超越当前最优攻击模型。

Voice anonymization masks vocal traits while preserving linguistic content, which may still leak speaker-specific patterns. To assess and strengthen privacy evaluation, we propose a dual-stream attacker that fuses spectral and self-supervised learning features via parallel encoders with a three-stage training strategy. Stage I establishes foundational speaker-discriminative representations. Stage II leverages the shared identity-transformation characteristics of voice conversion and anonymization, exposing the model to diverse converted speech to build cross-system robustness. Stage III provides lightweight adaptation to target anonymized data. Results on the VoicePrivacy Attacker Challenge (VPAC) dataset demonstrate that Stage II is the primary driver of generalization, enabling strong attacking performance on unseen anonymization datasets. With Stage III, fine-tuning on only 10% of the target anonymization dataset surpasses current state-of-the-art attackers in terms of EER.

评分理由: 论文《MotionAnymesh》主要研究将静态3D网格转换为可交互的铰接式资产，用于具身AI和机器人仿真。其核心贡献在于一个结合了视觉语言模型（VLM）与物理先验的零样本框架，以解决运动学幻觉和网格穿透问题。论文与绝大多数关键词无关，因为这些关键词主要涉及大语言模型（LLM）的技术原理、训练方法、推理优化、代理系统等，而本文聚焦于计算机视觉、3D几何和物理仿真，并未涉及LLM。仅有两个关键词有微弱关联：1）“Hallucination Mitigation”（5分）：论文明确提到解决VLM的"kinematic hallucinations”，这与缓解幻觉问题在概念上相关，但针对的是视觉-语言模型在3D任务中的幻觉，而非LLM的文本幻觉。2）“AI for Science”（5分）：论文应用于机器人仿真和数字孪生，可视为AI在科学/工程领域的应用，但并非生物信息学或化学信息学等典型科学AI子领域。因此，加权总分较低。

将静态三维网格模型转化为可交互的铰接式资产对于具身人工智能与机器人仿真至关重要。然而，现有的零样本流程因严重缺乏物理基础而在处理复杂资产时面临困难。具体而言，未基于物理的视觉语言模型常出现运动学幻觉问题，而无约束的关节估计则不可避免地导致物理仿真中灾难性的网格互穿现象。为弥合这一差距，我们提出MotionAnymesh——一个自动化的零样本框架，能够将非结构化的静态网格无缝转化为可直接用于仿真的数字孪生体。我们的方法具有以下特点：首先，通过搭载具备显式SP4D物理先验的动力学感知部件分割模块，将视觉语言模型的推理过程建立在物理基础之上，从而有效消除运动学幻觉；其次，提出几何-物理联合估计流程，将鲁棒的类型感知初始化与物理约束轨迹优化相结合，严格保证铰接运动过程中的无碰撞特性。大量实验表明，MotionAnymesh在几何精度与动态物理可执行性方面均显著优于现有先进基线，为下游应用提供了高度可靠的数字资产。

Converting static 3D meshes into interactable articulated assets is crucial for embodied AI and robotic simulation. However, existing zero-shot pipelines struggle with complex assets due to a critical lack of physical grounding. Specifically, ungrounded Vision-Language Models (VLMs) frequently suffer from kinematic hallucinations, while unconstrained joint estimation inevitably leads to catastrophic mesh inter-penetration during physical simulation. To bridge this gap, we propose MotionAnymesh, an automated zero-shot framework that seamlessly transforms unstructured static meshes into simulation-ready digital twins. Our method features a kinematic-aware part segmentation module that grounds VLM reasoning with explicit SP4D physical priors, effectively eradicating kinematic hallucinations. Furthermore, we introduce a geometry-physics joint estimation pipeline that combines robust type-aware initialization with physics-constrained trajectory optimization to rigorously guarantee collision-free articulation. Extensive experiments demonstrate that MotionAnymesh significantly outperforms state-of-the-art baselines in both geometric precision and dynamic physical executability, providing highly reliable assets for downstream applications.

评分理由: 该论文专注于工业制造中的具体应用，提出了一种基于视觉的层次化深度学习框架，用于自动分类磨料砂轮的磨损状态。论文的核心是计算机视觉和深度学习在工业检测中的应用，使用了EfficientNetV2架构和迁移学习。所有关键词均与大语言模型（LLM）或通用大模型技术直接相关，而本文未涉及任何LLM、基础模型或相关技术（如MoE、缩放定律、对齐、RAG、推理加速等）。唯一略有相关的是“Explainable AI”（通过Grad-CAM进行可解释性分析）和“AI for Science”（可视为工业工程领域的AI应用），但相关性较弱，并非核心。因此，绝大多数关键词评分为0，仅两个关键词评分为5（表示有一定关联但非核心）。

柔性砂布轮因其灵活性常用于复杂自由曲面的精加工。然而，这种灵活性会导致复杂的磨损形态，如叶片轮廓的凹形/凸形变形或叶片撕裂，从而影响磨削效果。本文提出了一种新颖的、基于视觉的分层分类框架，以实现砂布轮磨损状态监测的自动化。与单一分类方法不同，我们将问题分解为三个逻辑层次：(1) 状态检测（新 vs. 磨损），(2) 磨损类型识别（矩形、凹形、凸形）及叶片撕裂检测，以及(3) 严重程度评估（部分变形 vs. 完全变形）。研究构建了一个真实砂布轮图像的自定义数据集，并采用了基于EfficientNetV2架构的迁移学习方法。结果表明该方法具有很高的鲁棒性，分类准确率从93.8%（叶片撕裂）到99.3%（凹形严重度）不等。此外，研究利用梯度加权类激活映射（Grad-CAM）验证了模型学习到的是具有物理相关性的特征，并分析了错误分类的原因。所提出的分层方法为自动化砂布轮磨削中的自适应过程控制及磨损考量提供了基础。

Abrasive flap wheels are common for finishing complex free-form surfaces due to their flexibility. However, this flexibility results in complex wear patterns such as concave/convex flap profiles or flap tears, which influence the grinding result. This paper proposes a novel, vision-based hierarchical classification framework to automate the wear condition monitoring of flap wheels. Unlike monolithic classification approaches, we decompose the problem into three logical levels: (1) state detection (new vs. worn), (2) wear type identification (rectangular, concave, convex) and flap tear detection, and (3) severity assessment (partial vs. complete deformation). A custom-built dataset of real flap wheel images was generated and a transfer learning approach with EfficientNetV2 architecture was used. The results demonstrate high robustness with classification accuracies ranging from 93.8% (flap tears) to 99.3% (concave severity). Furthermore, Gradient-weighted Class Activation Mapping (Grad-CAM) is utilized to validate that the models learn physically relevant features and examine false classifications. The proposed hierarchical method provides a basis for adaptive process control and wear consideration in automated flap wheel grinding.

