数学推理论文趋势 2023–2026

主要结论

• 数学推理论文从 2023 年的 57 篇增长到 2026 年的 577 篇，三年间扩大到约 10 倍。
• NeurIPS 是发数学推理论文最多的会议，2023–2026 年共 469 篇，占四个会议总量的 43%。
• 交叉方向里涨得最快的是强化学习：同时属于两个主题的论文从 2023 年的 3 篇增至 2026 年的 140 篇。
• 在「神经符号」的 5 个主题里，它的 2026 年论文数排第 1；同组增长最快的可微推理是 2.5 倍，数学推理是 10 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 2026A mathematical theory of balancing relational generalization and memorization
• ICML 2026ASyMOB: Algebraic Symbolic Mathematical Operations Benchmark
• ICLR 2026AutoGPS: Automated Geometry Problem Solving via Multimodal Formalization and Deductive Reasoning
• CVPR 2026Affine Perspective-Three-Point Problem
• NeurIPS 2025Are Language Models Efficient Reasoners? A Perspective from Logic Programming
• ICML 2025Autoformulation of Mathematical Optimization Models Using LLMs
• ICLR 2025Advancing Mathematical Reasoning in Language Models: The Impact of Problem-Solving Data, Data Synthesis Methods, and Training Stages
• CVPR 2025Bridging Viewpoint Gaps: Geometric Reasoning Boosts Semantic Correspondence
• NeurIPS 2024A Careful Examination of Large Language Model Performance on Grade School Arithmetic
• ICML 2024Accelerating Legacy Numerical Solvers by Non-intrusive Gradient-based Meta-solving
• ICLR 2024Don't Trust: Verify -- Grounding LLM Quantitative Reasoning with Autoformalization
• CVPR 2024E-GPS: Explainable Geometry Problem Solving via Top-Down Solver and Bottom-Up Generator
• NeurIPS 2023Evaluating and Improving Tool-Augmented Computation-Intensive Math Reasoning
• ICML 2023Automated Search for Conjectures on Mathematical Constants using Analysis of Integer Sequences
• ICLR 2023Differentiable Mathematical Programming for Object-Centric Representation Learning
• CVPR 2023Abstract Visual Reasoning: An Algebraic Approach for Solving Raven's Progressive Matrices
• NeurIPS 2026A Mechanistic Analysis of Looped Reasoning Language Models
• ICML 2026Beyond Theorem Proving: Formulation, Framework and Benchmark for Formal Problem-Solving
• ICLR 2026CogFlow: Bridging Perception and Reasoning through Knowledge Internalization for Visual Mathematical Problem Solving
• CVPR 2026Hilbert-Geo Solving Solid Geometric Problems by Neural-Symbolic Reasoning

