HuggingFace 每日AI论文速递

duan
  • 5.0 (2)
  • Technology
  • Daily

每天10分钟，带您快速了解当日HuggingFace热门AI论文内容。每个工作日更新，欢迎订阅。 📢播客节目在小宇宙、Apple Podcast平台搜索【HuggingFace 每日AI论文速递】 🖼另外还有图文版，可在小红书搜索并关注【AI速递】

  1. 2h ago

    2026.11.09 | 高斯间距校准跨场景重建；参数先验加速多教师蒸馏

    ⏱️ Highlights[00:25] 🌐 SplatAnything: Gaussian Spacing Normalization for Feed-forward 3DGS across Arbitrary Scenes（SplatAnything：面向任意场景前馈式3DGS的高斯间距归一化）[01:02] 🧠 P^2-MOPD: Accelerating Multi-Teacher On-Policy Distillation via Parameter Priors（P²-MOPD：通过参数先验加速多教师在策略蒸馏）[01:43] 🎨 LongBanana: An Expert-Verified Benchmark for Long-Context Multi-Reference Image Synthesis（LongBanana：经专家核验的长上下文多参考图像合成基准）[02:21] 🔍 Beneath the Surface: What We Overlook When We Talk About Agent Evolution（表象之下：谈论智能体进化时，我们忽略了什么）[03:00] 🎬 EvoWorld: Towards Out-of-Sight World Evolution in Generative Video Models（EvoWorld：迈向生成式视频模型中的画外世界演化）[03:39] ⚙️ Fast and Exact Relative-Position Attention via Structure-Aware Kernels（通过结构感知内核实现快速且精确的相对位置注意力）[04:17] 🎙️ Confucius4-R2T2: A Stable-Prefix Streaming ASR Model for Real Real-Time Transcription（Confucius4-R2T2：面向真正实时转写的稳定前缀流式语音识别模型）[04:57] 🤝 AgentBaton: Efficient Model Collaboration in Agentic Tasks（AgentBaton：智能体任务中的高效模型协作）[05:37] 💻 Video-Coder: Reinforcing Agentic Coding for Video Generation（Video-Coder：通过强化智能体编码实现视频生成）[06:17] 🧩 Exploring Weight-Space Programming: Capabilities, Generalization, and Limitations（探索权重空间编程：能力、泛化与局限）[06:57] 🏗️ Frontier3DBench: Benchmarking Coding Agents for 3D Creation across Objects, Scenes, and Interactive Open Worlds（Frontier3DBench：跨物体、场景与交互式开放世界的编码智能体三维创作评测）[07:41] 📷 PREEN: Exemplar-Based Color Style Transfer with Smartphone Editor Presets（PREEN：使用智能手机编辑器预设的示例驱动色彩风格迁移）[08:18] 📊 LLMs Don't Know What They Know: Benchmarking Temporal Self-Knowledge for Financial Backtesting（大语言模型不知道自己知道什么：面向金融回测的时间自知能力评测）[09:00] 🔧 AutoCompress: Automated Hardware-Guided Model Compression（AutoCompress：自动化硬件引导模型压缩）[09:41] 🧱 OmniLatent: Unified Latent for 3D Part Modeling（OmniLatent：用于三维部件建模的统一潜表示） 【关注我们】您还可以在以下平台找到我们，获得播客内容以外更多信息小红书: AI速递 在小宇宙查看该单集文稿

  2. 2d ago

    【周末特辑】11月第2周最火AI论文 | 图形智能体的设计空间；错峰去噪生成视频

    ⏱️ Highlights[00:31] 🎨 Advances in Agentic Computer Graphics（智能体计算机图形学进展）[02:23] 🌊 Wave Forcing: Towards Speed-of-Light Streaming Video Generation（Wave Forcing：迈向光速的流式视频生成）[04:18] 🤖 Are Frontier VLM Agents Ready to Be Robot Generalists? An Empirical Study with the Embodied Agent Arena（前沿VLM智能体准备好成为机器人通才了吗？基于Embodied Agent Arena的实证研究）[06:20] 🛠️ CoworkOS:Scaling Agent Coworkers for Real-world Automation（CoworkOS：扩展智能体同事以实现真实世界自动化）[08:28] 🎬 Turns Matter: Advancing Streaming Video RL via Turn-Level Credit（轮次很重要：通过轮级信用分配推进流式视频强化学习） 【关注我们】您还可以在以下平台找到我们，获得播客内容以外更多信息小红书: AI速递 在小宇宙查看该单集文稿