评分理由: 论文专注于AI在气象预测和基础设施风险评估中的应用，具体开发了一个AI驱动的校正-降尺度框架（ACDF），用于将粗粒度的AI天气预测转化为高分辨率、无偏的风场和基础设施故障概率。论文的核心是AI在科学（气象学、工程学）领域的应用，与关键词列表中的绝大多数技术原理（如LLM、MoE、SFT、RLHF、RAG、CoT等）无直接关联。唯一高度相关的关键词是"AI for Science”，因为论文明确属于AI在科学（气象和工程）领域的应用研究。其他关键词均未在论文标题或摘要中提及或暗示，因此评分为0。

本文针对基础设施韧性领域的一项缺失能力：将快速、全球性的人工智能天气预测转化为资产尺度、可操作的风险信息。我们提出了基于人工智能的校正-降尺度框架（AI-based Correction-Downscaling Framework, ACDF），该框架可将粗分辨率的人工智能天气预测（AIWP）转化为500米分辨率、无偏的风场以及热带气旋影响下输电塔/线路的故障概率。ACDF将风暴尺度的偏差校正与地形感知的降尺度过程分离，在恢复主导结构载荷的亚公里尺度变异性的同时，避免了误差传播。通过对影响中国浙江的11场台风进行留一风暴交叉验证，ACDF将站点尺度风速的平均绝对误差较盘古天气模型降低了38.8%，其表现与同化观测的中尺度分析相当，而每个12小时预报周期在单GPU上仅需运行25秒。在台风“黑格比”的案例中，ACDF再现了观测到的高风速尾部分布，识别出一条沿海高风险走廊，并成功标记出实际发生故障的线路，展示了在塔线和线路尺度上提供可操作指导的能力。ACDF为从人工智能全球预报到关键基础设施的、基于影响的业务化预警，提供了一条端到端的实现路径。

This paper addresses a missing capability in infrastructure resilience: turning fast, global AI weather forecasts into asset-scale, actionable risk. We introduce the AI-based Correction-Downscaling Framework (ACDF), which transforms coarse AI weather prediction (AIWP) into 500-m, unbiased wind fields and transmission tower/line failure probabilities for tropical cyclones. ACDF separates storm-scale bias correction from terrain-aware downscaling, preventing error propagation while restoring sub-kilometer variability that governs structural loading. Tested on 11 typhoons affecting Zhejiang, China under leave-one-storm-out evaluation, ACDF reduces station-scale wind-speed MAE by 38.8% versus Pangu-Weather, matches observation-assimilated mesoscale analyses, yet runs in 25 s per 12-h cycle on a single GPU. In the Typhoon Hagupit case, ACDF reproduced observed high-wind tails, isolated a coastal high-risk corridor, and flagged the line that failed, demonstrating actionable guidance at tower and line scales. ACDF provides an end-to-end pathway from AI global forecasts to operational, impact-based early warning for critical infrastructure.

评分理由: 该论文主要研究水团簇中的氢原子漫游反应，属于计算化学和物理化学领域。论文使用了第一性原理计算和机器学习分析，但这里的机器学习是用于分析化学反应机制（如识别反应物偶极矩作为决定性开关），属于传统的可解释机器学习在科学计算中的应用，而非大模型或深度学习技术。因此，绝大多数关键词（涉及LLM、训练方法、推理优化、智能体等）完全不相关，得0分。仅有两个关键词有微弱关联：1）“Mechanistic Interpretability” OR “Explainable AI”：论文提到"interpretable machine learning analysis”，旨在解释化学反应机制，与可解释AI有一定关联，但非AI模型本身的可解释性，给5分。2）“AI for Science” OR “Bioinformatics” OR “Cheminformatics”：论文使用机器学习分析化学问题，属于AI在科学（具体是化学）中的应用，但非生物信息学或化学信息学核心领域，给5分。其他关键词均无直接关联。

水介导着广泛的化学反应，包括质子转移、键重排和常规自由基过程，这些构成了其不断扩展的本征反应体系。然而，漫游（roaming）作为一种基本反应机制——即解离片段绕过最小能量路径进行重组——尚未在水自身中被发现。本文通过高精度第一性原理（first-principles）的从头算（ab initio）计算，报道了在水团簇中氢原子漫游反应的发现。中性氢原子以自由基形式解离，在平坦的势能面上漫游，并通过与已知氢键网络重排连接相同反应物与产物的路径进行重组。可解释机器学习分析识别出反应物偶极矩是决定漫游是否发生的关键开关，其背后由交换排斥作用与静电相互作用支撑。一旦漫游启动，极化率和自旋布居决定能垒高度，而漫游氢原子的电荷分布则调控能垒宽度，这些共同由静电、轨道及色散贡献所塑造。这些发现确立了氢原子漫游作为水中一个先前未被认识的本征反应类别，为水反应性的机理图景补充了一个基本维度。

Water mediates a broad range of chemical reactions, including proton transfer, bond rearrangement, and conventional radical processes, defining a continuously expanding repertoire of intrinsic reactivity. However, roaming, a fundamental reaction mechanism that a departing fragment bypasses the minimum energy path to recombine, has not been identified in water itself. Here, we report the discovery of hydrogen-atom roaming reactions in water clusters through high-precision ab initio calculations of first-principles. A neutral hydrogen atom departs as a radical, roams across the flat potential energy surface, and recombines along pathways that connect the same reactants and products as known hydrogen-bond network rearrangements. Interpretable machine learning analysis identifies the reactant dipole moment as the decisive switch governing whether roaming occurs, underpinned by exchange-repulsion and electrostatic interactions. Once roaming is initiated, polarizability and spin population determine barrier heights, while the charge distribution of the roaming hydrogen atom governs barrier widths, collectively shaped by electrostatic, orbital, and dispersion contributions. These findings establish hydrogen-atom roaming as a previously unrecognized intrinsic reaction class in water, complementing a fundamental dimension to the mechanistic picture of water reactivity.

评分理由: 该论文主要研究利用视觉语言模型（VLM）增强水下图像复原的语义敏感性，属于计算机视觉与多模态AI交叉应用。论文核心是VLM在特定视觉任务（水下图像增强）中的应用，而非通用大语言模型（LLM）技术。所有关键词均围绕LLM技术原理、训练方法、推理优化、代理系统等，与论文的VLM应用焦点无直接关联。唯一略有相关的是“AI for Science”关键词，因为水下图像增强可视为AI在海洋科学或环境监测中的应用，但论文未明确提及科学领域，故给5分（有一定关联）。其他关键词均得0分（完全无关）。

近年来，基于学习的水下图像增强技术迅速发展。然而，高质量增强输出与自然图像之间的分布差异可能阻碍下游视觉任务对语义线索的提取，从而限制现有增强模型的适应性。为应对这一挑战，本研究提出一种新的学习机制，利用视觉-语言模型赋予水下图像增强模型语义感知能力。具体而言，我们的策略首先通过视觉-语言模型从退化图像生成关键对象的文本描述。随后，一个文本-图像对齐模型将这些相关描述重新映射到图像上，生成空间语义引导图。该引导图通过双引导机制——结合交叉注意力与显式对齐损失——引导水下图像增强网络。这迫使网络在图像重建过程中将修复能力集中于语义敏感区域，而非追求全局均匀的改善，从而确保关键对象特征的真实恢复。实验证实，当该策略应用于不同水下图像增强基线模型时，能显著提升其在感知质量指标上的表现，并增强其在检测与分割任务中的性能，验证了其有效性与适应性。

In recent years, learning-based underwater image enhancement (UIE) techniques have rapidly evolved. However, distribution shifts between high-quality enhanced outputs and natural images can hinder semantic cue extraction for downstream vision tasks, thereby limiting the adaptability of existing enhancement models. To address this challenge, this work proposes a new learning mechanism that leverages Vision-Language Models (VLMs) to empower UIE models with semantic-sensitive capabilities. To be concrete, our strategy first generates textual descriptions of key objects from a degraded image via VLMs. Subsequently, a text-image alignment model remaps these relevant descriptions back onto the image to produce a spatial semantic guidance map. This map then steers the UIE network through a dual-guidance mechanism, which combines cross-attention and an explicit alignment loss. This forces the network to focus its restorative power on semantic-sensitive regions during image reconstruction, rather than pursuing a globally uniform improvement, thereby ensuring the faithful restoration of key object features. Experiments confirm that when our strategy is applied to different UIE baselines, significantly boosts their performance on perceptual quality metrics as well as enhances their performance on detection and segmentation tasks, validating its effectiveness and adaptability.

