AI 智能体论文趋势 2023–2026

主要结论

• AI 智能体论文从 2023 年的 170 篇增长到 2026 年的 1,983 篇，三年间扩大到约 12 倍。
• NeurIPS 是发AI 智能体论文最多的会议，2023–2026 年共 1,317 篇，占四个会议总量的 40%。
• 交叉方向里涨得最快的是大模型基准：同时属于两个主题的论文从 2023 年的 1 篇增至 2026 年的 116 篇。
• 在「前沿系统」的 4 个主题里，它的 2026 年论文数排第 2；同组增长最快的世界模型是 11 倍，AI 智能体是 12 倍。

代表论文

按年份和会议均匀挑选，方便快速看到每一届的样貌

• NeurIPS 2026$\boldsymbol{f}$-OPD: Stabilizing Long-Horizon On-Policy Distillation with Freshness-Aware Control
• ICML 2026$\tau$-Knowledge: Evaluating Conversational Agents over Unstructured Knowledge
• ICLR 2026A Benchmark for Deep Information Synthesis
• CVPR 20263D-VCD Hallucination Mitigation in 3D-LLM Embodied Agents through Visual Contrastive
• NeurIPS 20254KAgent: Agentic Any Image to 4K Super-Resolution
• ICML 2025A Unified Approach to Routing and Cascading for LLMs
• ICLR 2025A Generalist Hanabi Agent
• CVPR 2025Collaborative Tree Search for Enhancing Embodied Multi-Agent Collaboration
• NeurIPS 2024AdaSociety: An Adaptive Environment with Social Structures for Multi-Agent Decision-Making
• ICML 2024A Human-Inspired Reading Agent with Gist Memory of Very Long Contexts
• ICLR 2024A Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis
• CVPR 2024Density-Adaptive Model Based on Motif Matrix for Multi-Agent Trajectory Prediction
• NeurIPS 2023AdaPlanner: Adaptive Planning from Feedback with Language Models
• ICML 2023A Game-Theoretic Framework for Managing Risk in Multi-Agent Systems
• ICLR 2023Achieving Near-Optimal Individual Regret & Low Communications in Multi-Agent Bandits
• CVPR 2023Behavioral Analysis of Vision-and-Language Navigation Agents
• NeurIPS 20262-Step Agent: How a Bayesian Decision Maker Learns from AI-Decision Support
• ICML 2026$\tau$-Voice: Benchmarking Full-Duplex Voice Agents on Real-World Domains
• ICLR 2026A Framework for Studying AI Agent Behavior: Evidence from Consumer Choice Experiments
• CVPR 2026A Multi-Agent Perception-Action Alliance for Efficient Long Video Reasoning

