👍 228
10/08 08:00
Online reinforcement learning (RL) fine-tunes diffusion models with black-box rewards, but reverse-process methods such as Flow-GRPO are limited in efficiency by poor credit assignment, while forward-process methods such as DiffusionNFT are efficient yet unstable without carefully tuned exponential-
中文介绍 针对扩散模型在线 RL 微调,提出 DiffusionART。逆过程方法(如 Flow-GRPO)受信用分配限制效率低,正过程方法(如 DiffusionNFT)效率高但稳定性差。DiffusionART 结合两者优势,实现高效、稳定且可扩展的扩散模型强化。
👍 201
10/09 20:48
Multimodal language models (MLLMs) exhibit promising spatial reasoning capabilities, yet the internal mechanisms supporting them remain poorly understood. We uncover S-Space, a linear representation subspace in which object-token representations encode continuous coordinates along horizontal, vertic
中文介绍 发现多模态语言模型内部存在线性表示子空间 S-Space,其中 object-token 编码水平、垂直方向的连续坐标。该发现揭示了 MLLM 空间推理能力的内部机制,为理解其空间智能提供了新视角。
👍 83
10/08 08:00
Computer-use agents (CUAs) are rapidly improving, yet synthesizing diverse and reliable software environments at scale remains challenging. Native application environments require substantial construction and integration effort, which limits application coverage and increases the risk of application
中文介绍 提出 SoftwareGym,通过大规模合成多样化软件环境来训练计算机使用智能体(CUA)。原生应用环境构建集成成本高、应用覆盖有限。SoftwareGym 以合成方式解决可扩展性瓶颈,降低训练智能体的应用集成风险。
👍 57
10/06 08:00
Multimodal large language models have become the dominant paradigm for visual understanding, but incur substantial costs by encoding inputs into dense, fixed-size patch tokens. However, visual information is unevenly distributed: some regions require fine-grained detail, while others admit compact r
中文介绍 提出 VisionWeave,将弹性视觉表示作为 MLLM 的固有功能。传统方法将图像编码为密集固定尺寸 patch token 代价高昂,而视觉信息分布不均。VisionWeave 按需分配表示粒度,平衡精细细节与紧凑压缩。
👍 42
10/08 08:00
Adversarial games have driven advances from heuristic search to reinforcement learning, yet learning and adapting strategies from limited samples remain challenging. AI agents offer an alternative by turning game experience into revisions of executable policies. Building on heuristic learning (HL),
中文介绍 在长时间运行的博弈 Agent 竞赛中评估启发式学习(HL)。传统方法依赖强化学习从少量样本学策略较难。该工作构建 HL 框架,将游戏经验转化为可执行策略修订,探索 AI Agent 自主学习进化能力。
👍 39
10/09 08:00
Computer-use agents can perform increasingly complex GUI tasks, yet unfamiliar software and workflows requiring human domain knowledge remain challenging. Human demonstrations convey procedural knowledge through the same GUI interaction space, but reuse requires interpreting actions in context and s
中文介绍 提出 Work2Skill,从人类 GUI 操作演示中捕获经验并归纳可复用 Agent 技能。现有 CUA 缺乏处理陌生软件和需领域知识工作流的能力。Work2Skill 将人类演示转化为可复用的 Agent 技能表示,提升泛化性。
👍 26
10/08 08:00
Speculative decoding accelerates large language model inference by using a lightweight drafter to propose multiple tokens that are subsequently verified in parallel by the target model. Recent diffusion-based drafters make this paradigm particularly attractive: deep attention-based neural backbones
中文介绍 针对推测解码提出上下文修正的并行草稿方法。扩散式 drafter 因深层注意力结构天然适合并行提议多 token,但草稿质量受限。该方法引入上下文修正机制提升草稿准确性,进一步加速 LLM 推理。
👍 20
10/08 08:00
Static coding benchmarks are fixed at release. Over time their tasks leak into training data, and stronger agents eventually saturate them. We argue that a benchmark should instead aim to be self-scaling: its tasks and oracles are created at evaluation time, so no content exists in advance to leak,
中文介绍 提出 SWE-Duel,一个自我扩展的编码 Agent 评测竞技场。传统静态基准随时间泄漏至训练数据且 Agent 会饱和。SWE-Duel 在评估时动态生成任务和验证器,实现无预先泄露内容的持续评测。
👍 19
09/29 08:00
Process reward models (PRMs) have become a key component for LLMs, as their step-level feedback supports both post-training and test-time reasoning. However, training strong PRMs remains costly: human step annotation is difficult to scale, while Monte Carlo estimation is computationally expensive an
中文介绍 提出仅从结果监督的强化学习归纳过程监督的方法。PRM 步骤级反馈有价值但训练成本高:人工标注难扩展,蒙特卡洛估算开销大。该方法无需步骤标注,直接从结果信号推断过程奖励模型。
👍 18
10/06 08:00
Automated research systems increasingly run LLM agents over long horizons, but more inference does not by itself produce more progress: agents replay growing histories, duplicate one another's work, or stop experimenting while token consumption continues. Yet most evaluations use short budgets or be
中文介绍 提出无状态语言 Agent 架构,扩展长时自动研究。现有 Agent 重放增长历史、重复工作或停止探索,但 token 消耗持续增加。Stateless Agent 消除冗余历史依赖,提升长时程研究效率。