评分理由: 论文核心研究LLM驱动的AI代理在计算材料科学中的应用，通过QMatSuite平台实现知识积累和反思，与LLM代理、AI for Science高度相关（10分）。涉及检索增强生成（RAG）、思维链推理、系统2深度思考、自我反思/校正、工具使用等机制（8分）。其他关键词如MoE、量化、对齐训练等未涉及（0分）。

尽管大语言模型（LLM）已将AI智能体转变为计算材料科学领域的熟练执行者，但执行上百次模拟并不能造就一名研究者。研究区别于常规执行的关键在于知识的渐进式积累——学习哪些方法会失败、识别不同体系间的模式，并将理解应用于新问题。然而，当前AI驱动的计算科学主流范式将每次执行视为孤立事件，很大程度上丢弃了不同运行间来之不易的洞见。为此，我们推出开源平台QMatSuite以弥合这一差距。智能体以完整溯源方式记录发现，在新计算前检索已有知识，并在专门的反思环节中修正错误结论，将观察结果综合为跨化合物规律。在六步量子力学模拟工作流的基准测试中，积累的知识使推理开销降低67%，并将准确度从与文献值47%的偏差提升至3%的偏差——当将所学知识迁移至陌生材料时，更实现了1%的偏差且流程失败率为零。

While large language models (LLMs) have transformed AI agents into proficient executors of computational materials science, performing a hundred simulations does not make a researcher. What distinguishes research from routine execution is the progressive accumulation of knowledge – learning which approaches fail, recognizing patterns across systems, and applying understanding to new problems. However, the prevailing paradigm in AI-driven computational science treats each execution in isolation, largely discarding hard-won insights between runs. Here we present QMatSuite, an open-source platform closing this gap. Agents record findings with full provenance, retrieve knowledge before new calculations, and in dedicated reflection sessions correct erroneous findings and synthesize observations into cross-compound patterns. In benchmarks on a six-step quantum-mechanical simulation workflow, accumulated knowledge reduces reasoning overhead by 67% and improves accuracy from 47% to 3% deviation from literature – and when transferred to an unfamiliar material, achieves 1% deviation with zero pipeline failures.

评分理由: 该论文主要研究基于扩散模型的人体运动生成，并提出了PhysMoDPO框架，该框架使用Direct Preference Optimization（DPO）来优化模型，使生成的物理模拟运动更符合物理约束和文本指令。因此，与关键词’RLHF OR RLAIF OR Direct Preference Optimization OR DPO’高度相关（10分），因为DPO是核心方法。与’AI for Science OR Bioinformatics OR Cheminformatics’有一定关联（5分），因为该研究涉及机器人控制和物理模拟，属于AI在科学领域的应用，但并非生物信息学或化学信息学。其他关键词主要涉及大语言模型（LLMs）的技术细节，如MoE、Scaling Laws、PEFT、RAG等，而本文专注于扩散模型和运动生成，与这些LLM技术无关，因此评分为0分。

近年来，文本条件人体运动生成领域的进展主要得益于基于大规模人体运动数据训练的扩散模型。在此基础上，近期研究尝试通过应用全身控制器（Whole-Body Controller, WBC）将扩散模型生成的运动转化为可执行轨迹，从而将其迁移至角色动画与真实机器人控制任务中。尽管WBC轨迹能够符合物理约束，但其结果可能与原始运动产生显著偏差。为解决此问题，本文提出PhysMoDPO——一种直接偏好优化框架。与以往依赖手工设计的物理感知启发式规则（如足部滑动惩罚）的方法不同，我们将WBC整合至训练流程中，并优化扩散模型，使得WBC的输出既符合物理规律，又忠实于原始文本指令。为训练PhysMoDPO，我们部署了基于物理和任务特定设计的奖励函数，并利用其对合成轨迹进行偏好标注。在文本到运动及空间控制任务上的大量实验表明，PhysMoDPO在仿真机器人的物理真实性与任务相关指标上均取得持续提升。此外，我们验证了PhysMoDPO在仿真环境中的零样本运动迁移任务以及G1人形机器人的实际部署中均能带来显著性能改进。

Recent progress in text-conditioned human motion generation has been largely driven by diffusion models trained on large-scale human motion data. Building on this progress, recent methods attempt to transfer such models for character animation and real robot control by applying a Whole-Body Controller (WBC) that converts diffusion-generated motions into executable trajectories. While WBC trajectories become compliant with physics, they may expose substantial deviations from original motion. To address this issue, we here propose PhysMoDPO, a Direct Preference Optimization framework. Unlike prior work that relies on hand-crafted physics-aware heuristics such as foot-sliding penalties, we integrate WBC into our training pipeline and optimize diffusion model such that the output of WBC becomes compliant both with physics and original text instructions. To train PhysMoDPO we deploy physics-based and task-specific rewards and use them to assign preference to synthesized trajectories. Our extensive experiments on text-to-motion and spatial control tasks demonstrate consistent improvements of PhysMoDPO in both physical realism and task-related metrics on simulated robots. Moreover, we demonstrate that PhysMoDPO results in significant improvements when applied to zero-shot motion transfer in simulation and for real-world deployment on a G1 humanoid robot.

评分理由: 论文主要研究视觉到代码任务中的强化学习奖励建模问题，核心贡献是Visual-ERM多模态奖励模型。与关键词的相关性分析如下：1）论文使用Qwen3-VL-8B-Instruct等大型视觉语言模型（LVLMs），与’Large Language Models’相关（8分）；2）涉及监督微调（SFT）和强化学习，与’Post-training/SFT’和’RLHF/DPO’相关（各8分）；3）提到通过反思和修订加强测试时扩展，与’Self-Correction/Self-Improvement’相关（8分）；4）奖励模型提供可解释反馈，与’Explainable AI’相关（8分）。其他关键词如MoE、SLMs、Scaling Laws、RAG、Agents等未在论文中涉及，得0分。

视觉到代码任务要求模型将结构化视觉输入（如图表、表格和SVG）重构为具有高视觉保真度的可执行或结构化表示。尽管近期的大型视觉语言模型通过监督微调取得了显著成果，但由于奖励信号失准，强化学习仍面临挑战。现有奖励机制要么依赖文本规则，要么依赖粗略的视觉嵌入相似度，两者均无法捕捉细粒度视觉差异且易受奖励破解影响。我们提出视觉等价奖励模型，这是一种多模态生成式奖励模型，可在渲染视觉空间中直接评估视觉到代码的质量，提供细粒度、可解释且与任务无关的反馈。该模型集成至强化学习后，将Qwen3-VL-8B-Instruct在图表到代码任务上的性能提升+8.4分，并在表格与SVG解析任务上实现稳定增益（平均提升+2.7和+4.1分），同时通过反思与修订机制进一步强化测试时扩展能力。我们还构建了VisualCritic-RewardBench基准，用于评估结构化视觉数据上细粒度的图像间差异判定能力。实验表明，8B参数的视觉等价奖励模型显著超越Qwen3-VL-235B-Instruct，并接近领先的闭源模型性能。我们的研究证明，无论任务特异性如何，细粒度视觉奖励监督对于视觉到代码的强化学习既是必要的也是充分的。

Vision-to-code tasks require models to reconstruct structured visual inputs, such as charts, tables, and SVGs, into executable or structured representations with high visual fidelity. While recent Large Vision Language Models (LVLMs) achieve strong results via supervised fine-tuning, reinforcement learning remains challenging due to misaligned reward signals. Existing rewards either rely on textual rules or coarse visual embedding similarity, both of which fail to capture fine-grained visual discrepancies and are vulnerable to reward hacking. We propose Visual Equivalence Reward Model (Visual-ERM), a multimodal generative reward model that provides fine-grained, interpretable, and task-agnostic feedback to evaluate vision-to-code quality directly in the rendered visual space. Integrated into RL, Visual-ERM improves Qwen3-VL-8B-Instruct by +8.4 on chart-to-code and yields consistent gains on table and SVG parsing (+2.7, +4.1 on average), and further strengthens test-time scaling via reflection and revision. We also introduce VisualCritic-RewardBench (VC-RewardBench), a benchmark for judging fine-grained image-to-image discrepancies on structured visual data, where Visual-ERM at 8B decisively outperforms Qwen3-VL-235B-Instruct and approaches leading closed-source models. Our results suggest that fine-grained visual reward supervision is both necessary and sufficient for vision-to-code RL, regardless of task specificity.

