Hugging Face每日论文

刚刚更新 · 去Hugging Face看原榜 → · 中文标题为机器翻译

1. 1SO-Arena：评测 System One 模型的能力谱系与可靠性SO-Arena: Benchmarking the Capability Spectrum and Reliability of System One Models136 赞
2. 2BirdDreams：面向合理且可反应驾驶世界模型的透视与鸟瞰视频联合生成BirdDreams: Joint Perspective and BEV Video Generation for a Plausible, Reactive Driving World Model50 赞
3. 3从结构化教科书中扩展数学证明监督Scaling Mathematical Proof Supervision from Structured Textbooks47 赞
4. 4PhysLoop：利用大语言模型的物理意识实现视频生成模型的自我改进PhysLoop: Harnessing LLMs' Physics Awareness for Self-Improving Video Generation Models46 赞
5. 5G-STAR：面向深度压缩自回归生成的分组缩放令牌容量G-STAR: Group-Scaled Token Capacity for Deeply Compressed Autoregressive Generation45 赞
6. 6Duo：面向全双工交互的互惠世界模型Duo: A Reciprocal World Model for Full-Duplex Interaction40 赞
7. 7SteerDistill：通过闭环教师引导量化大语言模型智能体SteerDistill: Quantizing LLM Agents via Closed-Loop Teacher Steering37 赞
8. 8Open Vision Banana：一套完全开放的生成式视觉感知方案Open Vision Banana: A Fully Open Recipe for Generative Visual Perception36 赞
9. 9行动者、裁判还是教师？理解长时程智能体中的决策模型Actor, Judge, or Teacher? Understanding Decision Models in Long-Horizon Agents34 赞
10. 10Ropedia Xperience 2：扩展物理 AI 环境与行为的黄金标准Ropedia Xperience 2: A Gold Standard for Scaling Physical AI Environments and Behaviors29 赞
11. 11论超连接的设计：简化、稳定性与可扩展性On the Design of Hyper-Connections: Simplification, Stability, and Scalability29 赞
12. 12TTB：面向仅解码器视图合成模型高效渲染的测试时烘焙TTB: Test-time Baking for Efficient Rendering of Decoder-only View Synthesis Models23 赞
13. 13少读多解：面向 AI 智能体的外部内容高效稀疏阅读Read Less, Solve More: Token-Efficient Sparse Reading of External Content for AI Agents22 赞
14. 14学习任务还是学习验证器？重新思考具身自我改进中的标准暴露问题Learning the Task or Learning the Verifier? Rethinking Criteria Exposure in Embodied Self-Improvement21 赞
15. 15解绑多模态视觉编码Unbundling Multimodal Vision Encoding16 赞
16. 16WorldDraft：如果代码构建世界，视频模型能否直接渲染它？WorldDraft: If Code Builds the World, Can Video Models Just Render It?16 赞
17. 17面向实用且可扩展 SOAP 的分解单侧 Adam 动量Factored One-sided Adam-Moment for Practical and Scalable SOAP14 赞
18. 18UniSpark：探索面向多模态理解与生成的混合注意力UniSpark: Exploring Hybrid Attention for Multimodal Understanding and Generation13 赞
19. 19Interactive Visual Gym：在交互式世界中用模拟用户评估视觉交互模型Interactive Visual Gym: Evaluating Visual Interaction Models with Simulated Users in Interactive Worlds13 赞
20. 20Evolving Infra：面向大规模部署的智能体自适应大语言模型服务Evolving Infra: Agent-Adaptive LLM Serving at Scale12 赞
21. 21谁记得什么？面向多用户、多智能体交互的多粒度记忆抽取Who Remembers What? Multi-Granularity Memory Extraction for Multi-User, Multi-Agent Interaction8 赞
22. 22思考的价值：在混合决策模型中学会何时推理The Value of Thinking: Learning When to Reason in Hybrid Decision Models7 赞
23. 23DeepStress：压力测试深度搜索智能体DeepStress: Stress-Testing Deep Search Agents7 赞
24. 24Agent-R^2：通过真实源数据合成扩展面向真实生产力工作的智能体Agent-R^2: Scaling Agents for Realistic Productivity Work via Real-Source Data Synthesis7 赞
25. 25视觉语言模型中的潜意识学习Subliminal Learning in Vision-Language Models7 赞