  3. 3d ago

    2026.11.06 | 视觉规划整块解码；借兄弟框架练智能体

    ⏱️ Highlights[00:22] 🧭 VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding（VFlash：通过闪速图元解码实现快速、准确的视觉规划）[00:54] 🛠️ Harness-S: Learning from Sibling Harnesses for Generalizable Agent Training（Harness-S：向兄弟执行框架学习，实现可泛化的智能体训练）[01:24] 🎲 Random Is Almost Enough? Stable and Low-Cost Benchmark Compression via BestRand（随机抽样几乎就够了吗？通过 BestRand 实现稳定、低成本的基准压缩）[01:53] 🎨 TailSteer: Trajectory-Tail Steering for Drift-Resistant Multi-turn Image Editing（TailSteer：通过轨迹尾段引导实现抗漂移的多轮图像编辑）[02:24] 🖼️ WeLM-Image: A Unified Image Generation Model towards Visual Intelligence（WeLM-Image：迈向视觉智能的统一图像生成模型）[02:59] 🤖 SLA: Scaling Spatial-Language-Action Model via Continual Pre-training（SLA：通过持续预训练扩展空间—语言—动作模型）[03:34] 🧠 EvoMemState: Evolving Memory State for Personal Agents（EvoMemState：为个人智能体演化记忆状态）[04:08] 🔎 VisSearch-Bench: Evaluating Long Horizon Visual Search in Multimodal Agents（VisSearch-Bench：评估多模态智能体的长程视觉搜索）[04:38] ⚖️ Towards Foundation Decision Models: Drex（迈向基础决策模型：Drex）[05:08] 🌉 On the Opportunities and Challenges of Prefill-to-Autoregressive Cross-Model Transfer（论预填充到自回归的跨模型迁移：机遇与挑战）[05:37] 👤 EDGAR: Efficient Diffusion-Driven Gaussian Avatar Refinement via Synthetic Failure（EDGAR：通过合成失败实现高效的扩散驱动高斯头像细化）[06:05] 📝 HeLM: Hierarchical Explicit Language Memory for Non-Markovian Manipulation（HeLM：用于非马尔可夫操作的分层显式语言记忆）[06:37] 🎬 Future Representation Distillation for Imagination-free World Action Models（用于免想象世界动作模型的未来表征蒸馏）[07:11] 🤝 From Autonomous Agents to Cognitive Symbionts: Toward AI-Augmented Human Intelligence（从自主智能体到认知共生体：迈向 AI 增强的人类智能）[07:42] 🎧 Decode Before You Hear: Speculative Listening for Chunk-Interleaved Full-Duplex Speech Language Models（听到之前先解码：面向分块交错全双工语音语言模型的推测式聆听） 【关注我们】您还可以在以下平台找到我们，获得播客内容以外更多信息小红书: AI速递 在小宇宙查看该单集文稿