评分理由: 论文研究神经网络中的隐私保护方法，通过识别关键权重并仅重写这些权重进行微调，以抵御成员推理攻击。与大多数关键词无关，因为论文未涉及大模型、深度学习技术原理创新或科学领域应用。仅与’Post-training OR Supervised Fine-tuning OR SFT’和’PEFT OR LoRA OR Parameter-efficient Fine-tuning’有一定关联（5分），因为论文提到’fine-tuning’和权重调整，但未明确涉及大模型或参数高效微调技术。其他关键词均得0分，因论文专注于通用神经网络隐私，而非大模型特定技术或应用。

先前用于成员隐私保护的方法通常需要更新或重新训练神经网络中的所有权重，这不仅成本高昂，还可能导致不必要的效用损失，甚至加剧训练数据与非训练数据之间预测结果的不对齐。在本研究中，我们观察到三个关键发现：i) 隐私漏洞仅存在于极少部分权重中；ii) 然而，这些权重中的大多数对模型效用性能具有关键影响；iii) 权重的重要性源于其位置而非具体数值。基于这些发现，为保护隐私，我们对关键权重进行评分，并选择不丢弃这些神经元，而是仅对这些权重进行回退（rewind）以进行微调。通过大量实验证明，该机制在多数情况下能有效抵御成员推理攻击（Membership Inference Attacks），同时保持模型效用。

Prior approaches for membership privacy preservation usually update or retrain all weights in neural networks, which is costly and can lead to unnecessary utility loss or even more serious misalignment in predictions between training data and non-training data. In this work, we observed three insights: i) privacy vulnerability exists in a very small fraction of weights; ii) however, most of those weights also critically impact utility performance; iii) the importance of weights stems from their locations rather than their values. According to these insights, to preserve privacy, we score critical weights, and instead of discarding those neurons, we rewind only the weights for fine-tuning. We show that, through extensive experiments, this mechanism exhibits outperforming resilience in most cases against Membership Inference Attacks while maintaining utility.

评分理由: 论文核心研究LLM在多智能体系统中的治理框架，高度相关关键词包括：LLMs（论文明确研究LLM生成影响策略）、Alignment（研究伦理对齐和操纵风险）、LLM Agents（研究LLM与智能体交互）、Multi-agent Systems（在80个智能体的网络中进行实验）。其他关键词如MoE、SLMs、Scaling Laws等与论文内容完全无关，论文未涉及这些技术细节或应用领域。

大语言模型（LLM）能够生成具有说服力的影响策略，从而改变多智能体群体中的合作行为，但一个关键问题依然存在：由此产生的合作是否反映了真正的亲社会对齐，还是掩盖了智能体自主性、认知完整性及分配公平性的侵蚀？我们提出了宪法多智能体治理框架（Constitutional Multi-Agent Governance, CMAG），这是一个介于LLM策略编译器与网络化智能体群体之间的两阶段框架，它结合了硬约束过滤与软惩罚效用优化，以平衡合作潜力与操纵风险及自主性压力。我们提出了伦理合作分数（Ethical Cooperation Score, ECS），这是一个由合作性、自主性、完整性和公平性相乘构成的复合指标，对通过操纵手段实现的合作进行惩罚。在对抗性条件下（70%违规候选者）对80个智能体组成的无标度网络进行的实验中，我们评估了三种机制：完整CMAG、朴素过滤和无约束优化。虽然无约束优化获得了最高的原始合作度（0.873），但由于严重的自主性侵蚀（0.867）和公平性下降（0.888），其ECS最低（0.645）。CMAG实现了0.741的ECS，提升了14.9%，同时将自主性保持在0.985，完整性保持在0.995，仅将合作度适度降低至0.770。朴素消融实验（ECS = 0.733）证实仅靠硬约束是不够的。帕累托分析表明CMAG主导了合作-自主性权衡空间，并且治理将中心-边缘节点的暴露差异降低了60%以上。这些研究结果表明，缺乏治理的合作本身并非必然可取：宪法约束对于确保LLM介导的影响力产生伦理稳定的结果而非操纵性均衡是必要的。

Large Language Models (LLMs) can generate persuasive influence strategies that shift cooperative behavior in multi-agent populations, but a critical question remains: does the resulting cooperation reflect genuine prosocial alignment, or does it mask erosion of agent autonomy, epistemic integrity, and distributional fairness? We introduce Constitutional Multi-Agent Governance (CMAG), a two-stage framework that interposes between an LLM policy compiler and a networked agent population, combining hard constraint filtering with soft penalized-utility optimization that balances cooperation potential against manipulation risk and autonomy pressure. We propose the Ethical Cooperation Score (ECS), a multiplicative composite of cooperation, autonomy, integrity, and fairness that penalizes cooperation achieved through manipulative means. In experiments on scale-free networks of 80 agents under adversarial conditions (70% violating candidates), we benchmark three regimes: full CMAG, naive filtering, and unconstrained optimization. While unconstrained optimization achieves the highest raw cooperation (0.873), it yields the lowest ECS (0.645) due to severe autonomy erosion (0.867) and fairness degradation (0.888). CMAG attains an ECS of 0.741, a 14.9% improvement, while preserving autonomy at 0.985 and integrity at 0.995, with only modest cooperation reduction to 0.770. The naive ablation (ECS = 0.733) confirms that hard constraints alone are insufficient. Pareto analysis shows CMAG dominates the cooperation-autonomy trade-off space, and governance reduces hub-periphery exposure disparities by over 60%. These findings establish that cooperation is not inherently desirable without governance: constitutional constraints are necessary to ensure that LLM-mediated influence produces ethically stable outcomes rather than manipulative equilibria.

评分理由: 论文提出MXNorm方法，通过重用MXFP8块尺度来高效估计RMS，减少归一化计算量，在Llama 3模型预训练中验证了有效性。核心相关关键词：1) ‘Large Language Models’ (10分)：论文在Llama 3模型上验证方法；2) ‘Pre-training’ (10分)：在Llama 3预训练中评估；3) ‘Quantization’ (10分)：专注于MXFP8低精度格式；4) ‘Speculative Decoding’ (10分)：通过减少计算量加速推理。其他关键词如MoE、SFT、RAG等与论文内容无关。

矩阵乘法性能长期以来一直是扩展深度学习工作负载的主要瓶颈，这促使了使用日益低精度数值格式的新型加速器设计。然而，矩阵乘法性能的提升速度远远超过了归约运算和逐元素计算性能的提升，后者目前仍在使用更高精度进行计算。在本工作中，我们提出MXNorm，一种可即插即用的RMSNorm替代方案，它仅利用作为MXFP8转换过程一部分计算出的块尺度来估计均方根值，从而将归一化所需的归约运算规模减小32倍。我们在参数规模为125M、1B和8B的Llama 3模型预训练中验证了该近似方法，发现与使用MXFP8矩阵乘法的RMSNorm基线相比，训练精度损失极小。我们还展示了仅通过torch.compile实现的实用内核加速——MXNorm相比RMSNorm最高可达2.4倍，这对应着MXFP8格式下Llama 3 8B Transformer层1.3%的加速，以及在NVFP4格式下2.6%的加速。