26. 26FAME-QP：用于实时模型预测控制的 GPU 加速二次规划求解器FAME-QP: A GPU-Accelerated Quadratic-Programming Solver for Real-Time Model Predictive Control6 赞
27. 27对齐在何处产生，又意味着什么？柏拉图表示假说的近距离审视Where Does Alignment Emerge and What Does It Mean? A Closer Look at the Platonic Representation Hypothesis6 赞
28. 28回到像素：将潜在视频扩散提升到像素空间Back to Pixels: Lifting Latent Video Diffusion into Pixel Space5 赞
29. 29经验交接强化学习：面向递归自我改进Experience-Handoff Reinforcement Learning for Recursive Self-Improvement5 赞
30. 30JustRUA：GPT-6 机器人使用智能体的受控研究JustRUA: A Controlled Study of GPT-6 Robot-Use Agents5 赞
31. 31人性化却仍然真实？审计 AI 文本人性化工具中的事实与检测Humanized but Still True? Auditing Facts and Detection in AI-Text Humanizers5 赞
32. 32按需精度：在小型本地模型不确定的解码步骤上提升权重位宽Precision on Demand: Escalating Weight Bits at Uncertain Decoding Steps in a Small Local Model5 赞
33. 33基于参考的无人机视频稳像中的共识捕获：动态城市场景基准Consensus Capture in Reference-Based UAV Video Stabilization: A Benchmark on Dynamic Urban Scenes4 赞
34. 34TurtLES：面向湍流三维神经 PDE 替代模型的大规模基准TurtLES: A Large-Scale Benchmark for Turbulent 3D Neural PDE Surrogates4 赞
35. 35HiVe-Mem：交错水平记忆扩展与垂直证据组合的多模态智能体HiVe-Mem: Interleaving Horizontal Memory Expansion and Vertical Evidence Composition for Multimodal Agents4 赞
36. 36面向具有模糊工具结果与部分执行的 AI 智能体的证据引导恢复Evidence-Guided Recovery for AI Agents with Ambiguous Tool Outcomes and Partial Execution4 赞
37. 37多模态模型能否对等价光谱作出一致判断？Do Multimodal Models Decide Consistently on Equivalent Optical Spectra?4 赞
38. 38从大语言模型到高效的 System One 模型：自适应计算之路From Large Language Models to Efficient System One Models via Adaptive Computation4 赞
39. 39逐层、逐模块探查 Transformer 表示Probing Transformer Representations Layer by Layer, Module by Module4 赞
40. 40叙事鸿沟：大语言模型能否帮助我们跨语言把握多样叙事？The Narrative Gap: Can LLMs Help Us Navigate Diverse Narratives Across Languages?3 赞
41. 41自我进化多模态智能体的跨场景综述与分析A Cross-Scenario Survey and Analysis of Self-Evolving Multimodal Agents3 赞
42. 42CARTBench：面向包含时间线的复杂模糊购物建议的评分细则评估CARTBench: Rubric-Based Evaluation of Shopping Advice for Complex and Ambiguous Requests with Timelines3 赞
43. 43比特还是排列？在小型本地语言模型中把固定推理预算花在权重精度还是选项顺序平均上Bits or permutations? Spending a fixed inference budget on weight precision or on option-order averaging in small local language models3 赞
44. 44混合信号：面向脑电与心电的权重空间编程Mixed Signals: Weight-Space Programming for EEG and ECG3 赞
45. 45面向图神经网络的模块感知优化Module-Aware Optimization for Graph Neural Networks3 赞
46. 46基于 Fisher 几何的结构化置信度扰动用于对抗训练Structured Confidence Perturbation via Fisher Geometry for Adversarial Training2 赞
47. 47ScrapeBench：评估 AI 智能体在网站抓取中的法律合规性ScrapeBench: Evaluating Legal Compliance of AI Agents in Website Scraping2 赞
48. 48PolyRoute：面向类型化决策路由的十二语言与代码混合基准，附基线与排行榜PolyRoute: A Twelve-Language and Code-Mixed Benchmark for Typed-Decision Routing, with Baselines and a Leaderboard2 赞
49. 49EgoX：借助具身迁移想象实现第一人称跨形态操作EgoX: Egocentric Cross-Embodiment Manipulation with Embodiment Dreaming1 赞
50. 50迈向人类神经影像的前沿机器智能A frontier machine intelligence for human neuroimaging1 赞