👍 100
09/29 08:00
Linear attention replaces growing KV caches with fixed-size recurrent states, yet these persistent states can become a substantial memory bottleneck under concurrent serving. Directly quantizing recurrent states to low precision often leads to severe accuracy degradation, as quantization errors prop
中文介绍 针对 linear attention 中循环状态量化导致的精度退化问题,STEPQuant 分析 delta-rule 递归状态中误差的重要性分布,提出「何时/何地」差异化的量化策略,在低比特下缓解精度损失,降低并发服务时的显存瓶颈。
👍 53
10/05 08:00
Multimodal Large Language Models (MLLMs) have demonstrated remarkable potential in video understanding, yet their reliance on retrospective summarization and text-centric priors often limits their ability to bridge unobserved causal transitions when applied to Video Event Prediction (VEP). To addres
中文介绍 面向视频事件预测中未观察因果过渡的难题,VepAgent 提出工具增强的强化学习框架,让 MLLM 在预测过程中调用外部工具补足因果链,减少对回顾式摘要的依赖,提升对未见事件的因果推断能力。
👍 48
10/01 08:00
Vision-language-action models benefit from the understanding and reasoning capabilities of pretrained vision-language models, but action-only supervision provides limited grounding in world dynamics. Conversely, world-action models inherit spatiotemporal priors from video generation models, yet rema
中文介绍 UniWAM 提出统一世界-动作模型,融合 VLM 的语义推理能力与视频生成模型的时空先验,解决动作监督对世界动力学约束不足、世界模型缺乏语义的问题,实现更贴近真实动态的具身智能决策。
👍 34
10/07 08:00
Much knowledge work produces new deliverables from files a workspace already holds, and LLM agents are beginning to take such work over. Through direct corpus interaction, an agent can search and read any of those files from a terminal with no indexing, and producing a deliverable from many of them
中文介绍 RunningTab 让 LLM agent 通过与工作区终端直接交互来搜索和读取文件,无需预建索引,即可从已有语料中生成新交付物,提升真实知识工作中跨文件检索与产出的效率。
👍 31
10/06 08:00
Reinforcement learning with verifiable rewards (RLVR) turns agent experience into learning signals primarily through scalar outcome rewards after interaction. For group-relative objectives, however, this signal vanishes when all rollouts receive the same reward, even though their trajectories may re
中文介绍 Self-Retrospection Distillation 针对 RLVR 中当所有 rollout 奖励相同时组相对信号消失的问题,将事后经验提炼为事前先验,让 agent 在推理前利用自我回顾蒸馏提升轨迹质量。
👍 29
10/07 08:00
General-purpose agents increasingly write code, use tools, and complete complex digital tasks, raising the question of how far these capabilities carry into the physical world. To investigate this, we introduce RobotWorld, a challenging simulation testbed for robot use: turning instructions and obse
中文介绍 RobotWorld 提出多模态机器人 agent 仿真测试平台,评估通用 agent 将指令与观察转化为可执行代码、工具调用来完成跨任务、跨本体机器人操作的能力,探索数字 agent 能力向物理世界的迁移边界。
👍 26
10/06 08:00
Astra can act, yet reliable manipulation depends on the system through which it observes and controls the world. We introduce PhysEvo, a framework for physical recursive self-improvement (RSI) around a single frozen model. A task agent executes robot tasks; a meta-agent uses the resulting trajectori
中文介绍 PhysEvo 围绕单一冻结模型构建物理递归自改进框架:任务 agent 执行机器人操作,元 agent 利用轨迹数据迭代改进感知与控制系统,实现物理场景下的闭环能力进化。
👍 25
10/07 08:00
We introduce SWE-Game, a benchmark of 247 tasks grounded in 41 executable reference Godot games spanning 13 gameplay categories in 2D and 3D. Five task types cover development from a brief, implementation from a game design document, skeleton completion, repair of 83 injected-fault cases, and Godot-
中文介绍 SWE-Game 是一个包含 247 个任务、覆盖 41 个 Godot 游戏和 13 类游戏玩法的基准,评估编码 agent 在 2D/3D 游戏开发、设计文档实现、缺陷修复及 Godot 代码生成等五种任务类型上的能力。
👍 20
10/03 08:00
Sparse-view novel view synthesis is a central problem in 3D content creation, but diffusion-based approaches remain limited by iterative denoising, making multi-view generation expensive at inference time. We introduce NAMVIS, a diffusion-free framework that reformulates multi-view image synthesis a
中文介绍 NAMVIS 提出免扩散的多视图图像自回归生成框架,将稀疏视图新视角合成重构为 next-scale 自回归任务,避免迭代去噪开销,显著降低多视角生成的推理成本。
👍 17
10/07 08:00
Flow policies capture rich and diverse action distributions, and fine-tuning them with off-policy RL to improve beyond the demonstrations has drawn growing interest. However, fine-tuning a flow policy against a learned value function is not trivial, because the policy generates its action over many
中文介绍 针对 flow policy 与学习到的价值函数离线 RL 微调的难题,本文提出基于标量伴随匹配的 Q-Learning 方法,解决策略跨多步生成动作与价值估计难以对齐的问题,提升超越示教的性能。