Matrix multiplication performance has long been the major bottleneck to scaling deep learning workloads, which has stimulated the design of new accelerators that use increasingly low-precision number formats. However, improvements in matrix multiplication performance have far outstripped improvements in performance on reductions and elementwise computations, which are still being performed in higher precision. In this work, we propose MXNorm, a drop-in replacement for RMSNorm that estimates the RMS using only the block scales calculated as part of the MXFP8 cast and enables a 32x decrease in the size of reduction needed for normalization. We validate our approximation method on pre-training of Llama 3 models of 125M, 1B and 8B parameters, finding minimal loss of training accuracy compared to a baseline using RMSNorm with MXFP8 matmuls. We also show practical kernel speedups using only torch.compile of up to 2.4x for MXNorm over RMSNorm, corresponding to a 1.3% speedup in Llama 3 8B transformer layers in MXFP8 and a 2.6% speedup in NVFP4.

评分理由: 该论文专注于使用机器学习方法（MiniRocket特征提取、降维和聚类）分析土星卫星系统的轨道动力学，属于AI在科学领域的应用。论文未涉及任何大模型（LLM）、深度学习技术原理、训练方法、推理优化、对齐技术、代理系统等关键词。唯一相关的关键词是’AI for Science OR Bioinformatics OR Cheminformatics’，因为论文将机器学习应用于天文学研究，属于AI for Science范畴，但并非核心创新点，只是传统机器学习应用，因此给予5分（有一定关联）。其他所有关键词均与论文内容完全无关，评分为0分。

土星卫星系统的动力学为研究轨道稳定性与共振相互作用提供了一个丰富的框架。分析此类系统的传统方法（包括傅里叶分析和稳定性度量指标）在处理现代数据集的规模和复杂性方面面临困难。本研究引入了一种基于机器学习的流程，用于对大约22,300条模拟卫星轨道进行聚类，通过先进的特征提取和降维技术应对这些挑战。该方法的关键在于使用MiniRocket算法，它能将400个时间步高效地转换为9,996维的特征空间，从而捕捉复杂的时间模式。额外的自动化特征提取与降维技术进一步优化了数据，实现了稳健的聚类分析。该流程揭示了土星卫星系统中的稳定区域、共振结构及其他关键行为，为其长期动力学演化提供了新的见解。通过将计算工具与传统天体力学技术相结合，本研究为分析大规模轨道数据集和推进行星动力学探索，提供了一种可扩展且可解释的方法论。

The dynamics of Saturn’s satellite system offer a rich framework for studying orbital stability and resonance interactions. Traditional methods for analysing such systems, including Fourier analysis and stability metrics, struggle with the scale and complexity of modern datasets. This study introduces a machine learning-based pipeline for clustering approximately 22,300 simulated satellite orbits, addressing these challenges with advanced feature extraction and dimensionality reduction techniques. The key to this approach is using MiniRocket, which efficiently transforms 400 timesteps into a 9,996-dimensional feature space, capturing intricate temporal patterns. Additional automated feature extraction and dimensionality reduction techniques refine the data, enabling robust clustering analysis. This pipeline reveals stability regions, resonance structures, and other key behaviours in Saturn’s satellite system, providing new insights into their long-term dynamical evolution. By integrating computational tools with traditional celestial mechanics techniques, this study offers a scalable and interpretable methodology for analysing large-scale orbital datasets and advancing the exploration of planetary dynamics.

评分理由: 论文核心研究LLM作为自主推理代理在科学问题解决中的语义不变性，与’Large Language Models’、‘Chain of Thought’、‘System 2 Thinking’、‘LLM Agents’和’AI for Science’高度相关（10分）。涉及多代理协调和可靠性评估，与’Multi-agent Systems’、‘Hallucination Mitigation’、‘Mechanistic Interpretability’有一定关联（5分）。其他关键词如MoE、量化、训练方法等未在摘要中提及，评为0分。

大型语言模型（LLM）日益在决策支持、科学问题求解与多智能体协同系统中扮演自主推理智能体的角色。然而，在关键应用中部署LLM智能体时，必须确保其推理在语义等价的输入变化下保持稳定——这一特性我们称之为语义不变性。现有的标准基准评估仅针对固定、规范的问题表述进行准确性测试，未能捕捉这一关键可靠性维度。为弥补此不足，本文提出一种蜕变测试框架，用于系统评估LLM推理智能体的鲁棒性。该框架在涵盖四种不同架构家族的七个基础模型上应用了八种语义保持变换（恒等变换、复述变换、事实重排序、扩展变换、压缩变换、学术语境转换、商业语境转换及对比式表述），这些模型包括：Hermes（70B, 405B）、Qwen3（30B-A3B, 235B-A22B）、DeepSeek-R1以及gpt-oss（20B, 120B）。我们的评估覆盖八个科学领域的19个多步推理问题。结果表明，模型规模并不能预测鲁棒性：较小的Qwen3-30B-A3B实现了最高的稳定性（79.6%的不变响应率，语义相似度0.91），而更大规模的模型反而表现出更强的脆弱性。

Large Language Models (LLMs) increasingly serve as autonomous reasoning agents in decision support, scientific problem-solving, and multi-agent coordination systems. However, deploying LLM agents in consequential applications requires assurance that their reasoning remains stable under semantically equivalent input variations, a property we term semantic invariance.Standard benchmark evaluations, which assess accuracy on fixed, canonical problem formulations, fail to capture this critical reliability dimension. To address this shortcoming, in this paper we present a metamorphic testing framework for systematically assessing the robustness of LLM reasoning agents, applying eight semantic-preserving transformations (identity, paraphrase, fact reordering, expansion, contraction, academic context, business context, and contrastive formulation) across seven foundation models spanning four distinct architectural families: Hermes (70B, 405B), Qwen3 (30B-A3B, 235B-A22B), DeepSeek-R1, and gpt-oss (20B, 120B). Our evaluation encompasses 19 multi-step reasoning problems across eight scientific domains. The results reveal that model scale does not predict robustness: the smaller Qwen3-30B-A3B achieves the highest stability (79.6% invariant responses, semantic similarity 0.91), while larger models exhibit greater fragility.

评分理由: 论文核心是开发用于印度孕产妇健康的聊天机器人，该系统结合了阶段感知分类、混合检索和基于LLM的证据条件生成。与关键词高度相关的是：1) ‘Large Language Models’（系统使用LLM进行证据条件生成，是核心组件）；2) ‘Retrieval-Augmented Generation’（系统采用混合检索和LLM生成，是RAG的典型应用）；3) ‘AI for Science’（应用于生物信息学/医疗健康领域，属于AI for Science范畴）。‘Hallucination Mitigation’得5分，因为论文关注可信医疗信息生成，涉及事实性和安全性，但未明确讨论幻觉缓解技术。其他关键词与论文内容无直接关联，论文未涉及MoE、SLMs、缩放定律、训练技术、推理优化、代理系统等主题。

利用电话聊天机器人提供可信赖的孕产健康信息的能力可能产生显著影响，这在用户健康素养较低且获得医疗服务机会有限的资源匮乏环境中尤为突出。然而，部署此类系统在技术上具有挑战性：用户查询通常简短、信息不完整，且存在跨语言的语码混合现象；回答需要结合地区特定的背景知识；而部分或缺失的症状描述使得安全的分诊决策变得困难。

