论文

本文将 S2G-RAG 的结构化充分性-缺口判断适配到冻结的 Search-R1 流程中，并在来自 900 个不相交 HotpotQA 问题的 3009 个状态上训练了一个 Qwen3.5-2B judge，以减少检索次数同时广泛保持答案准确性。This work adapts S2G-RAG's structured sufficiency-and-gap judgment to a frozen Search-R1 pipeline and trains a Qwen3.5-2B judge on 3,009 states from 900 disjoint HotpotQA questions to reduce retrieval while broadly preserving answer accuracy.

本文提出 Syfer，一种用于多语言多跳问答的 synthesizer-folding 框架，默认推迟翻译而非直接应用翻译，在保持具有竞争力准确性的同时，在性能与计算成本之间取得良好平衡。The method Syfer is introduced, a synthesizer-folding framework for multilingual multi-hop question answering that defers translation rather than applying it by default and attains competitive accuracy while striking a favourable balance between performance and computational cost.

联合部署可将攻击成功率从 67.4% 降至 14.0%，同时在未投毒检索上保留 41.3% 的 F1，证明无需投毒标签或可信语料库即可在语料入库和查询时提供实用的双重保护。Joint deployment reduces attack success from 67.4% to 14.0% while retaining 41.3% F1 on unpoisoned retrieval, demonstrating practical protection at both corpus ingestion and query time without poison labels or trusted corpora.

本文提出 Reduced Matrix Multiplication，一种免训练的输入自适应推理方法，通过沿收缩维度选取信息性切片来减少 Transformer 矩阵乘积，且不修改模型权重，并表明同一原理可扩展到多模态视觉-语言推理。Reduced Matrix Multiplication is proposed, a training-free, input-adaptive inference method that reduces Transformer matrix products by selecting informative slices along their contraction dimensions, without modifying model weights, and it is shown that the same principle extends to multimodal vision-language inference.

提出 Agentic Video Auto-Encoder（AVA-Encoder），一种由 agentic 自我进化驱动的新型自编码框架，用于学习 agent-native 视频表示，在 shot-level 和 keyframe-level system-prompt token 使用量减少 74.3% 的同时，性能优于精心人工调优的策略。The Agentic Video Auto-Encoder (AVA-Encoder), a novel auto-encoding framework driven by agentic self-evolution to learn agent-native video representations that outperforms a carefully human-tuned policy while using 74.3% fewer shot-level and keyframe-level system-prompt tokens.

论文表明，使用 Stanford Natural Language Inference 数据集有监督训练的通用句子表示，在广泛的迁移任务上能持续优于 SkipThought vectors 等无监督方法。It is shown how universal sentence representations trained using the supervised data of the Stanford Natural Language Inference datasets can consistently outperform unsupervised methods like SkipThought vectors on a wide range of transfer tasks.

提出 Context-Matched Distillation（CMD），一种因果 DMD 框架，将 teacher 监督信号与每个 target 生成时可用信息对齐，并可自然扩展到逐帧和逐 chunk 生成、长视频蒸馏以及相机条件蒸馏。Context-Matched Distillation (CMD) is introduced, a causal DMD framework that aligns teacher supervision with the information available when each target is generated, and naturally extends to frame-wise and chunk-wise generation, long video distillation, and camera-conditioned distillation.

提出 HPSE，构建一种混合 rollout，能够在 student 自身轨迹中 coverage 缺失的精确位置补齐缺失事实，同时在其他位置保持 on-policy。HPSE is proposed, which builds a hybrid rollout that steps in to place missing facts onto the student's own trajectory precisely where its coverage fails, while staying on-policy elsewhere.

本文报告了一项完整的、有完整记录的案例研究：在规范优先协议下，由 AI 编码 Agent 对大规模架构进行重构，期间无人工代码审查、无预先存在的预言机来验证目标行为。该任务是在一个大型相互依赖的代码库中拆除核心不变量，作者评估认为通过增量重构基本上不可行，这类变更通常需要重写。本文所述协议下，Agent 成功完成了任务。该系统包含 717,725 行This paper reports a single, fully instrumented case study of a large-scale architectural refactoring by an AI coding agent under a specification-first protocol, with no human review of the generated code and no pre-existing oracle to validate the target behaviour. The task, dismantling a central invariant across a large interdependent codebase, was assessed by the author as effectively infeasible through incremental refactoring, the kind of change that conventionally calls for a rewrite instead. Under the protocol described here, the agent completed it successfully. The system is a 717,725-li

Gambit 通过周期性剪除低潜力轨迹并立即从高质量前缀分支，利用轻量 scorer 探测 hidden state，将算力动态集中到最有潜力的推理路径上，同时保持持续的高硬件利用率。By periodically pruning unpromising trajectories and immediately branching from high-quality prefixes, Gambit dynamically concentrates compute onto the most promising reasoning traces via a light-weight scorer probing hidden states while maintaining continuous high hardware utilization.