  4. 3d ago

    【月末特辑】10月最火AI论文 | 把经验变成预见；让音画同步生成

    ⏱️ Highlights[00:28] 🔮 Self-Retrospection Distillation: Turning Post-hoc Experiences into Prior Foresight（自我回顾蒸馏：将事后经验转化为事前预见）[01:38] 🎬 Kandinsky 6.0 Video: Foundation Models for Synchronized Video and Audio Generation（Kandinsky 6.0 Video：用于同步视频与音频生成的基础模型）[02:49] 🌳 The Trinity:Capability-Driven Recursive Self-Improvement for Multimodal Agents（The Trinity：多模态智能体的能力驱动递归自我改进）[03:59] 🦾 PDRVLA: A Vision-Language-Action Model with Physical Dynamics Representations（PDRVLA：具有物理动力学表征的视觉—语言—动作模型）[05:09] 🧭 The Teacher Is a Direction, Not a Destination: Extrapolating RL-Induced Representation Residuals in On-Policy Distillation（教师是方向，而非终点：在同策略蒸馏中外推强化学习诱导的表征残差）[06:21] 🔍 False Frontiers: Diagnosing and Mitigating Co-Cheating in Self-Evolving Search Agents（虚假前沿：诊断与缓解自我进化搜索智能体中的共同作弊）[07:32] ⚖️ Jeva: System One Models from First Principles（Jeva：从第一性原理构建系统一模型）[08:43] 🧠 MemT-1.0: Towards Memory-Native Evolution（MemT-1.0：迈向记忆原生进化）[09:53] 🤖 Has GPT-6-Astra Achieved Embodied Intelligence?（GPT-6-Astra已经实现具身智能了吗？）[11:03] 🎨 UniEvo-VL: An On-policy Self-Distillation Training Recipe for Multimodal Model Self-improvement（UniEvo-VL：用于多模态模型自我改进的同策略自蒸馏训练配方） 【关注我们】您还可以在以下平台找到我们，获得播客内容以外更多信息小红书: AI速递 在小宇宙查看该单集文稿

  5. 4d ago

    2026.11.05 | 动作读取视觉层级；分钟级游戏世界数据

    ⏱️ Highlights[00:27] 🦾 From Pixels to Actions - Towards Native Vision-Language-Action Policies（从像素到动作——迈向原生视觉语言动作策略）[01:05] 🎮 GameVerse: A Minute-Scale Gameplay Dataset for Long-Horizon Interactive World Modeling（GameVerse：面向长时域交互世界建模的分钟级游戏数据集）[01:41] 💬 Turns Matter: Advancing Streaming Video RL via Turn-Level Credit（轮次很重要：通过轮级信用分配推进流式视频强化学习）[02:17] 🔬 FrontierPhysics: Evaluating Agents for End-to-End Physics Research（FrontierPhysics：评估端到端物理研究智能体）[02:57] 🧭 What Does Multi-Turn Reinforcement Learning Really Improve in LLM Agents?（多轮强化学习究竟改善了大语言模型智能体的什么能力？）[03:36] 🏭 PyRUA-DataFarm: LLM Agent as a Cost-Efficient Teleoperator for Real-Ready Robot Data（PyRUA-DataFarm：以大语言模型智能体作为低成本遥操作员，生成可用于实机的机器人数据）[04:15] 🎭 EmoReward: Self-Evolving Reward Calibration for Emotional Video Generation RL（EmoReward：用于情绪视频生成强化学习的自演化奖励校准）[04:52] 🎙️ StreamDuo: A Live Commentator and a Research Agent for Unseen Games（StreamDuo：面向未见游戏的实时解说员与研究智能体）[05:33] 🧩 Solv3R: 3D Looped Transformer as a Multi-view Solver（Solv3R：作为多视图求解器的三维循环Transformer）[06:13] 🤝 Human-to-Robot Transfer via Bridging Representation and Action Space in VLAs（通过桥接VLA的表征与动作空间实现人到机器人的迁移）[06:51] 🖼️ VA-TDM: A Unified View for Improving Few-Step Diffusion Models by Velocity Adversarial Distillation（VA-TDM：通过速度对抗蒸馏改进少步扩散模型的统一视角）[07:31] 💾 MUON-RT Reusing the Teacher's Trajectory for Momentum-Free Muon for LLM Self-Distillation（MUON-RT：复用教师轨迹，为大语言模型自蒸馏实现无动量Muon）[08:09] 🏠 AnchorComplete: Recovering Complete Scene Meshes from Partial Point Clouds（AnchorComplete：从局部点云恢复完整场景网格）[08:48] 🧠 VideoMemBench: How Well Do MLLMs Remember in Video Conversations?（VideoMemBench：多模态大语言模型在视频对话中记得有多好？）[09:24] 🌿 SpecOPD: Branch-Parallel On-Policy Distillation for Speculative Decoding（SpecOPD：用于推测解码的分支并行同策略蒸馏） 【关注我们】您还可以在以下平台找到我们，获得播客内容以外更多信息小红书: AI速递 在小宇宙查看该单集文稿