👍 16
10/07 08:00
Training and evaluating interactive language agents typically requires rich user interactions, yet collecting human feedback is expensive and difficult to scale. Simulated users offer a scalable alternative, but they must both resemble real user behavior and provide useful learning experiences for a
中文介绍 MIMESIS 将用户模拟器作为训练环境,使交互语言 agent 在无真实用户反馈下获得可扩展训练信号,模拟器既模仿真实用户行为又提供高质量学习信号,弥补人工反馈昂贵且难扩展的不足。
👍 16
10/07 08:00
Test-time scaling (TTS) improves the reasoning capabilities of large language models by allocating additional inference computation. Existing approaches to improving TTS efficiency largely optimize accuracy against one resource dimension at a time, advancing either the accuracy--cost or accuracy--la
中文介绍 该工作提出基于摊销 agentic 策略发现的个人化 test-time scaling 方法,从单维度准确率-成本折中扩展到多维偏好,让不同用户对推理计算资源的个性化分配更贴合自身需求。
👍 12
10/06 08:00
A central challenge for scientific agents is to turn analytical experience into reusable expertise grounded in physical evidence. Here we introduce Gan Jiang, a self-learning agent for powder X-ray diffraction built on a diffraction-analysis ecosystem we developed: XMatcher, XQueryer, XDecomposer an
中文介绍 Gan Jiang 是一个自学习的粉末 X 射线衍射分析 agent,依托 XMatcher、XQueryer、XDecomposer 等工具生态,将分析经验转化为可复用的物理证据驱动专家知识,实现 XRD 数据的自动化解析。
👍 12
10/07 08:00
We introduce QuadTok, a novel framework for visual tokenization and autoregressive image generation. Compared to traditional approaches using 2D grids or 1D token sequences, we propose a hierarchical quadtree structure, bridging the gap between 2D spatial binding and 1D sequence-level flexibility. T
中文介绍 QuadTok 提出基于四叉树的视觉 tokenizer 框架,用层次化四叉树结构替代传统二维网格或一维序列,在自回归图像生成中兼顾二维空间绑定与一维序列灵活性,提升生成质量。
👍 11
10/06 08:00
Speculative decoding accelerates autoregressive generation in large language models. In each drafting stage, a lightweight draft model proposes tokens that the target model subsequently verifies. With increasingly capable draft models, we find that the target model frequently accepts all tokens prod
中文介绍 DLoop 提出循环推测解码方法:当 draft 模型越来越强、target 模型常全部接受生成 token 时,利用循环机制进一步加速自回归生成,提升大模型推理吞吐。
👍 10
10/06 08:00
Reconstructing an editable CAD model from a 3D shape remains a challenging engineering task. Existing methods can propose CAD operations, but no single source of proposals works equally well across different part geometries and stages of reconstruction. We introduce CADFather, an autonomous agentic
中文介绍 CADFather 提出自治 CAD 重建 agentic 系统,通过协调多种工具的分工来提出 CAD 操作,弥补单一来源在不同零件几何和重建阶段表现不均的问题,实现从 3D 形状到可编辑 CAD 模型的自动恢复。
👍 10
10/05 08:00
Wrapping an image generation model in an agentic harness can effectively boost Text-to-Image task performance: the harness can leverage memory, skills, workflow orchestration, result verification, and iterative refinement to continually construct and revise prompts, thereby eliciting better images.
中文介绍 该工作提出 on-policy context distillation,将 agentic harness 在文生图任务中的提示构造、记忆与迭代修正经验内化到扩散模型权重中,在不依赖外部推理脚手架的情况下提升图像生成质量。
👍 8
10/04 08:00
Transcribing music into a human-readable score requires a coherent understanding of rhythm, harmony, melody, and form. Two obstacles limit this goal: annotated recordings are scarce, and accurate local predictions can still produce inconsistent musical sequences. We present SheetSage2, a unified mus
中文介绍 SheetSage2 面向乐谱转写,结合合成监督数据统一处理节奏、和声、旋律与曲式的一致性建模,解决标注数据稀缺与局部预测正确但全局序列不一致的双重问题,提升可读乐谱生成质量。
👍 8
09/27 08:00
Multimodal large language models can expose task-relevant evidence through generation, but producing useful evidence does not by itself determine how it enters a retrieval embedding. We study whether the semantic organization of that evidence can also specify where representations are read. To addre
中文介绍 该工作研究多模态大模型中任务相关证据的语义组织如何决定其在检索嵌入中的读取位置,提出 evidence-aligned readout 机制,让多模态嵌入的表示空间结构与证据的语义层次对齐。
👍 7
10/07 08:00
Memory-augmented reinforcement learning strengthens LLM agents' ability to solve complex long-horizon tasks. Skills are one such form of memory, pairing instructions with an applicability condition over task types. However, retaining every skill indiscriminately as the policy improves lets obsolete
中文介绍 SkillForge 提出动态技能生命周期机制,让技能与 agent 策略协同进化:随着策略提升自动淘汰过时技能、保留仍适用的技能,缓解记忆增强 RL 中技能累积带来的冗余与过时问题。