cs.AI（2025-04-21）

📊 共 30 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) (17) 支柱二：RL算法与架构 (RL & Architecture) (13 🔗4)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (17 篇)

#题目一句话要点标签🔗⭐
1 StableQuant: Layer Adaptive Post-Training Quantization for Speech Foundation Models 提出StableQuant，一种层自适应的语音基础模型后训练量化方法 large language model foundation model
2 Intrinsic Barriers to Explaining Deep Foundation Models 探讨深度基础模型的内在解释障碍 foundation model
3 AGI Is Coming... Right After AI Learns to Play Wordle 评估多模态Agent在Wordle游戏中的表现，揭示其在颜色识别方面的局限性 multimodal
4 Advancing AI-assisted Hardware Design with Hierarchical Decentralized Training and Personalized Inference-Time Optimization 提出分层分散训练和个性化推理优化框架，提升AI辅助硬件设计质量。 large language model
5 DONOD: Efficient and Generalizable Instruction Fine-Tuning for LLMs via Model-Intrinsic Dataset Pruning DONOD：通过模型内禀数据剪枝实现高效且泛化的LLM指令微调 large language model
6 Trends in Frontier AI Model Count: A Forecast to 2028 预测未来AI模型规模：基于训练算力阈值的模型数量预测至2028年 foundation model
7 Demand for LLMs: Descriptive Evidence on Substitution, Market Expansion, and Multihoming 基于LLM市场数据，揭示LLM需求的替代、扩展和多归属现象 large language model
8 LLM-Assisted Translation of Legacy FORTRAN Codes to C++: A Cross-Platform Study 利用大语言模型辅助将遗留FORTRAN代码翻译为C++，并进行跨平台研究 large language model
9 A Self-Improving Coding Agent 提出一种自改进编码Agent，通过自主编辑提升在代码任务上的性能 large language model
10 Synergistic Weak-Strong Collaboration by Aligning Preferences 提出协同弱-强模型框架，通过对齐偏好提升专业领域任务性能 large language model
11 Empowering AI to Generate Better AI Code: Guided Generation of Deep Learning Projects with LLMs DLCodeGen：提出规划引导的深度学习项目代码生成方法，提升LLM在复杂代码生成任务上的性能。 large language model
12 DualBreach: Efficient Dual-Jailbreaking via Target-Driven Initialization and Multi-Target Optimization 提出DualBreach框架，通过目标驱动初始化和多目标优化实现高效的大语言模型双重越狱攻击。 large language model
13 Evaluating Code Generation of LLMs in Advanced Computer Science Problems 评估LLM在高级计算机科学问题中的代码生成能力 large language model
14 AGI-Driven Generative Semantic Communications: Principles and Practices 提出生成式语义通信（GSC）框架，以支持AGI驱动的应用并应对其通信挑战。 foundation model
15 VLM as Policy: Common-Law Content Moderation Framework for Short Video Platform 提出KuaiMod框架，利用VLM和CoT解决短视频平台内容审核难题。 chain-of-thought
16 Retrieval is Not Enough: Enhancing RAG Reasoning through Test-Time Critique and Optimization 提出AlignRAG框架，通过测试时评判与优化增强RAG推理对齐性。 large language model
17 Automated Duplicate Bug Report Detection in Large Open Bug Repositories 提出基于机器学习的自动缺陷报告去重方法，提高大型开源项目问题追踪效率。 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🔗⭐
18 KGMEL: Knowledge Graph-Enhanced Multimodal Entity Linking KGMEL：提出知识图谱增强的多模态实体链接框架，提升实体对齐精度。 contrastive learning large language model multimodal ✅
19 Establishing Reliability Metrics for Reward Models in Large Language Models 提出RETA指标，用于量化评估大型语言模型奖励模型的可靠性。 reinforcement learning RLHF large language model
20 DRAGON: Distributional Rewards Optimize Diffusion Generative Models 提出DRAGON框架以优化生成模型的奖励函数 reinforcement learning RLHF DPO ✅
21 Integrating Symbolic Execution into the Fine-Tuning of Code-Generating LLMs 利用符号执行增强奖励模型，提升代码生成LLM微调效果 reinforcement learning direct preference optimization large language model
22 Text-to-Decision Agent: Offline Meta-Reinforcement Learning from Natural Language Supervision 提出T2DA，利用自然语言监督离线元强化学习，实现文本到决策的零样本泛化。 reinforcement learning world model ✅
23 Acting Less is Reasoning More! Teaching Model to Act Efficiently 提出OTC-PO，提升工具集成推理中LLM的效率，减少冗余工具调用。 reinforcement learning PPO large language model
24 Stop Summation: Min-Form Credit Assignment Is All Process Reward Model Needs for Reasoning 提出PURE，通过最小形式奖励分配解决过程奖励模型中的奖励利用问题 reinforcement learning large language model ✅
25 Mitigating Degree Bias in Graph Representation Learning with Learnable Structural Augmentation and Structural Self-Attention DegFairGT：通过可学习结构增强和结构自注意力缓解图表示学习中的度偏差 representation learning
26 A Self-supervised Learning Method for Raman Spectroscopy based on Masked Autoencoders 提出基于掩码自编码器的拉曼光谱自监督学习方法，提升光谱分析性能。 masked autoencoder
27 Learning Adaptive Parallel Reasoning with Language Models 提出自适应并行推理（APR）框架，提升语言模型在复杂推理任务中的性能和效率。 reinforcement learning chain-of-thought
28 Contemplative Artificial Intelligence 提出“沉思型人工智能”，通过内省原则提升AI安全性与合作性。 world model chain-of-thought
29 aiXamine: Simplified LLM Safety and Security aiXamine：简化LLM安全性和安全性的综合黑盒评估平台 distillation large language model
30 EducationQ: Evaluating LLMs' Teaching Capabilities Through Multi-Agent Dialogue Framework EducationQ：通过多智能体对话框架评估LLM的教学能力 teacher-student large language model

⬅️ 返回 cs.AI 首页 · 🏠 返回主页