  6. 5d ago

    2026.11.04 | 为模型与策略选路；让模拟来访者自然表达

    ⏱️ Highlights[00:27] 🧭 Route Anything: A Study of Model and Policy Routing for General-Purpose AI Systems（Route Anything：面向通用AI系统的模型与策略路由研究）[01:03] 💬 Words Caught in the Heart: Simulating Person-Specific Expression for Self-Opaque Clients in Psychotherapy（困在心中的话：为心理治疗中难以理解自身体验的来访者模拟个体化表达）[01:41] 🧠 Think Less, Locally: Local Preference Learning for Reasoning Compression（局部少想一点：用于推理压缩的局部偏好学习）[02:16] 🎬 MoviEdit: Long-form Multi-Shot Video Editing Made Easy（MoviEdit：让长篇多镜头视频编辑更轻松）[02:53] 🎞️ Behind Video Generation: Observations and Insights on Multimodal Understanding from Kling（视频生成的背后：来自Kling的多模态理解观察与洞见）[03:30] 🦾 DreamX: Video Diffusion Model is a Universal Latent Action Learner（DreamX：视频扩散模型是一种通用潜动作学习器）[04:05] 🤖 RoboColosseum: Evaluating Physical Intelligence at Scale In-The-Wild（RoboColosseum：在真实开放环境中规模化评估物理智能）[04:42] 🔎 Factorized MLLM Embeddings for Composed Image Retrieval（用于组合式图像检索的因子化多模态大语言模型嵌入）[05:19] 🛠️ Axion 1.0: Open Agent Foundation Models, Simplified（Axion 1.0：化繁为简的开放智能体基础模型）[05:54] 🔗 ToolSpace: Scaling Tool-Centric Environment Synthesis for Long-Horizon Agents（ToolSpace：为长程智能体扩展以工具为中心的环境合成）[06:32] 🗂️ YANAMI: A Large-Scale Multimodal Dataset and Benchmark for Multiturn Stateful Interaction（YANAMI：用于多轮有状态交互的大规模多模态数据集与基准）[07:10] 🤝 EvoTeam: Multi-Agent Collaboration via Evolving Teams（EvoTeam：通过演化团队实现多智能体协作）[07:47] 📽️ VidUEU-Agent: A Data-Curation Agent for Multimodal Understanding, Editing, and Unified Tasks（VidUEU-Agent：面向多模态理解、编辑与统一任务的数据策展智能体）[08:26] 🌐 WebHarbor: Docking Real Websites for Evolving GUI Agent Environments（WebHarbor：将真实网站接入可演进的GUI智能体环境）[09:04] ⚖️ ParetoSaddler: Pareto-Guided Optimization of Agent Harnesses for Quality and Cost（ParetoSaddler：以帕累托方法引导智能体运行框架的质量与成本优化） 【关注我们】您还可以在以下平台找到我们，获得播客内容以外更多信息小红书: AI速递 在小宇宙查看该单集文稿

