👍 124
11/04 08:00
Streaming video understanding requires models to repeatedly decide whether to respond or continue observing. Recently, Group Relative Policy Optimization (GRPO) has been applied to this setting, with existing approaches emphasizing reward design while retaining a single trajectory-level advantage ac
中文介绍 针对流式视频理解中GRPO奖励设计未区分轨迹内各回合贡献的问题,提出回合级信用分配方法,将优势分解至每个回合,改善长时交互场景下的训练稳定性与模型性能。
👍 56
11/05 02:59
Vision-language-action models are typically built on a simple premise: perception is first compressed by visual encoders, and action reasoning begins only afterwards. Yet manipulation often hinges on precisely the visual signals most vulnerable to this abstraction, e.g., local geometry, boundaries,
中文介绍 指出视觉编码器预压缩会丢失操纵任务所需的关键局部几何与边界信息,提出原生视觉-语言-动作策略,将视觉特征直接融入动作推理,减少信息瓶颈,提升操作精度。
👍 48
11/04 08:00
AI agents are playing an increasingly important role in physics research. However, existing physics benchmarks only evaluate fixed questions, isolated sub-tasks, or paper reproduction with given solutions, and rarely check research quality alongside correctness. We introduce FrontierPhysics, an end-
中文介绍 构建端到端物理研究基准FrontierPhysics,评估智能体从假设到论文的完整科研流程,不仅检查结果正确性,还评估研究质量,覆盖多子领域,揭示现有智能体在开放研究中的不足。
👍 34
11/04 21:24
While multi-turn agentic reinforcement learning (RL) excels on long-horizon tasks, its performance gains remain highly sensitive to interaction constraints. Evaluating across sampling breadth K and interaction depth T via Pass@(K,T), we show that trajectory-level RL advantages depend on sufficient t
中文介绍 基于Pass@(K,T)度量,系统分析多轮智能体强化学习中采样宽度与交互深度的影响,发现轨迹级优势依赖足够的任务探索,且性能提升对交互约束高度敏感,为训练设计提供指导。
👍 31
11/03 14:45
A frontier AI model is trained in an AI data center, on tens of thousands of accelerators joined by a dedicated network, and every training step ends with all of them exchanging data at once. Between racks, that network is optical, and each of its four layers, from lasers to switches, runs a fixed o
中文介绍 提出Halley,一种预测性控制平面,用于AI数据中心光网络,利用全局训练步信息动态调度光交换,降低网络拥塞和时延,提升大规模训练的效率与稳定性。
👍 24
11/04 02:25
Teleoperation remains the default way to collect robot demonstrations, but every episode costs robot and operator time, and every new task starts from scratch. We ask whether a coding agent can take over the teleoperator's job. In our DataFarm, the agent receives a data request and one calibration e
中文介绍 提出PyRUA-DataFarm,用代码智能体替代人工遥操作,依据数据请求和校准示例自动生成机器人操作代码,低成本收集真实世界演示数据,新任务无需从零开始,显著减少人工成本。
👍 23
10/31 08:00
Streaming video LLMs can now commentate live gameplay, but they are trained and tested on the same games. We study cross-game streaming commentary, in which a model trained on one game commentates games it has never seen, and find that current commentators transfer their timing but not their knowled
中文介绍 研究跨游戏流式解说场景,发现现有模型虽能迁移时间节奏但缺乏知识迁移。提出StreamDuo框架,结合实时解说员与研究智能体,对未见游戏进行有效评论与知识注入。
👍 22
11/04 08:00
Group Relative Policy Optimization (GRPO) has advanced generative model post-training in verifiable domains such as code and mathematics, but evaluating video quality, especially emotional expression, often depends on subjective aesthetics and has no uniquely correct target. Existing approaches typi
中文介绍 针对情绪视频生成中GRPO因主观审美而缺乏唯一正确目标的问题,提出EmoReward,自进化奖励校准机制,通过反馈逐步调整奖励信号,提升生成视频的情绪表达准确性与一致性。
👍 16
11/03 08:00
Multi-teacher on-policy distillation (MOPD) extends on-policy distillation (OPD) to consolidate capabilities from multiple teachers. Although prior work reports near-perfect in-distribution (IND) distillation performance, we find that this can mask incomplete capability consolidation beyond the supe
中文介绍 发现多教师在线策略蒸馏在分布内看似完美,但掩盖了能力整合不足的问题。系统研究其泛化局限性,证明需扩展评估范围才能确保跨任务能力真正被蒸馏,为后训练整合提供警示。
👍 14
11/01 22:05
LLM-based agentic systems are increasingly combined with CoT reasoning to solve tasks that require accurate reasoning, external knowledge, and multi-step deliberation. Yet, explicit CoT incurs substantial generation overhead, while implicit CoT, which replaces textual rationales with compact continu
中文介绍 提出TAC-CoT,将显式CoT令牌与隐式推理交替穿插,利用工具增强推理能力,在保持隐式高效性的同时提升多步任务的准确性,降低生成开销,平衡效率与性能。