全部数学推理论文

共 469 篇，第 3 / 10 页

101. NeurIPS 2026MCPHallu: Benchmarking Reasoning, Execution, and Memory Hallucinations in MCP Agents
102. NeurIPS 2026Measuring Weak-to-Strong Legibility of Reasoning Models
103. NeurIPS 2026MedZERO: Self-Evolving Agents for Open-Ended Medical Reasoning Through Controlled Knowledge Accumulation
104. NeurIPS 2026MemPilot: Learning Transferable Latent Memory Mechanisms for LLM Reasoning
105. NeurIPS 2026Mental Imagery That Matters: Curing Latent Collapse in Multimodal Reasoning
106. NeurIPS 2026MindLoom: Composing Thought Modes for Frontier-Level Reasoning Data Synthesis
107. NeurIPS 2026MinPath: Learning Efficient LLM Reasoning via Minimal Dependency Paths
108. NeurIPS 2026MIRA: Reinforcing Multimodal Reasoning via Deceptive Contextual Augmentation
109. NeurIPS 2026MIRAGE: Mobile Agents with Implicit Reasoning and Generative World Models
110. NeurIPS 2026Mitigating Factual Hallucination in Large Reasoning Models via Mixed-Mode Advantage Regularization
111. NeurIPS 2026Mitigating Overthinking in Large Reasoning Language Models via Reasoning Path Deviation Monitoring
112. NeurIPS 2026MMFineReason: Closing the Multimodal Reasoning Gap via Open Data-Centric Methods
113. NeurIPS 2026MotionHalluc: Diagnosing Kinematic Hallucinations in Fine-Grained Motion Reasoning
114. NeurIPS 2026MUX: Continuous Reasoning via Multiplexed Tokens
115. NeurIPS 2026Natural Synthesis: Outperforming Reactive Synthesis Tools with Large Reasoning Models
116. NeurIPS 2026On the Token Value Inequality in Efficient Reasoning
117. NeurIPS 2026OS-Pruner: Pruning Chains-of-Thought of Reasoning Models via Optimal Stopping
118. NeurIPS 2026Outcome-Based RL Provably Leads Transformers to Reason, but Only With the Right Data
119. NeurIPS 2026PAC Reasoning: Controlling the Performance Loss for Efficient Reasoning
120. NeurIPS 2026Pause and Reflect: Conformal Aggregation for Chain-of-Thought Reasoning
121. NeurIPS 2026Perceive, Interact, Reason: Building Tool-Augmented Visual Agents for Spatial Reasoning
122. NeurIPS 2026PhysGraphNet: Physical-State Scene Graphs via Latent Graph Reasoning and Counterfactual Supervision
123. NeurIPS 2026PhysVista: Benchmarking Physical Intelligence in VLMs via a Perception-Reasoning-Assessment Loop
124. NeurIPS 2026Plausibility Is Not Prediction: Contrastive Evidence for LLM-Based Cellular Perturbation Reasoning
125. NeurIPS 2026Position: Robust Reasoning Requires Internal Time, Not Scale Alone
126. NeurIPS 2026Precise but Uncoupled: Reviewer Precision Does Not Guarantee Critique Uptake in Multi-Agent Math Reasoning
127. NeurIPS 2026Preserving Exploration for LLM Reasoning via Mean Order-Statistic Alignment
128. NeurIPS 2026Probabilistic Recursive Reasoning
129. NeurIPS 2026Probing the Trajectories of Reasoning Traces in Large Language Models
130. NeurIPS 2026Problem-Dependent Dynamic Regret over a Predictor Class with One-Gradient Feedback
131. NeurIPS 2026Programmatic Reasoning with Structural Schema: A Unified Framework for Multi-Table Inference
132. NeurIPS 2026Quantized Reasoning Models Think They Need to Think Longer, but They Do Not
133. NeurIPS 2026Reasoning as Compression: Unifying Budget Forcing via the Conditional Information Bottleneck
134. NeurIPS 2026Reasoning over Coupled Receptive Fields: Eliminating Subgraph Redundancy at Scale
135. NeurIPS 2026Reasoning Pathologies in Large Language Models: A Diagnostic Perspective
136. NeurIPS 2026Reasoning Warm-up: Scaling Label-free RL via Verifiable Surrogate Rewards
137. NeurIPS 2026Reasoning with Undecoded Tokens in Diffusion Language Models
138. NeurIPS 2026Reasoning-Aware IRT: Explainable Evaluation of Test-Time Scaling in Reasoning Models
139. NeurIPS 2026Reasoning-based Spatial Prior (RSP): Learning Spatial Priors from Multimodal LLMs for Object Detection
140. NeurIPS 2026Reasoning-Trace Collapse: Evaluating the Loss of Explicit Reasoning During Fine-Tuning
141. NeurIPS 2026ReasoningShield: Safety Moderation over Reasoning Traces of Large Reasoning Models
142. NeurIPS 2026Rebellious Student: Reversing Teacher Signals for Reasoning Exploration with Self-Distilled RLVR
143. NeurIPS 2026Reinforcing Multimodal Reasoning Against Visual Degradation
144. NeurIPS 2026Reject, Resample, Repeat: Understanding Parallel Reasoning in Language Model Inference
145. NeurIPS 2026Reliable Chain-of-Thought via Prefix Consistency
146. NeurIPS 2026RepoMirage: Probing Repository Context Reasoning in Code Agents with Perturbations
147. NeurIPS 2026ReSET: Accurate Latency-Critical NVFP4 Reasoning via Step-Aware Temperature Scaling
148. NeurIPS 2026RoboWits: Unexpected Challenges for Robotic Creative Problem Solving
149. NeurIPS 2026s2n-bignum-bench: A practical benchmark for evaluating low-level code reasoning of LLMs
150. NeurIPS 2026SAGE: Mitigating Long-Horizon Reasoning Biases via Topological Guidance

常见问题

2023–2026 年四大会议一共有多少篇数学推理论文？

共 1,085 篇，按年份分别是 2023 年 57 篇、2024 年 137 篇、2025 年 314 篇、2026 年 577 篇。

NeurIPS 2026 有多少篇数学推理论文？

NeurIPS 2026 有 208 篇数学推理论文，上一届是 190 篇。

ICLR 2026 有多少篇数学推理论文？

ICLR 2026 有 175 篇数学推理论文，上一届是 61 篇。

ICML 2026 有多少篇数学推理论文？

ICML 2026 有 113 篇数学推理论文，上一届是 45 篇。

CVPR 2026 有多少篇数学推理论文？

CVPR 2026 有 81 篇数学推理论文，上一届是 18 篇。

哪个会议的数学推理论文最多？

NeurIPS，2023–2026 年共 469 篇。