我们介绍一款为印度孕产健康开发的聊天机器人，它由学术研究人员、一家健康科技公司、一个公共卫生非营利组织以及一家医院合作开发。该系统整合了以下组件：(1) 阶段感知分诊，将高风险查询路由至专家设计的模板；(2) 对经过整理的孕产/新生儿指南进行混合检索；(3) 基于大型语言模型（LLM）的证据条件生成。我们的核心贡献是在有限专家监督下，为高风险场景部署设计了一套评估工作流程。针对组件级和端到端测试，我们引入了：(i) 一个带标注的分诊基准数据集（N=150），实现了86.7%的紧急情况召回率，并明确报告了漏报紧急情况与过度升级之间的权衡；(ii) 一个包含分块级证据标签的合成多证据检索基准（N=100）；(iii) 使用临床医生共同设计的标准，对真实查询（N=781）进行LLM即评判员的比较评估；(iv) 专家验证。我们的研究结果表明，在多语言、高噪声环境下构建可信赖的医疗助手，需要采用深度防御设计并结合多方法评估，而非依赖单一模型或评估方法的选择。

The ability to provide trustworthy maternal health information using phone-based chatbots can have a significant impact, particularly in low-resource settings where users have low health literacy and limited access to care. However, deploying such systems is technically challenging: user queries are short, underspecified, and code-mixed across languages, answers require regional context-specific grounding, and partial or missing symptom context makes safe routing decisions difficult. We present a chatbot for maternal health in India developed through a partnership between academic researchers, a health tech company, a public health nonprofit, and a hospital. The system combines (1) stage-aware triage, routing high-risk queries to expert templates, (2) hybrid retrieval over curated maternal/newborn guidelines, and (3) evidence-conditioned generation from an LLM. Our core contribution is an evaluation workflow for high-stakes deployment under limited expert supervision. Targeting both component-level and end-to-end testing, we introduce: (i) a labeled triage benchmark (N=150) achieving 86.7% emergency recall, explicitly reporting the missed-emergency vs. over-escalation trade-off; (ii) a synthetic multi-evidence retrieval benchmark (N=100) with chunk-level evidence labels; (iii) LLM-as-judge comparison on real queries (N=781) using clinician-codesigned criteria; and (iv) expert validation. Our findings show that trustworthy medical assistants in multilingual, noisy settings require defense-in-depth design paired with multi-method evaluation, rather than any single model and evaluation method choice.

评分理由: 论文核心研究大语言模型（LLMs）在ESG报告分析中的幻觉缓解问题，因此与’Large Language Models’高度相关（10分）。论文明确使用Chain-of-Thought（CoT）提示策略，因此与’Chain of Thought’高度相关（10分）。论文主要研究幻觉缓解，因此与’Hallucination Mitigation’高度相关（10分）。论文涉及ESG领域应用，与’AI for Science’有一定关联（5分）。论文处理长文档ESG报告，与’Long Context LLMs’有一定关联（5分）。论文提到微调LLMs，与’Post-training’有一定关联（5分）。其他关键词如MoE、SLMs、Scaling Laws、RLHF、RAG、Quantization等未在论文中涉及，评分为0分。

随着企业责任日益纳入环境、社会和治理（ESG）标准，ESG报告正逐渐成为许多地区的法定要求，也成为记录可持续发展实践和评估企业长期与伦理绩效的关键渠道。然而，ESG披露内容的篇幅和复杂性使其难以被可靠地解读或实现自动化分析。为支持可扩展且可信的分析，本文提出了ESG-Bench——一个用于大语言模型（LLM）理解ESG报告并缓解幻觉现象的基准数据集。ESG-Bench包含基于真实ESG报告语境的人工标注问答对，其细粒度标签可指示模型输出是否得到事实支持或存在幻觉。通过将ESG报告分析构建为具有可验证性约束的问答任务，本研究系统评估了LLM提取和推理ESG内容的能力，并提供了一个新的应用场景：在社会敏感、合规关键的环境中缓解幻觉问题。我们设计了针对特定任务的思维链（CoT）提示策略，并利用带有CoT标注推理过程的数据对多个前沿LLM进行微调。实验表明，这些基于CoT的方法在减少幻觉方面显著优于标准提示和直接微调，且其优势可迁移至ESG领域之外的现有问答基准。

As corporate responsibility increasingly incorporates environmental, social, and governance (ESG) criteria, ESG reporting is becoming a legal requirement in many regions and a key channel for documenting sustainability practices and assessing firms’ long-term and ethical performance. However, the length and complexity of ESG disclosures make them difficult to interpret and automate the analysis reliably. To support scalable and trustworthy analysis, this paper introduces ESG-Bench, a benchmark dataset for ESG report understanding and hallucination mitigation in large language models (LLMs). ESG-Bench contains human-annotated question-answer (QA) pairs grounded in real-world ESG report contexts, with fine-grained labels indicating whether model outputs are factually supported or hallucinated. Framing ESG report analysis as a QA task with verifiability constraints enables systematic evaluation of LLMs’ ability to extract and reason over ESG content and provides a new use case: mitigating hallucinations in socially sensitive, compliance-critical settings. We design task-specific Chain-of-Thought (CoT) prompting strategies and fine-tune multiple state-of-the-art LLMs on ESG-Bench using CoT-annotated rationales. Our experiments show that these CoT-based methods substantially outperform standard prompting and direct fine-tuning in reducing hallucinations, and that the gains transfer to existing QA benchmarks beyond the ESG domain.

评分理由: 论文提出了一种改进GRPO（Group Relative Policy Optimization）的方法，属于强化学习对齐技术，与RLHF/DPO高度相关（10分）。论文专注于推理模型的训练，与Chain of Thought推理高度相关（10分），并涉及深度推理过程（8分）。论文提到大模型在数学推理中的应用，与LLMs相关（8分）。论文的对比学习机制涉及自我改进概念（5分）。其他关键词如MoE、量化、RAG等与论文内容无关（0分）。

群体相对策略优化（GRPO）已成为训练推理模型的有效方法。尽管其基于群体均值计算优势函数，但GRPO在优化过程中将每个输出视为独立样本，忽略了一个关键的结构性信号：同一群体内正确与错误解答之间的天然对比性，从而未能充分利用通过显式对比成功与失败推理轨迹可获得的丰富比较数据。为利用这一潜力，我们提出了GRPO的对比式重构，证明GRPO目标函数隐式地最大化正确与错误样本的策略比率之间的边际。基于此洞见，我们提出双边上下文条件化（Bilateral Context Conditioning, BICC）机制，使模型在优化过程中能够交叉参考成功与失败的推理轨迹，实现跨样本的直接信息流动。我们进一步引入奖励-置信度校正（Reward-Confidence Correction, RCC），通过基于方差最小估计量的一阶近似推导出的奖励-置信度协方差，动态调整GRPO中的优势基线以稳定训练。两种机制均无需额外采样或辅助模型，并可适配所有GRPO变体。在数学推理基准测试上的实验表明，该方法在多种模型与算法中均取得了一致的性能提升。代码发布于 \href{https://github.com/Skylanding/BiCC}{https://github.com/Skylanding/BiCC}。

Group Relative Policy Optimization (GRPO) has emerged as an effective method for training reasoning models. While it computes advantages based on group mean, GRPO treats each output as an independent sample during the optimization and overlooks a vital structural signal: the natural contrast between correct and incorrect solutions within the same group, thus ignoring the rich, comparative data that could be leveraged by explicitly pitting successful reasoning traces against failed ones. To capitalize on this, we present a contrastive reformulation of GRPO, showing that the GRPO objective implicitly maximizes the margin between the policy ratios of correct and incorrect samples. Building on this insight, we propose Bilateral Context Conditioning (BICC), a mechanism that allows the model to cross-reference successful and failed reasoning traces during the optimization, enabling a direct information flow across samples. We further introduce Reward-Confidence Correction (RCC) to stabilize training by dynamically adjusts the advantage baseline in GRPO using reward-confidence covariance derived from the first-order approximation of the variance-minimizing estimator. Both mechanisms require no additional sampling or auxiliary models and can be adapted to all GRPO variants. Experiments on mathematical reasoning benchmarks demonstrate consistent improvements across comprehensive models and algorithms. Code is available at \href{https://github.com/Skylanding/BiCC}{https://github.com/Skylanding/BiCC}.