全部AI 智能体论文

共 178 篇，第 3 / 4 页

101. ICML 2025Multi-agent Architecture Search via Agentic Supernet
102. ICML 2025MVA: Linear Attention with High-order Query-Keys Integration and Multi-level Vocabulary Decomposition
103. ICML 2025Near Optimal Best Arm Identification for Clustered Bandits
104. ICML 2025Nemotron-CORTEXA: Enhancing LLM Agents for Software Engineering Tasks via Improved Localization and Solution Diversity
105. ICML 2025O-MAPL: Offline Multi-agent Preference Learning
106. ICML 2025Of Mice and Machines: A Comparison of Learning Between Real World Mice and RL Agents
107. ICML 2025Offline Opponent Modeling with Truncated Q-driven Instant Policy Refinement
108. ICML 2025On the Resilience of LLM-Based Multi-Agent Collaboration with Faulty Agents
109. ICML 2025OrcaLoca: An LLM Agent Framework for Software Issue Localization
110. ICML 2025PAC Learning with Improvements
111. ICML 2025PaperBench: Evaluating AI’s Ability to Replicate AI Research
112. ICML 2025PatchPilot: A Cost-Efficient Software Engineering Agent with Early Attempts on Formal Verification
113. ICML 2025PhantomWiki: On-Demand Datasets for Reasoning and Retrieval Evaluation
114. ICML 2025PINNsAgent: Automated PDE Surrogation with Large Language Models
115. ICML 2025Plan-and-Act: Improving Planning of Agents for Long-Horizon Tasks
116. ICML 2025PokéChamp: an Expert-level Minimax Language Agent
117. ICML 2025Policy-Regret Minimization in Markov Games with Function Approximation
118. ICML 2025Position: AI Agents Need Authenticated Delegation
119. ICML 2025Position: Build Agent Advocates, Not Platform Agents
120. ICML 2025Position: LLM Social Simulations Are a Promising Research Method
121. ICML 2025Position: Scaling LLM Agents Requires Asymptotic Analysis with LLM Primitives
122. ICML 2025Position: The Most Expensive Part of an LLM *should* be its Training Data
123. ICML 2025Position: Theory of Mind Benchmarks are Broken for Large Language Models
124. ICML 2025Position: Truly Self-Improving Agents Require Intrinsic Metacognitive Learning
125. ICML 2025Position: Uncertainty Quantification Needs Reassessment for Large Language Model Agents
126. ICML 2025Position: We Need An Algorithmic Understanding of Generative AI
127. ICML 2025Prediction-Aware Learning in Multi-Agent Systems
128. ICML 2025Principal-Agent Bandit Games with Self-Interested and Exploratory Learning Agents
129. ICML 2025Proactive Agents for Multi-Turn Text-to-Image Generation Under Uncertainty
130. ICML 2025Proposer-Agent-Evaluator (PAE): Autonomous Skill Discovery For Foundation Model Internet Agents
131. ICML 2025Provably Efficient Algorithm for Best Scoring Rule Identification in Online Principal-Agent Information Acquisition
132. ICML 2025Prune 'n Predict: Optimizing LLM Decision-making with Conformal Prediction
133. ICML 2025QLASS: Boosting Language Agent Inference via Q-Guided Stepwise Search
134. ICML 2025R3DM: Enabling Role Discovery and Diversity Through Dynamics Models in Multi-agent Reinforcement Learning
135. ICML 2025RE-Bench: Evaluating Frontier AI R&D Capabilities of Language Model Agents against Human Experts
136. ICML 2025Reflection-Bench: Evaluating Epistemic Agency in Large Language Models
137. ICML 2025Reidentify: Context-Aware Identity Generation for Contextual Multi-Agent Reinforcement Learning
138. ICML 2025Reinforce LLM Reasoning through Multi-Agent Reflection
139. ICML 2025RepoAudit: An Autonomous LLM-Agent for Repository-Level Code Auditing
140. ICML 2025ResearchTown: Simulator of Human Research Community
141. ICML 2025Revisiting Cooperative Off-Policy Multi-Agent Reinforcement Learning
142. ICML 2025Risk-Sensitive Theory of Mind: Coordinating with Agents of Unknown Bias using Cumulative Prospect Theory
143. ICML 2025RLEF: Grounding Code LLMs in Execution Feedback with Reinforcement Learning
144. ICML 2025Robust Multi-Agent Reinforcement Learning with Stochastic Adversary
145. ICML 2025Robust Multi-bit Text Watermark with LLM-based Paraphrasers
146. ICML 2025Robust Noise Attenuation via Adaptive Pooling of Transformer Outputs
147. ICML 2025Sable: a Performant, Efficient and Scalable Sequence Model for MARL
148. ICML 2025SafeArena: Evaluating the Safety of Autonomous Web Agents
149. ICML 2025Self-Play $Q$-Learners Can Provably Collude in the Iterated Prisoner's Dilemma
150. ICML 2025ShieldAgent: Shielding Agents via Verifiable Safety Policy Reasoning

常见问题

2023–2026 年四大会议一共有多少篇AI 智能体论文？

共 3,266 篇，按年份分别是 2023 年 170 篇、2024 年 329 篇、2025 年 784 篇、2026 年 1,983 篇。

NeurIPS 2026 有多少篇AI 智能体论文？

NeurIPS 2026 有 729 篇AI 智能体论文，上一届是 362 篇。

ICLR 2026 有多少篇AI 智能体论文？

ICLR 2026 有 500 篇AI 智能体论文，上一届是 204 篇。

ICML 2026 有多少篇AI 智能体论文？

ICML 2026 有 640 篇AI 智能体论文，上一届是 178 篇。

CVPR 2026 有多少篇AI 智能体论文？

CVPR 2026 有 114 篇AI 智能体论文，上一届是 40 篇。

哪个会议的AI 智能体论文最多？

NeurIPS，2023–2026 年共 1,317 篇。