提出 Intermittent Low-Frequency Lockout（ILL），一种基于通用波形模板在黑盒设置下评估该风险的不可听 red teaming 方法；同时提出 Distributional Requery Guard（DRG），用于缓解该风险。This paper proposes Intermittent Low-Frequency Lockout (ILL), an inaudible red teaming method that evaluates this risk using a universal waveform template in a black box setting and proposes Distributional Requery Guard (DRG), an inaudible red teaming method to mitigate this risk.

一种具有固定大小记忆的循环 Transformer 架构，可泛化 sliding-window attention，同时在训练期间保持并行性，并在验证 loss 和下游预训练 benchmark 上优于 sliding-window 和 latent recurrent Transformer 基线。A recurrent Transformer architecture with fixed-size memory that generalizes sliding-window attention while remaining parallelizable during training and improves validation loss and downstream pretraining benchmarks over sliding-window and latent recurrent transformer baselines.

本文提出一种权重裁剪的替代方案：对 critic 相对于其输入的梯度范数施加惩罚。其性能优于标准 WGAN，能以几乎无需调参的方式稳定训练多种 GAN 架构。This work proposes an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input, which performs better than standard WGAN and enables stable training of a wide variety of GAN architectures with almost no hyperparameter tuning.

当将监督规模扩展到 680,000 小时的多语言、多任务数据时，所得到的模型在标准 benchmark 上泛化良好，在 zero-shot transfer 设置下常可与此前全监督方法的结果相当，且无需任何微调。When scaled to 680,000 hours of multilingual and multitask supervision, the resulting models generalize well to standard benchmarks and are often competitive with prior fully supervised results but in a zero-shot transfer setting without the need for any fine-tuning.

建立一个可复现的选择性 3D 定位框架，并指出跨视角对应（cross-view correspondence）是主要操作瓶颈。The study establishes a reproducible framework for selective 3D localization and identifies cross-view correspondence as the dominant operational bottleneck.

首个同时使用 LLM 推理和 tag-aware 翻译来明确处理并评估英罗 MT 中性别偏见的方法。This is the first method to explicitly address and evaluate gender bias in English-Romanian MT using both LLM inference and tag-aware translation.

对八个法律 RAG 系统在 GDPR 和某国国内民法两个法律语料上的幻觉行为进行细粒度分析，发现含有错误假设的 false-premise 问题在人工编写问题上产生高幻觉率。A fine-grained analysis of hallucination behavior in eight legal RAG systems across two legal corpora, the GDPR and a national civil law, finds that false-premise questions, containing incorrect assumptions that must be rejected, produce high hallucination rates on the manually-drafted questions.

基于 CPI-Bench 对主流图像编辑模型的评测结果显示，CPI-Bench 增强了模型间的性能区分度；排名分析表明 CPI-Bench 与 Arena Image Edit Leaderboard 的对齐度最高，与公开人类偏好排名具有更强的一致性。Evaluation results of mainstream image editing models based on CPI-Bench demonstrate that CPI-Bench enhances performance differentiation among models, and ranking analysis reveals that CPI-Bench achieves the highest alignment with the Arena Image Edit Leaderboard, indicating stronger consistency with public human preference rankings.