评分理由: 论文提出了一种名为Steve-Evolving的开放世界具身智能体自进化框架，核心是LLM规划器与经验诊断、知识蒸馏的闭环集成。因此，与’Large Language Models’高度相关（LLM作为规划器），与’LLM Agents’高度相关（具身智能体应用），与’Self-Correction’高度相关（通过诊断和闭环控制实现自我进化）。其他关键词如MoE、SFT、RAG、量化等未在摘要中提及，故评分为0。

开放世界的具身智能体需解决长周期任务，其核心瓶颈并非单步规划质量，而是交互经验的组织与演进方式。为此，我们提出Steve-Evolving——一种非参数化的自演进框架，通过闭环机制将细粒度执行诊断与双轨知识蒸馏紧密耦合。该方法包含三个阶段：经验锚定、经验蒸馏与知识驱动的闭环控制。具体而言，经验锚定将每个子目标尝试固化为具有固定模式（前置状态、动作、诊断结果与后置状态）的结构化经验元组，并通过多维索引（如条件特征、空间哈希与语义标签）及滚动摘要将其组织至三层经验空间中，以实现高效且可追溯的检索。为确保归因所需的信息密度，执行层提供超越二元结果的组合式诊断信号，包括状态差异摘要、枚举式失败原因、连续指标及停滞/循环检测。此外，经验蒸馏阶段将成功轨迹泛化为具有明确前置条件与验证标准的可复用技能，同时将失败案例提炼为可执行的防护规则，这些规则能捕捉根本原因并在子目标与任务粒度上禁止风险操作。在知识驱动的闭环控制中，检索到的技能与防护规则被注入大语言模型（LLM）规划器，而诊断触发的局部重规划会在线更新动态约束，形成无需更新模型参数的持续演进过程。在《我的世界》MCU长周期任务套件上的实验表明，该方法相较于静态检索基线取得了持续的性能提升。

Open-world embodied agents must solve long-horizon tasks where the main bottleneck is not single-step planning quality but how interaction experience is organized and evolved. To this end, we present Steve-Evolving, a non-parametric self-evolving framework that tightly couples fine-grained execution diagnosis with dual-track knowledge distillation in a closed loop. The method follows three phases: Experience Anchoring, Experience Distillation, and Knowledge-Driven Closed-Loop Control. In detail, Experience Anchoring solidifies each subgoal attempt into a structured experience tuple with a fixed schema (pre-state, action, diagnosis-result, and post-state) and organizes it in a three-tier experience space with multi-dimensional indices (e.g., condition signatures, spatial hashing, and semantic tags) plus rolling summarization for efficient and auditable recall. To ensure sufficient information density for attribution, the execution layer provides compositional diagnosis signals beyond binary outcomes, including state-difference summaries, enumerated failure causes, continuous indicators, and stagnation/loop detection. Moreover, successful trajectories of Experience Distillation are generalized into reusable skills with explicit preconditions and verification criteria, while failures are distilled into executable guardrails that capture root causes and forbid risky operations at both subgoal and task granularities. Besides, Knowledge-Driven Closed-Loop Control retrieved skills and guardrails are injected into an LLM planner, and diagnosis-triggered local replanning updates the active constraints online, forming a continual evolution process without any model parameter updates. Experiments on the long-horizon suite of Minecraft MCU demonstrate consistent improvements over static-retrieval baselines.

评分理由: 论文核心是开发一个用于评估大语言模型（LLMs）的PsyCogMetrics AI平台，因此与’Large Language Models OR LLMs OR Foundation Models’高度相关（10分）。论文涉及AI在科学（认知科学）中的应用，与’AI for Science OR Bioinformatics OR Cheminformatics’有一定关联（5分）。其他关键词主要涉及大模型的技术原理、训练方法、优化技术或特定应用领域，而本文专注于评估平台开发和方法论，未涉及这些具体技术，因此评分为0分。

本研究介绍了PsyCogMetrics AI实验室平台（psycogmetrics.ai）的开发，这是一个集成的云端平台，旨在将心理测量学与认知科学方法应用于大语言模型（LLM）评估。研究采用三循环行动设计科学框架：在关联性循环中，识别了当前评估方法的关键局限性与未满足的利益相关者需求；严谨性循环借鉴了波普尔可证伪性、经典测验理论和认知负荷理论等核心理论，推导出演绎性设计目标；设计循环则通过嵌套的“构建-干预-评估”迭代过程，将这些目标具体实施。本研究贡献了一个新颖的信息技术制品——一套经过验证的大语言模型评估设计方案，有助于推动人工智能、心理学、认知科学以及社会与行为科学交叉领域的研究。

This study presents the development of the PsyCogMetrics AI Lab (psycogmetrics.ai), an integrated, cloud-based platform that operationalizes psychometric and cognitive-science methodologies for Large Language Model (LLM) evaluation. Framed as a three-cycle Action Design Science study, the Relevance Cycle identifies key limitations in current evaluation methods and unfulfilled stakeholder needs. The Rigor Cycle draws on kernel theories such as Popperian falsifiability, Classical Test Theory, and Cognitive Load Theory to derive deductive design objectives. The Design Cycle operationalizes these objectives through nested Build-Intervene-Evaluate loops. The study contributes a novel IT artifact, a validated design for LLM evaluation, benefiting research at the intersection of AI, psychology, cognitive science, and the social and behavioral sciences.

评分理由: 论文主要研究视频语言模型（VideoLLMs）中的相机运动理解，属于大模型（VideoLLMs）在视觉领域的应用，与’Large Language Models OR LLMs OR Foundation Models’有一定相关性（8分），因为VideoLLMs是视觉-语言基础模型的一种。论文涉及几何信号提取、数据集构建和轻量级注入方法，但未深入探讨其他关键词如MoE、SLMs、训练技术、推理优化、代理系统等具体技术，因此这些关键词评分为0分。

相机运动是塑造视觉感知与电影风格的基础几何信号，然而当前具备视频处理能力的视觉语言模型（VideoLLMs）很少显式表征相机运动，且常在细粒度运动基元上表现不佳。我们通过一个包含基准测试、诊断与注入的框架来弥补这一不足。我们构建了CameraMotionDataset——一个具有显式相机控制的大规模合成数据集，将相机运动建模为约束感知的多标签识别任务，并创建了一个视觉问答基准——CameraMotionVQA。在对多种现有VideoLLMs的测试中，我们观察到其在识别相机运动基元时存在显著错误。对Qwen2.5-VL视觉编码器的探测实验表明，相机运动线索的表征较弱，尤其在更深的ViT模块中，这有助于解释观察到的失败模式。为了在不进行昂贵训练或微调的情况下弥补这一缺陷，我们提出了一种轻量级、模型无关的流程：从三维基础模型（3DFMs）中提取几何相机线索，通过时序分类器预测受约束的运动基元，并通过结构化提示将其注入下游VideoLLM的推理过程。实验证明，该方法提升了运动识别能力，并生成了更具相机感知的模型响应，凸显了几何驱动的线索提取与结构化提示作为实现相机感知VideoLLM和视觉语言智能系统（VLA）的实用步骤。数据集与基准已公开于https://hf.co/datasets/fengyee/camera-motion-dataset-and-benchmark。