👍 18
10/09 08:00
In this paper, we study physically grounded articulated 3D asset generation---moving the goal from reproducing how objects look to capturing how they move and function. Inspired by the literature on procedural and agentic generation, we present Artisan, a system that extends the latest method Articr
中文介绍 研究物理接地的铰接 3D 资产生成,从复制外观转向捕捉运动和功能。提出 Artisan 多 Agent 流水线,在 Articraft 基础上引入物理约束,确保生成资产在碰撞、关节运动等维度合理可行。
👍 16
10/04 08:00
Recent GPU kernel optimization agents retain what they learn in knowledge bases or as distilled skills. Kernel benchmarks score each attempt's implementation for correctness and speed but leave the reuse value of retained experience unmeasured. We introduce MetaKernelBench, which measures whether ex
中文介绍 提出 MetaKernelBench,评估 GPU 内核优化 Agent 的知识迁移能力。现有基准仅衡量单次实现的正确性和速度,忽略保留经验的可复用价值。MetaKernelBench 衡量经验跨任务复用效果,揭示知识保留实际效用。
👍 16
10/08 08:00
Language model agents increasingly operate over long interaction streams, requiring online adaptation to distribution shifts through continual learning. However, existing agent harnesses typically either evolve their states only through forward information flow or rely on additional replay or valida
中文介绍 提出循环 Harness 用于 Agent 持续学习。现有 Agent Harness 仅通过前向信息流更新状态,或依赖回放和验证集。循环 Harness 引入反馈回路,支持 Agent 在线适应分布漂移,无需额外验证数据。
👍 14
10/08 08:00
Humans perceive far more in a scene than what is explicitly depicted: a single glance captures past causes and future trajectories; a quick peek determines if a vehicle can fit between two parked cars; a few seconds of video reveals who holds authority in a room; and a fleeting clip highlights subtl
中文介绍 提出「第六感」基准,评测多模态模型的直觉视觉推理。人类能感知场景中隐含信息:过往因果、未来轨迹、空间适配性、社会权力关系等。该基准考察模型超越显式视觉内容的深度直觉理解能力。
👍 13
10/08 08:00
Data video is a media form that integrates data visualization with video narrative, widely adopted in news reporting and business analysis. Compared with general video understanding, data video understanding places greater emphasis on accurately reading data from animated charts, integrating evidenc
中文介绍 提出 DataVista,诊断多模态 LLM 在数据视频理解上的能力。数据视频将可视化与视频叙事结合,需准确读取动态图表数据、整合证据并推理。DataVista 系统评估模型在数据密集型视频中的表现。
👍 13
10/09 08:00
Decoder hidden states from different layers of large language models (LLMs) often encode complementary information. However, existing speculative decoding methods typically rely on a small set of target-model layers chosen using fixed, position-based heuristics. Selecting an optimal subset of layers
中文介绍 分析 LLM 隐藏状态的信息与几何特性以优化推测解码。现有方法基于固定位置启发式选择少量目标层。本文研究激活波特征,提出自适应层选择策略,提升 drafter 与信息对齐效率。
👍 13
09/29 08:00
Video understanding agents acquire evidence through an executable harness that controls what they observe and how they use those observations. However, execution traces contain only the evidence acquired by the current harness, leaving competing explanations for failure unresolved and limiting the b
中文介绍 提出 Video-RSI,通过 Harness 进化实现视频理解 Agent 的递归自改进。当前 Harness 仅获取已选证据,无法消除失败的多重解释。Video-RSI 通过递归进化 Harness 探索更多证据路径,提升理解鲁棒性。
👍 13
10/09 08:00
Large language models (LLMs) are increasingly explored for graph understanding and decision-making, while System One models such as Jev select directly from supplied options. However, the capabilities of System One models on graph-related tasks remain unclear. We introduce GraphDecide, a model-indep
中文介绍 提出 GraphDecisionBench,评测 System One 模型在图任务上的表现。LLM 已用于图理解和决策,但 Jev 等 System One 模型的能力尚不清楚。GraphDecisionBench 为独立模型评测提供标准化基准。
👍 12
09/30 08:00
Foundation models can serve as clinical agents through tool-use harnesses. However, conventional medical benchmarks assess reasoning over preselected evidence rather than the ability to seek it across clinical records and longitudinal imaging. We propose CASE: a series of role-specific Clinical Agen
中文介绍 提出 CASE,面向纵向医学推理的 Agentic 学习框架。传统医学基准评估预选证据上的推理能力,未考察主动检索跨记录证据的能力。CASE 让临床 Agent 主动搜索纵向病历和影像证据,提升推理可靠性。
👍 12
10/09 08:00
Efficient visual tokenization requires balancing representation compactness with the preservation of fine-grained image content. Conventional grid-based tokenizers allocate tokens uniformly across spatial locations despite non-uniform visual complexity, whereas compact 1D tokenizers aggregate global
中文介绍 提出 MixTok,自适应视觉分词方法,结合全局 1D token 与稀疏 2D 细化。传统网格分词均匀分配 token 忽略视觉复杂度不均,紧凑 1D 分词丢失局部细节。MixTok 平衡紧凑性与细粒度信息保留。