  7. 6d ago

    2026.11.03 | 视频生成波前流水线；视频智能体的反馈边界

    ⏱️ Highlights[00:26] 🌊 Wave Forcing: Towards Speed-of-Light Streaming Video Generation（Wave Forcing：迈向光速的流式视频生成）[01:01] 🎬 Video Generation Meets Agent Harness: Agentic Video Generation, Agentic World Modeling and Recursive Self-Improvement（当视频生成遇上智能体运行框架：智能体式视频生成、智能体式世界建模与递归自我改进）[01:40] 🤝 CoworkOS:Scaling Agent Coworkers for Real-world Automation（CoworkOS：扩展面向现实世界自动化的智能体同事）[02:18] 🦾 Democratizing Robot-Use Agents: A Unified Harness with Open Models（让机器人使用智能体更普及：采用开放模型的统一运行框架）[02:55] 🖥️ CUA-Master: Scaling Computer-use Agents through Open-World Recursive Improvements（CUA-Master：通过开放世界递归改进扩展计算机使用智能体）[03:33] 🧭 NorthCUA Technical Report: Turbocharging Computer-Use Agents through Experience at Scale（NorthCUA 技术报告：通过规模化经验加速计算机使用智能体）[04:13] ↩️ Lost in Recovery: Demystifying Recovery Behavior of Computer-Use Agents via Goal-Oriented Action Replay（迷失于恢复：通过目标导向动作重放揭示计算机使用智能体的恢复行为）[04:51] 👁️ Reviso: Not All Visual Evidence Deserves Equal Attention in Video Captioning（Reviso：视频描述中，并非所有视觉证据都值得同等关注）[05:30] 🧩 From Fragmentation to Fusion: A Survey and Roadmap for Omnimodal Embeddings（从碎片化到融合：全模态嵌入的综述与路线图）[06:07] 💻 CrossSWE: Benchmarking Cross-Repository Joint Development for Coding Agents（CrossSWE：评测编程智能体的跨仓库联合开发）[06:48] 📷 PhotoGuider: Composition-Aware Cropping and 360^{circ} View Framing with Explainable Photographic Assessments（PhotoGuider：结合可解释摄影评估的构图感知裁剪与360°视角取景）[07:25] 🙋 Avatar Intelligence: Amplifying Human Agency Through Persistent Personal Agents（Avatar Intelligence：通过持久个人智能体增强人的自主行动能力）[08:05] 🔢 Diagnosing the Reference Gap: What Visual Scaffolds Can and Cannot Fix in Multimodal Reasoning（诊断指代鸿沟：视觉支架在多模态推理中能与不能修复什么）[08:42] 📑 DecisionRank: On-Policy Distillation for Decision-Model Rerankers（DecisionRank：面向决策模型重排器的在策略蒸馏）[09:21] 🔎 DeFT-RAG: Decomposing Multi-Hop Questions into Self-Contained Sub-Questions with Reusable Answers（DeFT-RAG：将多跳问题分解为答案可复用的自足子问题） 【关注我们】您还可以在以下平台找到我们，获得播客内容以外更多信息小红书: AI速递 在小宇宙查看该单集文稿

  8. 6d ago

    【周末特辑】10月第5周最火AI论文 | 把物理预测学进行动；多视图视频补上几何

    ⏱️ Highlights[00:29] 🤖 PDRVLA: A Vision-Language-Action Model with Physical Dynamics Representations（PDRVLA：具有物理动力学表征的视觉–语言–动作模型）[01:58] 🎥 OmniShowcase: Omnidirectional Subject-Consistent Video Generation from Multi-View Images（OmniShowcase：基于多视图图像的全方位主体一致视频生成）[03:27] 🎙️ Efficient Streaming Audio-Visual Target Speaker Extraction for Real-World Acoustic Scenes（面向真实世界声学场景的高效流式音视目标说话人提取）[05:02] 👁️ Look Only When Needed: Certificate-Guided Visual Routing for Efficient Diffusion VLMs（只在需要时看图：以凭据引导视觉路由的高效扩散视觉语言模型）[06:39] 🌍 Hy-Nomas: Towards Free-form Interactive World Generation（Hy-Nomas：迈向自由形式的交互式世界生成） 【关注我们】您还可以在以下平台找到我们，获得播客内容以外更多信息小红书: AI速递 在小宇宙查看该单集文稿

See All (739)

Ratings & Reviews

5
out of 5
2 Ratings

About

每天10分钟，带您快速了解当日HuggingFace热门AI论文内容。每个工作日更新，欢迎订阅。 📢播客节目在小宇宙、Apple Podcast平台搜索【HuggingFace 每日AI论文速递】 🖼另外还有图文版，可在小红书搜索并关注【AI速递】

Information

You Might Also Like