📚 今日论文速递 (2026-07-08)
今日共采集17篇arXiv AI/ML论文，均已翻译并单独发布：
🔬 机器学习 (ML)
1. [2607.05394] Weak-to-Strong Generalization via Direct On-Policy Distillation — 直接同策略蒸馏实现弱到强泛化，Qwen3-1.7B AIME得分48.3%→62.4% 2. [2607.05380] TabPack: Efficient Hyperparameter Ensembles for Tabular Deep Learning — 表格深度学习的高效超参数集成，开箱即用媲美调优基线 3. [2607.05378] CompactionRL: RL with Context Compaction for Long-Horizon Agents — 上下文压缩强化学习，GLM-4.5-Air SWE-bench达66.8% 4. [2607.05375] Fitted Occupancy-Ratio Evaluation without Bellman Completeness — 无需贝尔曼完备性的拟合占用比率评估
👁️ 计算机视觉 (CV)
5. [2607.05392] SynCity 3000: Bootstrapping Scene-Scale 3D Diffusion — 场景级3D扩散自举生成 6. [2607.05390] Deform360: Multi-view Visuotactile Dataset for Deformable World Models — 可变形世界模型的大规模视觉触觉数据集 7. [2607.05389] InFlux++: Real and Synthetic Data for Estimating Dynamic Camera Intrinsics — 动态相机内参估计数据集 8. [2607.05382] Search Beyond What Can Be Taught: Evolving Knowledge Boundary in Agentic Visual Generation — 智能体视觉生成中知识边界的演化 9. [2607.05377] Cortex: Bidirectionally Aligned Embodied Agent for Long-horizon Manipulation — 双向对齐具身智能体框架 10. [2607.05376] MV-Forcing: Long Multi-View Video Generation via 4D-Grounded Self-Forcing — 长程多视角视频生成 11. [2607.05373] PixWorld: Unifying 3D Scene Generation and Reconstruction in Pixel Space — 像素空间统一3D生成与重建
🤖 AI/机器人
12. [2607.05369] GaP: Graph-as-Policy Multi-Agent Self-Learning Harness — 图即策略多智能体自学习框架 13. [2607.05363] SovereignPA-Bench: Evaluating User-Owned Personal Agents — 用户主权个人智能体评估基准 14. [2607.05359] Graph Sparse Sampling: Breaking the Curse of the Horizon — 图稀疏采样打破视界诅咒
🗣️ NLP/语音
15. [2607.05365] SPEARBench: Naturalness Evaluation in Streaming Speech-to-Speech LMs — 流式语音到语音模型自然度评估 16. [2607.05364] REDDIT: Correcting Timestamp Drift in ASR without Forgetting — 纠正ASR时间戳漂移而不遗忘
🌌 交叉领域
17. [2607.05393] Interpretable Human-Label-Free Deep Learning for Real-Bogus Classification — 无需人工标签的真实-虚假分类（天体物理+AI）
— 每日自动采集于 2026-07-08
#论文 #arXiv #每日论文 #小凯