形式验证 ML论文趋势 2023–2026

主要结论

• 形式验证 ML论文从 2023 年的 10 篇增长到 2026 年的 37 篇，三年间扩大到约 3.7 倍。
• NeurIPS 是发形式验证 ML论文最多的会议，2023–2026 年共 32 篇，占四个会议总量的 38%。
• 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 0 篇增至 2026 年的 6 篇。
• 在「程序综合」的 5 个主题里，它的 2026 年论文数排第 4；同组增长最快的定理证明是 13 倍，形式验证 ML是 3.7 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 2026Certified but Private: Scalable Zero-Knowledge Proofs for the Formal Verification of Neural Networks
• ICML 2026Propose, Solve, Verify: Self-Play Through Formal Verification
• ICLR 2026Local Success Does Not Compose: Benchmarking Large Language Models for Compositional Formal Verification
• CVPR 2026Lipschitz Optimization for Formal Verification of Homographies
• NeurIPS 2025Formal Models of Active Learning from Contrastive Examples
• ICML 2025Position: Formal Mathematical Reasoning—A New Frontier in AI
• ICLR 2025Sensitivity Verification for Additive Decision Tree Ensembles
• CVPR 2025Neuro-Symbolic Evaluation of Text-to-Video Models using Formal Verification
• NeurIPS 2024Critically Assessing the State of the Art in Neural Network Verification
• ICML 2024Relational DNN Verification With Cross Executional Bound Refinement
• ICLR 2024CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing
• CVPR 2024Towards General Robustness Verification of MaxPool-based Convolutional Neural Networks via Tightening Linear Approximation
• NeurIPS 2023Provably Bounding Neural Network Preimages
• ICML 2023Trapdoor Normalization with Irreversible Ownership Verification
• ICLR 2023Fundamental Limits in Formal Verification of Message-Passing Neural Networks
• CVPR 2023Efficient Verification of Neural Networks Against LVM-Based Specifications
• NeurIPS 2026End-to-End Verification of Neuro-symbolic Automata via Contrastive Logit-Gaps
• ICML 2026Pushing the Boundaries of Natural Reasoning: Interleaved Bonus from Formal-Logic Verification
• ICLR 2026Out of the Shadows: Exploring a Latent Space for Neural Network Verification
• CVPR 2026Goldilocks Test Sets for Face Verification

全部形式验证 ML论文

共 84 篇，第 2 / 2 页

51. NeurIPS 2025Formal Models of Active Learning from Contrastive Examples
52. NeurIPS 2025Grammars of Formal Uncertainty: When to Trust LLMs in Automated Reasoning Tasks
53. NeurIPS 2025Learning Simple Interpolants for Linear Integer Arithmetic
54. NeurIPS 2025Let a Neural Network be Your Invariant
55. NeurIPS 2025SpecMAS: A Multi-Agent System for Self-Verifying System Generation via Formal Model Checking
56. NeurIPS 2025Traversal Verification for Speculative Tree Decoding
57. CVPR 2024Towards General Robustness Verification of MaxPool-based Convolutional Neural Networks via Tightening Linear Approximation
58. ICLR 2024CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing
59. ICLR 2024Incremental Randomized Smoothing Certification
60. ICLR 2024Interpreting Robustness Proofs of Deep Neural Networks
61. ICLR 2024Lemur: Integrating Large Language Models in Automated Program Verification
62. ICLR 2024Prediction without Preclusion: Recourse Verification with Reachable Sets
63. ICML 2024ED-Copilot: Reduce Emergency Department Wait Time with Language Model Diagnostic Assistance
64. ICML 2024Relational DNN Verification With Cross Executional Bound Refinement
65. ICML 2024Verification of Machine Unlearning is Fragile
66. ICML 2024VNN: Verification-Friendly Neural Networks with Hard Robustness Guarantees
67. NeurIPS 2024Critically Assessing the State of the Art in Neural Network Verification
68. NeurIPS 2024FVEL: Interactive Formal Verification Environment with Large Language Models via Theorem Proving
69. NeurIPS 2024Neural Model Checking
70. NeurIPS 2024Scalable Neural Network Verification with Branch-and-bound Inferred Cutting Planes
71. NeurIPS 2024The Expressive Capacity of State Space Models: A Formal Language Perspective
72. NeurIPS 2024The Power of Hard Attention Transformers on Data Sequences: A formal language theoretic perspective
73. NeurIPS 2024Verifiably Robust Conformal Prediction
74. NeurIPS 2024Verified Safe Reinforcement Learning for Neural Network Dynamic Models
75. CVPR 2023Efficient Verification of Neural Networks Against LVM-Based Specifications
76. ICLR 2023Formal Mathematics Statement Curriculum Learning
77. ICLR 2023Fundamental Limits in Formal Verification of Message-Passing Neural Networks
78. ICLR 2023Iterative Circuit Repair Against Formal Specifications
79. ICLR 2023Sound Randomized Smoothing in Floating-Point Arithmetic
80. ICLR 2023Wasserstein Auto-encoded MDPs: Formal Verification of Efficiently Distilled RL Policies with Many-sided Guarantees
81. ICML 2023Data-Copying in Generative Models: A Formal Framework
82. ICML 2023Trapdoor Normalization with Irreversible Ownership Verification
83. NeurIPS 2023Compositional Policy Learning in Stochastic Control Systems with Formal Guarantees
84. NeurIPS 2023Provably Bounding Neural Network Preimages

常见问题

2023–2026 年四大会议一共有多少篇形式验证 ML论文？

共 84 篇，按年份分别是 2023 年 10 篇、2024 年 18 篇、2025 年 19 篇、2026 年 37 篇。

NeurIPS 2026 有多少篇形式验证 ML论文？

NeurIPS 2026 有 12 篇形式验证 ML论文，上一届是 10 篇。

ICLR 2026 有多少篇形式验证 ML论文？

ICLR 2026 有 11 篇形式验证 ML论文，上一届是 5 篇。

ICML 2026 有多少篇形式验证 ML论文？

ICML 2026 有 12 篇形式验证 ML论文，上一届是 3 篇。

CVPR 2026 有多少篇形式验证 ML论文？

CVPR 2026 有 2 篇形式验证 ML论文，上一届是 1 篇。

哪个会议的形式验证 ML论文最多？

NeurIPS，2023–2026 年共 32 篇。