cs.LG（2024-10-07）

📊 共 41 篇论文 | 🔗 9 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (24 🔗7) 支柱二：RL算法与架构 (RL & Architecture) (13 🔗1) 支柱一：机器人控制 (Robot Control) (3 🔗1) 支柱八：物理动画 (Physics-based Animation) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (24 篇)

#题目一句话要点标签🔗⭐
1 ACDC: Autoregressive Coherent Multimodal Generation using Diffusion Correction 提出ACDC，结合自回归模型与扩散模型，实现高质量连贯的多模态生成。 large language model multimodal
2 fLSA: Learning Semantic Structures in Document Collections Using Foundation Models fLSA：利用基础模型学习文档集合中的语义结构，提升文本重建与生成质量。 large language model foundation model ✅
3 From Sparse Dependence to Sparse Attention: Unveiling How Chain-of-Thought Enhances Transformer Sample Efficiency 揭示CoT提升Transformer样本效率的机制：从稀疏依赖到稀疏注意力 large language model chain-of-thought
4 RespLLM: Unifying Audio and Text with Multimodal LLMs for Generalized Respiratory Health Prediction RespLLM：利用多模态LLM统一音频和文本，实现广义呼吸系统健康预测 large language model multimodal
5 Compression via Pre-trained Transformers: A Study on Byte-Level Multimodal Data 利用预训练Transformer实现高效数据压缩，超越传统算法。 foundation model multimodal
6 Wireless-Friendly Window Position Optimization for RIS-Aided Outdoor-to-Indoor Networks based on Multi-Modal Large Language Model 提出基于多模态大语言模型的无线友好型窗户位置优化方法，用于RIS辅助的室外到室内网络。 large language model
7 PrefixQuant: Eliminating Outliers by Prefixed Tokens for Large Language Models Quantization PrefixQuant通过前缀化异常token解决LLM量化中的token级离群点问题 large language model ✅
8 GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models GSM-Symbolic：揭示大语言模型在数学推理上的局限性 large language model
9 Recent Advances of Multimodal Continual Learning: A Comprehensive Survey 首个多模态持续学习综述，系统梳理方法并展望未来方向。 multimodal ✅
10 TLDR: Token-Level Detective Reward Model for Large Vision Language Models 提出TLDR：一种Token级别判别奖励模型，提升大型视觉语言模型性能。 large language model multimodal
11 AnyAttack: Towards Large-scale Self-supervised Adversarial Attacks on Vision-language Models AnyAttack：面向视觉-语言模型的大规模自监督对抗攻击框架 foundation model multimodal
12 Chain and Causal Attention for Efficient Entity Tracking 提出链式与因果注意力机制，高效解决Transformer在实体追踪任务中的局限性 large language model
13 Transformers learn variable-order Markov chains in-context 研究Transformer上下文学习变阶马尔可夫链能力，并提出CTW算法的Transformer构造。 large language model
14 Transformers are Efficient Compilers, Provably 证明Transformer能以对数复杂度高效编译类C语言，优于RNN large language model
15 SecAlign: Defending Against Prompt Injection with Preference Optimization SecAlign：利用偏好优化防御大语言模型的提示注入攻击 large language model ✅
16 Can LLMs Understand Time Series Anomalies? 探索LLM在时间序列异常检测中的能力，揭示其理解机制与局限性 large language model ✅
17 Density estimation with LLMs: a geometric investigation of in-context learning trajectories 利用LLM进行密度估计：上下文学习轨迹的几何分析 large language model ✅
18 Interactive Event Sifting using Bayesian Graph Neural Networks 提出基于贝叶斯图神经网络的交互式事件筛选方法，用于法庭分析中社交媒体数据的快速过滤。 multimodal
19 Model-GLUE: Democratized LLM Scaling for A Large Model Zoo in the Wild Model-GLUE：面向大规模模型库的普适性LLM扩展方案 large language model ✅
20 TidalDecode: Fast and Accurate LLM Decoding with Position Persistent Sparse Attention TidalDecode：利用位置持久稀疏注意力加速LLM解码并保持精度 large language model
21 TOAST: Transformer Optimization using Adaptive and Simple Transformations TOAST：利用自适应简单变换优化Transformer，无需额外训练。 foundation model
22 Strong Model Collapse 揭示大规模模型训练中由合成数据引起的强模型崩溃现象 large language model
23 Generating CAD Code with Vision-Language Models for 3D Designs 提出CADCodeVerify，利用视觉-语言模型迭代验证并改进CAD代码生成的3D对象。 large language model
24 Deeper Insights Without Updates: The Power of In-Context Learning Over Fine-Tuning ICL在隐式模式学习上优于微调，无需模型更新即可获得更深层次的理解 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (13 篇)

