cs.LG（2026-10-27）

📊 共 44 篇论文 | 🔗 10 篇有代码

🎯 兴趣领域导航

支柱二：RL算法与架构 (RL & Architecture) (20 🔗2) 支柱九：具身大模型 (Embodied Foundation Models) (18 🔗6) 支柱八：物理动画 (Physics-based Animation) (2) 支柱一：机器人控制 (Robot Control) (2) 支柱五：交互与反应 (Interaction & Reaction) (1 🔗1) 支柱三：空间感知与语义 (Perception & Semantics) (1 🔗1)

🔬 支柱二：RL算法与架构 (RL & Architecture) (20 篇)

#题目一句话要点标签🔗⭐
1 Understanding the limits of latent world models 提出忠实潜在世界模型以解决环境动态建模问题 world model world models latent dynamics
2 MLE-World: Predicting Machine Learning Experiment Outcomes with a Learned World Model 提出MLE-World以预测机器学习实验结果，提升测试时间效率 world model world models ✅
3 Repulsive Population Policy Gradients for Behaviorally Diverse Reinforcement Learning 提出RePPO以解决行为多样性强化学习问题 reinforcement learning PPO large language model
4 On the Sparsity of Direct Preference Optimization: Weight Disentanglement in the NTK Regime 提出双稀疏性概念以解释直接偏好优化中的稀疏参数更新 RLHF DPO direct preference optimization
5 Live-OPSD: On-Policy Self-Distillation for Proactive Streaming Large Video Models 提出Live-OPSD以解决流媒体助手响应决策问题 distillation multimodal
6 CoHARL: Co-Evolving Agent Policies and Executable Harnesses with Reinforcement Learning 提出CoHARL框架以优化强化学习中的策略与执行工具的协同演化 reinforcement learning large language model
7 Exploring the Limits of the Delta Learning Hypothesis: Single-Pair Generator Biases in Direct Preference Optimization 探讨单对生成器偏差在直接偏好优化中的极限 DPO direct preference optimization
8 Hyperparameter and Neural Architecture Search via Evolutionary Optimization for Reinforcement Learning: HPNAS-RL 提出双分支进化框架以优化强化学习中的神经架构与超参数 reinforcement learning teacher-student
9 EnvSched: Environment-Aware Scheduling for Rollout-Efficient Agent Reinforcement Learning 提出EnvSched以解决有限预算下的强化学习调度问题 reinforcement learning
10 Graph Representation Learning via Adaptive Expectile Aggregation 提出ExpectileGCN以解决异质图中的聚合信息不足问题 representation learning
11 How Many and How Much? Contrastive Learning with Negative Samples for Knowledge Graph Embedding 提出RADAR框架以优化知识图谱嵌入中的负样本学习 contrastive learning
12 Action-Value Calibration via Continuation Sharing for Long-Horizon Agentic Reinforcement Learning 提出ConCal以解决长时间序列强化学习中的信用分配问题 reinforcement learning ✅
13 Physics-Guided Flow Matching: A Generative AI Framework for Global Total Electron Content Forecasting 提出物理引导流匹配框架以提高全球电子总量预测精度 flow matching
14 HiLo-RD: Cross-Budget Self-Distillation for Masked Diffusion Language Models 提出HiLo-RD以解决有限生成预算下的MDLMs问题 distillation
15 When All Rollouts Fail: Policy-Compatible Joint Decoding for Hard-Problem Reasoning 提出PICO以解决强化学习中的困难问题推理 reinforcement learning large language model
16 SetGRPO: Calibrating Grader-Induced Policy Optimization 提出SetGRPO以解决评分者随机性对策略优化的影响 reinforcement learning large language model
17 Frequency-Uniform Log-Regret Reinforcement Learning for Diffusions 提出频率均匀的对数遗憾强化学习以解决扩散控制问题 reinforcement learning
18 Conservative Dueling Bandits: Preference Learning under Baseline Constraints 提出保守对决赌博算法以解决用户偏好学习中的基线约束问题 preference learning
19 Balanced State Subspaces for Mamba-2 State Compression 提出平衡状态子空间方法以优化Mamba-2状态压缩 Mamba
20 Interpretable Meta-Policies for Neural Decision Making 提出NEXUS框架以解决深度强化学习的可解释性问题 reinforcement learning deep reinforcement learning

🔬 支柱九：具身大模型 (Embodied Foundation Models) (18 篇)

