强化学习论文趋势 2023–2026

主要结论

• 强化学习论文从 2023 年的 1,070 篇增长到 2026 年的 2,985 篇，三年间扩大到约 2.8 倍。
• NeurIPS 是发强化学习论文最多的会议，2023–2026 年共 2,494 篇，占四个会议总量的 36%。
• 交叉方向里涨得最快的是推理模型：同时属于两个主题的论文从 2023 年的 9 篇增至 2026 年的 437 篇。

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\texttt{FEROM}: Frontier Endogenous Reveal-Order Marginal Policy Optimization for Masked Diffusion LMs
• ICML 2026$\text{DT}^\text{2}$: Decision-Targeted Digital Twins
• ICLR 2026$\nabla$-Reasoner: LLM Reasoning via Test-Time Gradient Descent in Textual Space
• CVPR 2026A Causal Marriage between VLM and IRM from Understanding to
• NeurIPS 2025$\texttt{G1}$: Teaching LLMs to Reason on Graphs with Reinforcement Learning
• ICML 2025A Computationally Efficient Algorithm for Infinite-Horizon Average-Reward Linear MDPs
• ICLR 2025$\phi$-Update: A Class of Policy Update Methods with Policy Convergence Guarantee
• CVPR 2025All-Optical Nonlinear Diffractive Deep Network for Ultrafast Image Denoising
• NeurIPS 2024A Critical Evaluation of AI Feedback for Aligning Large Language Models
• ICML 2024${\rm E}(3)$-Equivariant Actor-Critic Methods for Cooperative Multi-Agent Reinforcement Learning
• ICLR 2024A Black-box Approach for Non-stationary Multi-agent Reinforcement Learning
• CVPR 2024DMR: Decomposed Multi-Modality Representations for Frames and Events Fusion in Visual Reinforcement Learning
• NeurIPS 2023“Why Not Looking backward?” A Robust Two-Step Method to Automatically Terminate Bayesian Optimization
• ICML 2023A Connection between One-Step RL and Critic Regularization in Reinforcement Learning
• ICLR 2023$O(T^{-1})$ Convergence of Optimistic-Follow-the-Regularized-Leader in Two-Player Zero-Sum Markov Games
• CVPR 2023"Seeing" Electric Network Frequency From Events
• NeurIPS 2026$\varphi$TD: Distributional Reinforcement Learning using Characteristic Functions
• ICML 2026$\textit{S}$-SPPO: Semantic-Calibrated Self-Play Preference Optimization
• ICLR 2026$AutoDrive\text{-}P^3$: Unified Chain of Perception–Prediction–Planning Thought via Reinforcement Fine-Tuning
• CVPR 2026AdaptVision Efficient Vision-Language Models via Adaptive Visual Acquisition

