InfoVis：统一的幻灯片图像生成与编辑#
幻灯片生成与编辑需同时做到准确文字渲染、指令遵循、视觉定位、版式美感与精确编辑。作者定义四项幻灯片任务，构建含 1300 多万条持续预训练数据和 190 万条监督数据的数据引擎及诊断评测，训练统一模型。
▲ 28 AI 摘要 arXiv GitHub Jiahao Wang, Xiyan Jiang, Chenyuan Wu, Haoqiang Zhang · Beijing Academy of Artificial Intelligence
AI 深研院 · 共 8 条
AI 深研院 · 论文
Hugging Face Daily Papers 与 arXiv 精选
Data World：从零扩展数据分析智能体的训练环境#
数据智能体训练受环境构建成本高、隐私与版权限制和开放式结论难验证所困。Data World 先设定文件背后的“世界”再生成文件，用多智能体引擎合成环境、任务与可验证答案，以规模化训练数据分析智能体。
▲ 28 AI 摘要 arXiv GitHub Zhanli Li, Zeyuan Xu, Chaoxu Pang, Zeyuan Liu · LongCat
MFVSR：多风格细粒度视频字幕去除基准#
现有视频字幕去除评测分辨率低、测试集零散，且沿用通用视频修复指标。MFVSR 构建大规模多风格基准，针对局部修复错误、误改、时序不稳和残留等问题设计细粒度、可靠性导向的评估。
▲ 12 AI 摘要 arXiv GitHub Xiyan Huang, Qingdong He, Peng Tang, Bo Yin
EdgeWise: Bringing Long-Horizon Agents to the Edge via Selective Cloud Delegation#
Complex and long-horizon agent tasks often rely on frontier models deployed in the cloud to achieve strong performance. However, using these models for every step of a long-horizon trajectory incurs substantial inferenc…
▲ 26 未翻译 arXiv Zeru Shi, Jindong Jiang, Ryo Hachiuma, Yonggan Fu · NVIDIA
Terminal-RSI: Recursive Self-Improving Terminal Agents via Recursive Self-Rewrite and Policy-Refine#
Converting verifier outcomes into useful policy updates and partial progress into complete task resolution is crucial for terminal agents. We present Terminal-RSI, which alternates Recursive Self-Rewrite (RSR) of verifi…
▲ 26 未翻译 arXiv Junyao Yang, Zongxia Li, Yucheng Shi, Zhongzhi Li · Tencent
Intern-Decision: Efficient Multimodal Decision Making with Autoregressive Masked Language Modeling#
Interactive agents repeatedly choose among a finite set of actions, labels, or ordinal outcomes. Generating a free-form response for each choice adds latency and complicates the interface between a model and its environ…
▲ 25 未翻译 arXiv GitHub Tong Zhu, Zhe Xu, Zefeng He, Yunkai Chai · Intern Large Models
Routing Policy Optimization for Online Diffusion Reinforcement Learning#
Reinforcement learning can align visual generators with reward signals beyond their pretraining objectives, but existing score-function methods typically introduce stochasticity by perturbing the reverse denoising traje…
▲ 23 未翻译 arXiv Qiuhong Shen, Yiding Yang, Yuxin Zhang, Xinwei Huang · National University of Singapore
Trajectory-Level Speculation with Semantic Verification for Tool-Using Agents#
Large language model agents solve tasks through long sequences of tool calls, and the prevailing design has a single strong model generate every step, although many steps are routine enough for a much smaller model. Exi…
▲ 21 未翻译 arXiv GitHub Zilin Wang, Yinhong Liu, Chao Chen, Qun Liu
更多 34 条未进正文的候选，按重要度排序；默认折叠
- Velox：用空间记忆让视频世界模型保持高效HF Daily Papers · 10月15日
- SWE-at-Scale: Scalable Issue Synthesis from Historical Bug FixesHF Daily Papers · 10月15日
- Do Better Visual Representations Always Lead to Better End-to-End Autonomous Driving?HF Daily Papers · 10月7日
- SimVTA: Simulation-Grounded Video-Trajectory Alignment for Driving World Action ModelsHF Daily Papers · 10月9日
- WorldMemBench: A Comprehensive Benchmark for Memory of Video World ModelsHF Daily Papers · 10月13日
- Athanor: An Agent-Native System for Autonomous Post-Training ResearchHF Daily Papers · 10月15日
- Virtual Head AttentionHF Daily Papers · 10月13日
- 在去中心化智能体群中学习自组织HF Daily Papers · 10月15日
- CarryOn: A Dual-Loop Harness for Personalized and Reliable Robot AssistanceHF Daily Papers · 10月15日
- FlashMask-3: Efficient and Expressive Mask-Aware Distributed AttentionHF Daily Papers · 10月13日
- CoDA：数据引擎与动作策略协同进化HF Daily Papers · 10月15日
- Beyond Policy Rollouts: Embedding-Space Directional Exploration with Anchored Noise for LLM RLVRHF Daily Papers · 10月12日
- Cowork-Gym: Synthesizing Interactive Work Environments at ScaleHF Daily Papers · 10月13日
- LEASH: Uncertainty-Aware Weak-to-Strong Escalation and Oversight for Computer-Use AgentsHF Daily Papers · 10月15日
- SUPRA-Tex：超对齐纹理的单图 3D 生成HF Daily Papers · 10月15日
- Memory as Weights: Internalizing Long-Term History for Streaming VideosHF Daily Papers · 10月15日
- Calibrated Typed Decisions for Entity, Relation, and Event ExtractionHF Daily Papers · 10月12日
- Gen4P：视频生成器可作前馈 4D 感知器HF Daily Papers · 10月15日
- CLIFT: Conformal Self-Verification for Web Agent Training and Test-Time ScalingHF Daily Papers · 10月5日
- ANTE: Anticipate, Then Execute -- Learning Internal World Models for Coding Agents from Verified BeliefsHF Daily Papers · 10月11日
- LeanPRM: Lean-Grounded Process Supervision for Natural Language Proof VerificationHF Daily Papers · 10月13日
- Copy-Paste Language ModelsHF Daily Papers · 10月13日
- SERO: Learning from Executable Oracle Continuations for Long-Horizon Agentic Reinforcement LearningHF Daily Papers · 10月14日
- Mine, Sample, Render: Test-Time Diversity Alignment Of Generative ModelsHF Daily Papers · 10月15日
- MMSearch-OPD: Easy Made Hard, Hard Made Learnable for Long-Horizon Multimodal Search AgentsHF Daily Papers · 10月15日
- Scalable Weak-to-Strong Verifier Hardening for Agent EnvironmentsHF Daily Papers · 10月11日
- Recursive Environment Scaling for General Long-Horizon AgentsHF Daily Papers · 10月12日
- PN-Mix: Pair-Norm Guided Mixed-Precision Key Cache Quantization for Efficient Long-Output LLM InferenceHF Daily Papers · 10月14日
- H-DDCM: Hadamard Codebooks for Fast Generative Image and Video CompressionHF Daily Papers · 10月13日
- Learning to Read Out: Unembedding Dynamics in Language Model PretrainingHF Daily Papers · 10月14日
- RepForcing: Representation-Space Initialization for Autoregressive Video DistillationHF Daily Papers · 10月14日
- PathFM: Fast Generative Forecasting of Multivariate TrajectoriesHF Daily Papers · 10月15日
- Explaining the Roles of Experts in Mixture-of-Experts LLMs for PersonalizationHF Daily Papers · 10月15日
- Skill-Decoupled On-Policy Expert Distillation for Unified Vision-Language ModelsHF Daily Papers · 10月15日