#题目一句话要点标签🔗⭐
21 When Visual Evidence Is Insufficient: How Instructions, Post-Training, and Internal Activations Shape Refusal in Multimodal Language Models 提出多模态语言模型拒绝机制的优化策略 large language model multimodal
22 EmbeddingGemma 2: Unified and Efficient Multimodal Representations 提出EmbeddingGemma 2以实现统一高效的多模态表示 multimodal
23 PRISM: Pathology Reliability In Scarce-label Medicine. A Tile-level Benchmark of Pathology Foundation Models 提出PRISM以解决病理基础模型在稀缺标签医学中的可靠性问题 foundation model ✅
24 CoS-VQ: Coordinated Shared-Scale Vector Quantization for Large Language Models 提出CoS-VQ以解决大语言模型的低比特权重量化问题 large language model ✅
25 MM-Net: A Multimodal Multi-Task Network for Sleep Staging and Sleep Apnea Assessment After Ischemic Stroke 提出MM-Net以解决缺血性中风后睡眠分期与呼吸暂停评估问题 multimodal
26 Scaling Laws for Tool-Use Agents via Controllable Environment Synthesis 提出SynthAPI以优化工具使用代理的训练效果 large language model
27 Follow the Gradient: Tracing Safety Degradation Back to Training Data 提出SimpleLoRIF以解决大语言模型安全性下降问题 large language model
28 TRIDENT-Seq: Profile-Matched Prefix-Future Contrast for Label-Efficient Payment Risk Modeling 提出TRIDENT-Seq以解决支付风险建模中的标签效率问题 foundation model
29 SimplexLoRA: Dynamic Rank Updating via Alternating Minimization with Simplex Projection 提出SimplexLoRA以动态调整适配器秩解决参数效率问题 large language model ✅
30 CaLLM: Continual Adaptation of LLMs via Non-Parametric Adapter Routing 提出CaLLM以解决大语言模型的持续适应问题 large language model
31 Advancement or Acceleration? How Data Curation Improves LLM Training 提出数据策划方法以提升大语言模型训练效果 large language model
32 LLM knowledge for improved Out-of-Distribution Tabular Prediction - A Cross-Validated Regularisation Approach 利用大语言模型知识改善分布外表格预测 large language model ✅
33 Robust LLM Data Selection via Content-Invariant Gradient Attribution 提出内容不变梯度归因方法以解决数据选择不稳定问题 large language model
34 Dream-COP: Guiding Algorithm Discovery for Combinatorial Optimization Problems in an Experience World 提出Dream-COP以优化组合优化问题的算法发现 large language model
35 Recall or Relearn? A Mixture of In-Context Experts for Recurring Concept Drift 提出MICE以解决重复概念漂移问题 foundation model
36 Less is more: learning what to profile for GPU kernel optimization 提出AutoProfiler以解决GPU内核优化中的动态剖面选择问题 large language model ✅
37 Learning from Points and Patches: A Pre-trained Hybrid Attention Transformer for Time Series Analysis 提出TimeHAT以解决异构时间序列分析问题 zero-shot transfer ✅
38 NextEval: An Agent-Based Solver and Benchmark for Derivative-Free Optimization 提出NextEval以解决无导数优化问题 large language model

🔬 支柱八：物理动画 (Physics-based Animation) (2 篇)

#题目一句话要点标签🔗⭐
39 GAS-PFN: Graph-Aware Spatiotemporal Prior-Data Fitted Networks for Zero-Shot Traffic Forecasting 提出GAS-PFN以解决零样本交通预测中的图结构依赖问题 spatiotemporal
40 Quantum Evolution Kernel for Node-Attributed Graphs 提出量子演化核以解决节点属性图的分类问题 PULSE

🔬 支柱一：机器人控制 (Robot Control) (2 篇)

#题目一句话要点标签🔗⭐
41 When Is a Relation Worth Trusting? Manipulation Costs, Exposure, and the Trust Trap in Multi-Relational Graph Learning 提出多关系图学习中的信任陷阱解决方案 manipulation
42 Quantifying and Prioritizing Stealthy Source-Manipulation Threats Against Multi-Source Anomaly Detectors 提出任务条件可见性边际以量化多源异常检测的隐蔽源操控威胁 manipulation

🔬 支柱五：交互与反应 (Interaction & Reaction) (1 篇)

#题目一句话要点标签🔗⭐
43 V-LUMEN: Visual Lookup Memory for Embedding Scaling in Vision-Language Models 提出V-LUMEN以解决视觉语言模型的嵌入扩展问题 CHOIS ✅

🔬 支柱三：空间感知与语义 (Perception & Semantics) (1 篇)

#题目一句话要点标签🔗⭐
44 Generative Refinement Learning for Temporal Interpolation in Scientific Machine Learning 提出iFLEX以解决科学机器学习中的时间插值问题 optical flow ✅

⬅️ 返回 cs.LG 首页 · 🏠 返回主页