全部强化学习论文

共 760 篇，第 14 / 16 页

651. NeurIPS 2026Temporal Gradient Inversion for Private Trajectory Reconstruction in Embodied Reinforcement Learning
652. NeurIPS 2026Temporal Selective Exploration for Reinforcement Learning-Guided Continuous-Discrete Flow Matching in 3D Molecular Design
653. NeurIPS 2026Test-time Multi-agent Coordination by Decomposed Value Gradient Flow
654. NeurIPS 2026Test-time Risk Adaptation with Mixture of Agents
655. NeurIPS 2026TGPO: Temporal Grounded Policy Optimization for Signal Temporal Logic Tasks
656. NeurIPS 2026TGPO: Trace-Guided Policy Optimization for Robot Task Planning via Verifiable Subgoal Generation
657. NeurIPS 2026TGRL: Temperature-Grouped Reinforcement Learning for Efficient Exploration in LLMs
658. NeurIPS 2026The Benefits of Temporal Correlations: SGD Efficiently Learns k-Juntas from Random Walks
659. NeurIPS 2026The Capability Frontier of GRPO
660. NeurIPS 2026The Cost of Mismatch: Noise Amplification in Zeroth-Order Reinforcement Learning
661. NeurIPS 2026The Era of Agentic Organization: Learning to Organize with Language Models
662. NeurIPS 2026The Horizon Threshold in Cooperative Multi-Agent Reward-Free Exploration
663. NeurIPS 2026The Illusion of Diversity: Aligning LLM Exploration via Effective Entropy
664. NeurIPS 2026The Laplacian Keyboard: Beyond the Linear Span
665. NeurIPS 2026The Many Faces of On-Policy Distillation: Pitfalls, Mechanisms, and Fixes
666. NeurIPS 2026The Mirage of Optimizing Training Policies: Monotonic Inference Policies as the Real Objective for LLM Reinforcement Learning
667. NeurIPS 2026The Post-Training Dilemma: Why We Should Rethink the Sequential SFT-RL Paradigm
668. NeurIPS 2026The Reasoning Boundary Paradox: How Reinforcement Learning Constrains Language Models
669. NeurIPS 2026The Reciprocity Gradient
670. NeurIPS 2026The Reward Was in Your Data All Along: Correcting Flow Matching with Discriminator-Guided RL.
671. NeurIPS 2026The Sample Complexity of Multiple Change Point Identification under Bandit Feedback
672. NeurIPS 2026The Sampling Complexity of Condorcet Winner Identification in Dueling Bandits
673. NeurIPS 2026The Surprising Effectiveness of Deleting Weights in LLM Reasoning and Adaptation
674. NeurIPS 2026The Type Theory of Stationary MDPs: Rare Events and Uncertainty Quantification
675. NeurIPS 2026Theoretical Limits of Language Model Alignment
676. NeurIPS 2026There are Levels to It: Red Teaming LLMs with Hierarchical Reinforcement Learning
677. NeurIPS 2026Thinking with Images as Continuous Policy: Numerical Visual Chain-of-Thought
678. NeurIPS 2026ThinkSafe: Self-Generated Safety Alignment for Reasoning Models
679. NeurIPS 2026TIGER: Bridging the Multimodal Reasoning-Access Gap via Modality Counterfactuals
680. NeurIPS 2026Tool Verification for Test-Time Reinforcement Learning
681. NeurIPS 2026ToolSearcher: Optimizing Tool Selection at Scale via Reinforcement Learning
682. NeurIPS 2026Topic-Aware Contextual Cascading Bandits
683. NeurIPS 2026TOPPO: Rethinking PPO for Multi-Task Reinforcement Learning with Critic Balancing
684. NeurIPS 2026ToS: Tree-of-Skill Reinforcement Learning driven by Behavior Tree for Long-Horizon Manipulation
685. NeurIPS 2026Toward Cognitive Supersensing in Multimodal Large Language Models
686. NeurIPS 2026Toward Embodied World Agents via Embodied-Planning Dataset and Interactive World Models
687. NeurIPS 2026Toward Executable Multi-framework Front-end Code Generation with Self-Correction
688. NeurIPS 2026Toward Online Robust Zero-Sum Markov Games with Function Approximation
689. NeurIPS 2026Towards Convergence of PPO: An Approximate Descent Approach
690. NeurIPS 2026Towards Differentially Private Reinforcement Learning with General Function Approximation
691. NeurIPS 2026Towards On-Policy Data Evolution for Visual-Native Multimodal Deep Search Agents
692. NeurIPS 2026Towards Optimal Pre-training Data Mixtures for Empowering Reinforcement Learning in LLMs
693. NeurIPS 2026TPRL: Adaptive Visual Token Pruning in LVLMs via Language-Guided Reinforcement Learning
694. NeurIPS 2026Train at the Moving Edge: Rollout-Efficient RL for Large Reasoning Models
695. NeurIPS 2026Train the Agent, Not the Expert: Learning to Harness Heterogeneous Experts for Multi-Turn Visual Reasoning
696. NeurIPS 2026Training Agent to Scale Inference-Time Reasoning
697. NeurIPS 2026Training Generalizable Collaborative Agents via Strategic Risk Aversion
698. NeurIPS 2026Training on Documents About Monitoring Leads to CoT Obfuscation
699. NeurIPS 2026Training-Induced Escape from Token Clustering in a Mean-Field Formulation of Transformers
700. NeurIPS 2026Trajectory-Consistent Diffusion Policies for Offline Reinforcement Learning

常见问题

2023–2026 年四大会议一共有多少篇强化学习论文？

共 6,885 篇，按年份分别是 2023 年 1,070 篇、2024 年 1,190 篇、2025 年 1,640 篇、2026 年 2,985 篇。

NeurIPS 2026 有多少篇强化学习论文？

NeurIPS 2026 有 760 篇强化学习论文，上一届是 784 篇。

ICLR 2026 有多少篇强化学习论文？

ICLR 2026 有 911 篇强化学习论文，上一届是 379 篇。

ICML 2026 有多少篇强化学习论文？

ICML 2026 有 1,021 篇强化学习论文，上一届是 405 篇。

CVPR 2026 有多少篇强化学习论文？

CVPR 2026 有 293 篇强化学习论文，上一届是 72 篇。

哪个会议的强化学习论文最多？

NeurIPS，2023–2026 年共 2,494 篇。