👍 14
10/30 08:00
Self-distillation methods for large language models with an exponential moving average (EMA) teacher already retain a history of student updates. Muon nevertheless stores a separate momentum buffer, increasing persistent training memory. We propose Muon with Reused Teacher (Muon-RT), which reuses th
中文介绍 提出Muon-RT,复用EMA教师的轨迹作为优化器动量,消除Muon的独立动量缓冲区,在LLM自蒸馏中减少持久内存占用,同时保持收敛速度和性能。
👍 14
11/02 08:00
Speculative decoding (SD) accelerates large language model inference by reducing the sequential cost of token-by-token generation. Advanced SD methods reduce proposal cost with parallel drafters that predict entire blocks in one forward pass, but overall speedup still depends on acceptance length. O
中文介绍 提出SpecOPD,分支并行在线策略蒸馏,用于投机解码,训练多个并行草稿分支以优化接受长度,减少token顺序生成开销,加速LLM推理同时保持输出质量。
👍 13
11/03 08:00
Multimodal large language models (MLLMs) can now process hour-long videos, making extended multi-turn dialogue over a single video a natural mode of use. Sustaining such a dialogue requires more than answering each question in isolation: the model must retrieve and reuse information established earl
中文介绍 提出VideoMemBench,评估多模态LLM在长视频多轮对话中的记忆能力,测试模型能否检索并复用早期信息,发现现有模型在长时间对话中记忆退化严重,为视频对话系统提供基准。
👍 13
11/03 08:00
Pretrained Vision-Language-Action (VLA) policies provide a strong starting point for deployment, but adapting them to a target robot and environment still requires costly embodiment-specific demonstrations. Human demonstrations are far easier to collect, yet directly leveraging them is challenging d
中文介绍 针对VLA策略需特定于具体机器人的演示数据问题,提出桥接表示与动作空间的方法,利用人类演示数据对齐到目标机器人,减少对昂贵本体演示的依赖,提升泛化与部署效率。
👍 13
10/29 08:00
On-policy Self-distillation (OPSD) has proven effective for post-training large language models by constructing a teacher conditioned on privileged references. However, existing approaches keep the teacher fixed either throughout training by consistently exposing it to the full reference, or within
中文介绍 提出多教师自适应在线策略自蒸馏,每个令牌使用不同教师,按需动态调整蒸馏强度,避免固定教师导致的学习偏差,增强LLM后训练中的能力整合与分布外泛化。
👍 11
11/02 08:00
We introduce Dory1.0-4B, a 4-billion-parameter looped (latent-recurrent) language model with elastic depth: a single checkpoint serves different inference compute budgets. Dory1.0-4B is built on a normalized Transformer (nGPT) backbone with interleaved sliding-window and full-attention layers, organ
中文介绍 发布Dory1.0-4B,一种40亿参数的循环潜在递归语言模型,基于归一化Transformer并交织滑窗与全注意力,支持弹性深度,单个检查点可适应不同推理算力预算。
👍 10
10/30 16:38
Computer-use agents increasingly tackle tasks ranging from everyday desktop activities to long-horizon professional workflows. Yet interfaces centered on screenshots and mouse-and-keyboard actions can impose a human-interaction prior, requiring repeated screen interpretation while leaving the comput
中文介绍 提出Hybrid Computer Use统一接口,结合程序化、语义与视觉操作,让智能体直接调用计算机功能而非仅模拟鼠标键盘,提升长程任务效率并减少屏幕解读负担。
👍 10
11/05 00:46
Coding agents powered by models such as GPT-6-Astra connect spatial reasoning with executable robot behavior, raising the question of how such agents can produce lasting improvements in robot capabilities from physical interaction. In this work, we study embodied recursive self-improvement (RSI) thr
中文介绍 通过GPT-6-Astra驱动的编码智能体在真实机器人上实现具身递归自我改进,使智能体从物理交互中持续学习并提升能力,探索自主改进机器人技能的新路径。
👍 9
11/02 08:00
Generative AI is billed per token, yet tokens are not equally expensive. Each token enters the KV-cache and raises the cost of those that follow, so energy grows non-linearly along a sequence whose tokens look identical. No analytical model states what a token costs, and the most accurate SOTA model
中文介绍 提出SIMPLE定律,解析描述LLM推理中能量成本与序列位置的关系,每token的KV缓存成本累积导致能耗非线性增长,为推理计费与优化提供精确模型。
👍 9
11/04 02:16
Looped transformers are a parameter-efficient alternative to standard transformers for language modeling. By reapplying the same layers across multiple loop steps, recent looped language models match the performance of non-looped models several times their size. While this architecture effectively r
中文介绍 提出LoopVQ,面向循环Transformer的KV缓存量化方法,在保持模型性能的同时实现亚1比特量化,显著降低循环语言模型的显存占用,提升长序列推理效率。