cs.LG（2024-06-27）

📊 共 19 篇论文 | 🔗 4 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (10 🔗2) 支柱二：RL算法与架构 (RL & Architecture) (9 🔗2)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (10 篇)

#题目一句话要点标签🔗⭐
1 MissionGNN: Hierarchical Multimodal GNN-based Weakly Supervised Video Anomaly Recognition with Mission-Specific Knowledge Graph Generation MissionGNN：基于层级多模态GNN与任务知识图谱的弱监督视频异常识别 large language model multimodal ✅
2 LICO: Large Language Models for In-Context Molecular Optimization LICO：基于大语言模型的分子优化上下文学习框架 large language model
3 A look under the hood of the Interactive Deep Learning Enterprise (No-IDLE) No-IDLE原型系统：探索交互式深度学习在非专家用户中的应用 large language model multimodal
4 The Remarkable Robustness of LLMs: Stages of Inference? LLM推理阶段鲁棒性研究：通过层删除与交换揭示模型内部运作机制 large language model
5 Granite-Function Calling Model: Introducing Function Calling Abilities via Multi-task Learning of Granular Tasks 提出Granite-20B-FUNCTIONCALLING，通过多任务学习提升LLM函数调用能力。 large language model
6 Jump Starting Bandits with LLM-Generated Prior Knowledge 利用LLM生成先验知识，加速Contextual Bandit算法学习 large language model
7 From Artificial Needles to Real Haystacks: Improving Retrieval Capabilities in LLMs by Finetuning on Synthetic Data 通过合成数据微调提升LLM在长文本中的信息检索能力 large language model
8 Towards Learning Abductive Reasoning using VSA Distributed Representations 提出ARLC模型，利用VSA分布式表示学习归纳推理，解决抽象推理任务。 large language model ✅
9 A Teacher Is Worth A Million Instructions 利用大模型知识蒸馏与领域对齐，提升小模型指令调优性能 large language model
10 UnUnlearning: Unlearning is not sufficient for content regulation in advanced generative AI 揭示大语言模型内容监管困境：仅靠遗忘不足以抵御上下文学习带来的知识重引入 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (9 篇)

#题目一句话要点标签🔗⭐
11 OmniJARVIS: Unified Vision-Language-Action Tokenization Enables Open-World Instruction Following Agents OmniJARVIS：统一视觉-语言-动作 Token 化实现开放世界指令跟随智能体 imitation learning vision-language-action VLA
12 From Efficient Multimodal Models to World Models: A Survey 综述多模态大模型：迈向通用人工智能与世界模型的关键技术与挑战 world model large language model multimodal
13 Curriculum Learning with Quality-Driven Data Selection 提出基于质量驱动数据选择的课程学习方法，提升多模态大语言模型性能 curriculum learning large language model multimodal
14 Efficient World Models with Context-Aware Tokenization 提出Δ-IRIS，通过上下文感知 Tokenization 实现高效世界模型，刷新 Crafter 基准。 reinforcement learning deep reinforcement learning world model ✅
15 Averaging log-likelihoods in direct alignment 提出一种长度不变的直接对齐方法，优化LLM与人类判断的一致性。 reinforcement learning RLHF large language model
16 Instance Temperature Knowledge Distillation 提出基于强化学习的实例温度知识蒸馏方法，提升学生网络性能。 reinforcement learning distillation ✅
17 Leveraging Contrastive Learning for Enhanced Node Representations in Tokenized Graph Transformers GCFormer：利用对比学习增强Token化图Transformer中的节点表示，提升节点分类性能。 contrastive learning
18 Contrastive Policy Gradient: Aligning LLMs on sequence-level scores in a supervised-friendly fashion 提出对比策略梯度(CoPG)，用于在序列级奖励下对齐LLM，且兼容监督学习。 reinforcement learning large language model
19 Decoding-Time Language Model Alignment with Multiple Objectives 提出多目标解码（MOD）算法，用于解码时对齐语言模型以优化多个目标。 PPO DPO

⬅️ 返回 cs.LG 首页 · 🏠 返回主页