在对 1,000 只美股进行小时级收益预测时，我们观察到一种意外现象：预测结果近乎平坦，且截面相关性衡量的股票排序能力很差，我们将其称为"预测坍缩"。令人惊讶的是，在同一设置下对交易量进行预测时，该现象基本消失。我们在多种时序基础模型（TSFMs）、12 个深度学习预测模型以及 97 个公开基准配置中系统考察了这一现象，发现其与目标可预测性密切相关，并识别出背后的两类成因：低 p……When forecasting hourly returns for 1,000 US equities, we observe an unexpected phenomenon: predictions become nearly flat and show poor stock ranking, as measured by cross-sectional correlation. We call this forecast collapse. Surprisingly, the phenomenon largely disappears when forecasting trading volume under the same setting. We investigate forecast collapse across time-series foundation models (TSFMs), twelve deep-learning forecasting models, and 97 public benchmark configurations, and find that it is closely tied to target predictability. We identify two distinct reasons behind it: low p

提出 Second Thought，一种免训练的推理框架，在每个 Thought 阶段结束时立即 fork 四个辅助分支，与主循环并发解码，并在环境 observation 到达时将生成的 thought 合并回去。This work proposes Second Thought, a training-free inference framework that forks four auxiliary branches the instant each Thought phase concludes, decodes them concurrently with the main loop, and merges the generated thoughts back when the environment observation arrives.

一个机器人过程评估工具包，将 rollout 视频转化为稠密进度曲线并衍生多项细粒度指标；引入 RoboPulse++ 用于评估过程奖励模型（PRM）的可靠性，为评测者提供更准确的测试平台。A toolkit for robot process assessment that turns rollout videos into dense progress curves and derives multiple fine metrics, and introduces RoboPulse++ to evaluate the reliability of process reward models (PRM), providing evaluators with a more accurate testing platform.

提出 Latent On-Policy Self-Distillation（LOPD），不再提出另一种手工设计、附带新形式 privileged context 的 OPSD 变体，而是让 teacher 的 privileged context 本身可从经验端到端学习。This work introduces Latent On-Policy Self-Distillation (LOPD), which, rather than proposing another hand-crafted OPSD variant with a newly prescribed form of privileged context, makes the teacher's privileged context itself learnable end-to-end from experience.

提出 MobileMem，一个面向设备端长期记忆研究的 benchmark 和框架，基于长达一年的移动端经验集合，使 agent 能够记忆过去、理解当下并适应未来。This work introduces MobileMem, a benchmark and framework for studying on-device long-term memory, grounded in a year-scale collection of mobile experiences, and enables agents to remember the past, understand the present, and adapt to the future.

MMDiff 是一种多模态 model-diffing 框架，训练多模态 SAE 并将其转化为特征级接口，用于发现和控制多模态行为；研究表明多模态 SAE 不仅可作为可解释性工具，还可作为审计、引导和控制 MLLM 行为的机制，以实现更安全、更具能力的生成。MMDiff, a multimodal model-diffing framework that trains multimodal SAEs and turns them into feature-level interfaces for discovering and controlling multimodal behavior, shows that multimodal SAEs can serve not only as interpretability tools, but as mechanisms for auditing, steering, and controlling MLLMs behavior toward safer and more capable generations.

展望 ALD/E-ImageMiner benchmark 如何指导未来科学图像挑战赛，以及 Bloom-informed 问题设计如何支持更深入的科学理解。A forward-looking perspective is presented on how the ALD/E-ImageMiner benchmark can guide future scientific-image challenges and how Bloom-informed question design can support deeper scientific understanding.

提出 Self-Supervised Visual On-Policy Distillation（S²VOPD），一种简单有效的方法，通过非对称增强视图构建 on-policy 学习信号，系统地探索了视觉增强的广阔设计空间，并发现非对称性至关重要。Self-Supervised Visual On-Policy Distillation (S$^2$VOPD), a simple yet effective method that constructs on-policy learning signals from asymmetric augmented views, systematically explores a broad design space of visual augmentations and uncover that asymmetry matters.

引入 student reference KL 损失并 mask 特殊终止 token 的 advantage，以缓解生成过长和频繁截断的问题；在 HLE 和 HiPhO 等科学 benchmark 上取得改进，表明 OPD 传递的推理能力可泛化至数学训练领域之外。This work introduces a student reference KL loss and mask the advantages of special termination tokens to mitigate the problem of excessive generation length and frequent truncation, and improves on science benchmarks such as HLE and HiPhO, suggesting that OPD transfers reasoning capabilities that generalize beyond the mathematical training domain.

提出 Mimir v1，一个基于 Hierarchical Reasoning Model（HRM）架构的 10 亿参数语言模型，从头训练，在英语上具有高度竞争力，并仅使用合规的后训练数据在丹麦语上创下新的 SOTA。Mimir v1 is introduced, a 1-billion-parameter language model based on the Hierarchical Reasoning Model (HRM) architecture that is trained from scratch and delivers highly competitive performance for English and sets a new state of the art for Danish using only permissible post-training data.