#题目一句话要点标签🔗⭐
25 A Predictive and Optimization Approach for Enhanced Urban Mobility Using Spatiotemporal Data 提出时空数据驱动的预测与优化方法，提升城市交通效率。 predictive model spatiotemporal
26 HERO: Human-Feedback Efficient Reinforcement Learning for Online Diffusion Model Finetuning 提出HERO框架以高效利用人类反馈进行在线扩散模型微调 reinforcement learning representation learning ✅
27 Meta-Dynamical State Space Models for Integrative Neural Data Analysis 提出元动态状态空间模型，用于整合神经数据分析并快速学习潜在动态 latent dynamics state space model
28 Better than Your Teacher: LLM Agents that learn from Privileged AI Feedback LEAP：利用特权AI反馈提升LLM Agent决策能力 behavior cloning privileged information large language model
29 Efficient Model-Based Reinforcement Learning Through Optimistic Thompson Sampling 提出基于乐观Thompson采样的模型强化学习方法，提升机器人稀疏奖励环境探索效率。 reinforcement learning
30 Unsupervised Representation Learning from Sparse Transformation Analysis 提出基于稀疏变换分析的无监督表征学习方法，用于解耦序列数据中的潜在因素。 representation learning
31 SimO Loss: Anchor-Free Contrastive Loss for Fine-Grained Supervised Contrastive Learning 提出SimO损失的无锚点对比学习方法，实现细粒度监督对比学习 contrastive learning
32 DOPL: Direct Online Preference Learning for Restless Bandits with Preference Feedback 提出DOPL算法以解决带偏好反馈的Restless Bandits问题 preference learning
33 Progressive distillation induces an implicit curriculum 渐进式蒸馏通过隐式课程学习加速学生模型训练 distillation
34 Hyper-Representations: Learning from Populations of Neural Networks 提出超表示学习方法，从神经网络群体中学习任务无关的通用权重表示。 representation learning foundation model
35 FreSh: Frequency Shifting for Accelerated Neural Representation Learning FreSh：通过频率偏移加速神经表示学习，避免昂贵的超参数搜索。 representation learning
36 Spatio-Temporal 3D Point Clouds from WiFi-CSI Data via Transformer Networks 提出基于Transformer网络的室内环境3D点云生成方法 predictive model multimodal
37 Mastering Chinese Chess AI (Xiangqi) Without Search 提出一种无需搜索的中国象棋AI，性能媲美顶尖人类棋手 reinforcement learning PPO

🔬 支柱一：机器人控制 (Robot Control) (3 篇)

#题目一句话要点标签🔗⭐
38 Diffusion Model Predictive Control 提出基于扩散模型的预测控制（D-MPC），用于提升离线强化学习的规划性能。 MPC model predictive control reinforcement learning
39 Diffusion Imitation from Observation 提出基于扩散模型的观测模仿学习框架DIFO，提升连续控制任务性能。 locomotion manipulation policy learning ✅
40 Neural Fourier Modelling: A Highly Compact Approach to Time-Series Analysis 提出神经傅里叶建模（NFM），用于紧凑高效的时间序列分析。 manipulation

🔬 支柱八：物理动画 (Physics-based Animation) (1 篇)

#题目一句话要点标签🔗⭐
41 Evaluating the Generalization Ability of Spatiotemporal Model in Urban Scenario 提出时空分布外基准测试，评估城市场景时空模型泛化能力 spatiotemporal

⬅️ 返回 cs.LG 首页 · 🏠 返回主页