Camera motion is a fundamental geometric signal that shapes visual perception and cinematic style, yet current video-capable vision-language models (VideoLLMs) rarely represent it explicitly and often fail on fine-grained motion primitives. We address this gap with a framework of $\textbf{benchmarking}$, $\textbf{diagnosis}$, and $\textbf{injection}$. We curate $\textbf{CameraMotionDataset}$, a large-scale synthetic dataset with explicit camera control, formulate camera motion as constraint-aware multi-label recognition, and construct a VQA benchmark–$\textbf{CameraMotionVQA}$. Across diverse off-the-shelf VideoLLMs, we observe substantial errors in recognizing camera motion primitives. Probing experiments on a Qwen2.5-VL vision encoder suggest that camera motion cues are weakly represented, especially in deeper ViT blocks, helping explain the observed failure modes. To bridge this gap without costly training or fine-tuning, we propose a lightweight, model-agnostic pipeline that extracts geometric camera cues from 3D foundation models (3DFMs), predicts constrained motion primitives with a temporal classifier, and injects them into downstream VideoLLM inference via structured prompting. Experiments demonstrate improved motion recognition and more camera-aware model responses, highlighting geometry-driven cue extraction and structured prompting as practical steps toward a camera-aware VideoLLM and VLA system. The dataset and benchmark is publicly available at https://hf.co/datasets/fengyee/camera-motion-dataset-and-benchmark.

评分理由: 论文研究主动学习（Active Learning）中的样本选择策略，提出了一种名为BoSS的可扩展Oracle策略，用于大规模数据集和深度神经网络场景。虽然摘要中提到’foundation models’，但这是作为背景提及（这些模型使识别有价值实例更容易），并非论文的核心研究内容。论文的核心是主动学习策略的评估和改进，特别是Oracle策略的设计和集成方法。所有评分关键词都直接针对大模型/深度学习的技术原理、训练方法、推理优化、对齐、应用等具体方面，而本文专注于传统的主动学习框架和策略评估，没有涉及任何评分关键词中的具体技术。因此，所有关键词的相关度均为0。

主动学习（Active Learning, AL）旨在通过迭代选择有价值的样本，在最大化模型性能的同时降低标注成本。尽管基础模型使得识别这些样本变得更加容易，但现有的选择策略在不同模型、标注预算和数据集之间仍缺乏鲁棒性。为了揭示现有AL策略的潜在缺陷并为研究提供参考基准，我们探索了预言家策略（oracle strategies），即通过访问实际AL场景中无法获得的真实标注信息来逼近最优选择的策略。然而，当前的预言家策略难以有效扩展到大规模数据集和复杂的深度神经网络。为应对这些局限性，我们提出了最佳策略选择器（Best-of-Strategy Selector, BoSS），这是一种专为大规模AL场景设计的可扩展预言家策略。BoSS通过集成多种选择策略构建候选批次集合，随后选择能带来最高性能增益的批次。作为一种策略集成方法，BoSS能够轻松纳入未来出现的新前沿策略，从而确保其持续作为可靠的预言家策略。我们的评估表明：i) BoSS优于现有预言家策略；ii) 当前前沿的AL策略仍明显落后于预言家性能，尤其在具有多类别的大规模数据集中；iii) 应对AL策略性能不稳定的一个可行解决方案可能是采用基于集成的方法进行样本选择。

Active learning (AL) aims to reduce annotation costs while maximizing model performance by iteratively selecting valuable instances. While foundation models have made it easier to identify these instances, existing selection strategies still lack robustness across different models, annotation budgets, and datasets. To highlight the potential weaknesses of existing AL strategies and provide a reference point for research, we explore oracle strategies, i.e., strategies that approximate the optimal selection by accessing ground-truth information unavailable in practical AL scenarios. Current oracle strategies, however, fail to scale effectively to large datasets and complex deep neural networks. To tackle these limitations, we introduce the Best-of-Strategy Selector (BoSS), a scalable oracle strategy designed for large-scale AL scenarios. BoSS constructs a set of candidate batches through an ensemble of selection strategies and then selects the batch yielding the highest performance gain. As an ensemble of selection strategies, BoSS can be easily extended with new state-of-the-art strategies as they emerge, ensuring it remains a reliable oracle strategy in the future. Our evaluation demonstrates that i) BoSS outperforms existing oracle strategies, ii) state-of-the-art AL strategies still fall noticeably short of oracle performance, especially in large-scale datasets with many classes, and iii) one possible solution to counteract the inconsistent performance of AL strategies might be to employ an ensemble-based approach for the selection.

评分理由: 论文评估了四种最先进的视觉语言模型（VLMs）在机器人运动空间推理方面的能力，属于大模型在机器人领域的应用研究。与关键词的相关性分析如下：1）‘Large Language Models OR LLMs OR Foundation Models’（8分）：论文明确研究视觉语言模型（VLMs），这是基础模型的一种，属于大模型范畴，但论文聚焦于视觉-语言任务而非纯语言模型。2）‘Post-training OR Supervised Fine-tuning OR SFT’（5分）：论文提到在较小模型上进行了微调（fine-tuning），这属于后训练技术，但并非核心内容。3）‘LLM Agents OR Autonomous Agents OR Agentic Workflow’（8分）：论文研究机器人运动规划，涉及智能体系统，与LLM智能体概念相关。其他关键词如MoE、Scaling Laws、RLHF、RAG等均未在论文中涉及，因此评分为0。论文未涉及生物信息学等科学AI应用。

理解用户指令与周围环境中物体的空间关系，对于智能机器人系统协助人类完成多样化任务至关重要。视觉语言模型（Vision-Language Models, VLMs）的自然语言理解与空间推理能力，有望提升机器人规划器在新任务、新物体及新运动规范上的泛化性能。尽管基础模型已被应用于任务规划，但其是否具备足够的空间推理能力来满足用户对运动的偏好或约束——例如与物体的期望距离、拓扑属性或运动风格偏好——仍不明确。本文通过四种不同的查询方法，评估了四种前沿视觉语言模型在机器人运动空间推理方面的能力。实验结果表明，在性能最佳的查询方法下，Qwen2.5-VL 实现了 71.4% 的零样本准确率，经微调后的小型模型准确率达 75%，而 GPT-4o 的表现则相对较低。我们评估了两种运动偏好类型（物体接近度与路径风格），并分析了准确率与计算成本（以 token 数量衡量）之间的权衡。这项工作揭示了视觉语言模型与机器人运动规划流程结合的潜在前景。

Understanding user instructions and object spatial relations in surrounding environments is crucial for intelligent robot systems to assist humans in various tasks. The natural language and spatial reasoning capabilities of Vision-Language Models (VLMs) have the potential to enhance the generalization of robot planners on new tasks, objects, and motion specifications. While foundation models have been applied to task planning, it is still unclear the degree to which they have the capability of spatial reasoning required to enforce user preferences or constraints on motion, such as desired distances from objects, topological properties, or motion style preferences. In this paper, we evaluate the capability of four state-of-the-art VLMs at spatial reasoning over robot motion, using four different querying methods. Our results show that, with the highest-performing querying method, Qwen2.5-VL achieves 71.4% accuracy zero-shot and 75% on a smaller model after fine-tuning, and GPT-4o leads to lower performance. We evaluate two types of motion preferences (object-proximity and path-style), and we also analyze the trade-off between accuracy and computation cost in number of tokens. This work shows some promise in the potential of VLM integration with robot motion planning pipelines.