提出 LITTLECURRICULUM，一个为美国小学课程量身定制的 88B token 精选预训练语料，并由此训练得到 LITTLELEARNER；该模型具备开放式评估所需的语言能力，同时其知识和能力边界清晰可映射到可解释的课程指南。This work introduces LITTLECURRICULUM, a curated 88B-token pretraining corpus tailored to U.S. elementary school material, which yields LITTLELEARNER, a model with sufficient language competence for open-ended evaluation, yet with clear knowledge and capability boundaries mapped to interpretable curriculum guidelines.

提出 UniProbe，一个轻量、统一、可学习的检测器，通过单次前向传播建模冻结 LVLM 的异构计算 trace，在 token 级和物体级幻觉检测上达到 SOTA。UniProbe is introduced, a lightweight, unified, learnable detector that models a frozen LVLM's heterogeneous computational trace from a single forward pass and achieves state-of-the-art token-level and object-hallucination detection.

提出 UNMASK，一个全自动 pipeline，可在无需额外人工标注的情况下发现、因果验证并缓解文本分类器中的伪相关，并证明其发现与验证阶段可泛化至奖励模型的偏好数据。U N M ASK is presented, a fully automated pipeline that discovers, causally verifies, and mitigates spurious correlations in text classifiers without additional human annotation, and demonstrates that the discovery and validation stages generalize to reward model preference data.

在多个 benchmark 数据集上的实证评估表明，RAEF 在准确率和推理开销方面均优于 RAF，并且与零样本及微调基础模型的全面对比显示，RAEF 在避免微计算负担的同时取得了与微调相当或更优的性能。Empirical evaluation across multiple benchmark datasets demonstrates that RAEF outperforms RAF in both accuracy and inference overhead, and comprehensive comparisons with zero-shot and fine-tuned foundation models show that RAEF achieves competitive or superior performance to fine-tuning while avoiding its computational burden.

提出一种 chunked-prefill 策略，可缓解由此带来的内存容量惩罚，在 32 GiB 共享内存上将允许的上下文宽度扩展 $2.7 \times$，但即便降低了内存开销，仍需打补丁才能使 Nanbeige4.2-3B 可用。A chunked-prefill strategy is introduced which alleviates the incurred memory-capacity penalty, extending allowable context width by $2.7 \times$ on 32~GiB shared memory, however, even with the reduced memory overhead, it is shown that patches are required to render Nanbeige4.2-3B usable.

发现面向推理的训练并未优先放大具有最高 Lift 的行为，这促使研究者采用过程级目标，以奖励经过校准且有依据的推理，而不仅仅是表面形式。It is found that reasoning-oriented training does not preferentially amplify the highest-Lift behaviors, motivating process-level objectives that reward calibrated and grounded reasoning rather than surface form alone.

2023 年，一位纽约法官在 Mata v. Avianca 案中制裁了两位律师，因其提交的 brief 包含了由 ChatGPT 生成的虚构引用。此类失误大多能被数据库检索发现；但更棘手的问题在于检测那些指向真实案例、却不支持其所述命题的引用——这一失效模式是现有面向法律场景的 LLM 评测基本忽略的。本文通过对来自两个法律语料库的真实法律引用进行受控扰动（替换引用In 2023, a New York judge sanctioned two attorneys in Mata v. Avianca for filing a brief with hallucinated citations generated by ChatGPT. Such failures are largely caught by database lookups; the harder problem is detecting citations that point to real cases but do not support the propositions for which they are offered -- a failure mode that existing evaluations of LLMs for legal use cases largely overlook. In this paper, we study proposition-level citation support verification through controlled perturbations of real legal citations obtained from two legal corpora, either replacing the cite

提出 Apodex Discovery，一个通过 heavy-duty solver 构建和评估发现型 AI 的框架；该 solver 包含一个 foundation model、harness、工具和控制策略，用于执行长期的、有状态的、可验证的探索。This work introduces Apodex Discovery, a framework for building and evaluating discoverative AI through the heavy-duty solver, a system comprising a foundation model, harness, tools, and control policies that pursues extended, stateful, verifiable investigations.