机制解释论文趋势 2023–2026

主要结论

• 机制解释论文从 2023 年的 17 篇增长到 2026 年的 138 篇，三年间扩大到约 8.1 倍。
• ICML 是发机制解释论文最多的会议，2023–2026 年共 119 篇，占四个会议总量的 38%。
• 交叉方向里涨得最快的是深度学习 · large language models：同时属于两个主题的论文从 2023 年的 1 篇增至 2026 年的 23 篇。
• 在「可解释性」的 5 个主题里，它的 2026 年论文数排第 2；同组增长最快的特征归因是 1.9 倍，机制解释是 8.1 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 Mechanistic Interpretability Study of an Astronomical Foundation Model
• ICML 2026Contribution Weights: A Geometrical Analysis of Self-Attention Transformers
• ICLR 2026Circuit Insights: Towards Interpretability Beyond Activations
• CVPR 2026CIGMA Causal Information-Gain Mechanistic Attribution of Attention Heads in Vision
• NeurIPS 2025Beyond Components: Singular Vector-Based Interpretability of Transformer Circuits
• ICML 2025Algorithm Development in Neural Networks: Insights from the Streaming Parity Task
• ICLR 2025Arithmetic Without Algorithms: Language Models Solve Math with a Bag of Heuristics
• CVPR 2025Dissecting and Mitigating Diffusion Bias via Mechanistic Interpretability
• NeurIPS 2024InterpBench: Semi-Synthetic Transformers for Evaluating Mechanistic Interpretability Techniques
• ICML 2024Mechanistic Neural Networks for Scientific Machine Learning
• ICLR 2024Fine-Tuning Enhances Existing Mechanisms: A Case Study on Entity Tracking
• CVPR 2024SI-MIL: Taming Deep MIL for Self-Interpretability in Gigapixel Histopathology
• NeurIPS 2023A Logic for Expressing Log-Precision Transformers
• ICML 2023Mechanistic Mode Connectivity
• ICLR 2023Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 Small
• NeurIPS 2026Discovering Mechanistic Models of Neural Activity: System Identification in an in Silico Zebrafish
• ICML 2026CorrSteer: Generation-Time LLM Steering via Correlated Sparse Autoencoder Features
• ICLR 2026CLUE: Conflict-guided Localization for LLM Unlearning Framework
• CVPR 2026From Weights to Concepts Data-Free Interpretability of CLIP via Singular
• NeurIPS 2025Bigram Subnetworks: Mapping to Next Tokens in Transformer Language Models

全部机制解释论文

共 119 篇，第 2 / 3 页

51. ICML 2026Representational Curvature Modulates Behavioral Uncertainty in Large Language Models
52. ICML 2026Revisiting Anisotropy in Language Transformers: The Geometry of Learning Dynamics
53. ICML 2026RouterInterp: Understanding Superposed Specialisation in Mixture of Experts Routing
54. ICML 2026Safety Anchor: Defending Harmful Fine-tuning via Geometric Bottlenecks
55. ICML 2026Seeing to Generalize: How Visual Data Corrects Binding Shortcuts
56. ICML 2026Shared Lexical Task Representations Explain Behavioral Variability In LLMs
57. ICML 2026Singular Vectors of Attention Heads Align with Features
58. ICML 2026Spurious Rewards Paradox: Mechanistically Understanding How RLVR Activates Memorization Shortcuts in LLMs
59. ICML 2026SVD as a Fast Interpretability Method for Transformers
60. ICML 2026Symmetries in language statistics shape the geometry of model representations
61. ICML 2026Syntax vs. Semantics: How Transformers Learn Deep Dependencies
62. ICML 2026Temporal Context Reinstatement Drives Episodic-Like Order Memory in Long-Context Language Models
63. ICML 2026Textual Supervision Enhances Geospatial Representations in Vision-Language Models
64. ICML 2026The ACUTE Protocol: Operationalizing Language Model Activations for Better Calibration, Utility, and Trust
65. ICML 2026The Cylindrical Representation Hypothesis for Language Model Steering
66. ICML 2026The Geometry of Representational Failures in Vision Language Models
67. ICML 2026The Mechanistic Emergence of Symbol Grounding in Language Models
68. ICML 2026The Truth Lies Somewhere in the Middle (of the Generated Tokens)
69. ICML 2026Tracing the Persona Circuit: How Large Language Models Encode and Express Character Traits
70. ICML 2026Translation Heads: Disentangling meaning from language in LLM-based machine translation
71. ICML 2026Uncovering Grounding IDs: How External Cues Shape Multi-Modal Binding
72. ICML 2026Weights to Code: Extracting Interpretable Algorithms from the Discrete Transformer
73. ICML 2026WET: Mitigating World-Conditioned Knowledge Conflicts via World Entropy Tethering
74. ICML 2026What Makes Effective Supervision in Latent Chain-of-Thought: An Information-Theoretic Analysis
75. ICML 2025Algorithm Development in Neural Networks: Insights from the Streaming Parity Task
76. ICML 2025Are Sparse Autoencoders Useful? A Case Study in Sparse Probing
77. ICML 2025Beyond Induction Heads: In-Context Meta Learning Induces Multi-Phase Circuit Emergence
78. ICML 2025ConceptAttention: Diffusion Transformers Learn Highly Interpretable Features
79. ICML 2025Constrained Belief Updates Explain Geometric Structures in Transformer Representations
80. ICML 2025Emergence and Effectiveness of Task Vectors in In-Context Learning: An Encoder Decoder Perspective
81. ICML 2025Emergent Symbolic Mechanisms Support Abstract Reasoning in Large Language Models
82. ICML 2025Evaluating Neuron Explanations: A Unified Framework with Sanity Checks
83. ICML 2025Geometric Hyena Networks for Large-scale Equivariant Learning
84. ICML 2025How Do Transformers Learn Variable Binding in Symbolic Programs?
85. ICML 2025In-Context Linear Regression Demystified: Training Dynamics and Mechanistic Interpretability of Multi-Head Softmax Attention
86. ICML 2025Inference-Time Decomposition of Activations (ITDA): A Scalable Approach to Interpreting Large Language Models
87. ICML 2025Learning Multi-Level Features with Matryoshka Sparse Autoencoders
88. ICML 2025Locate-then-edit for Multi-hop Factual Recall under Knowledge Editing
89. ICML 2025Looking Beyond the Top-1: Transformers Determine Top Tokens in Order
90. ICML 2025Mechanistic Unlearning: Robust Knowledge Unlearning and Editing via Mechanistic Localization
91. ICML 2025MIB: A Mechanistic Interpretability Benchmark
92. ICML 2025Models of Heavy-Tailed Mechanistic Universality
93. ICML 2025Position: Explainable AI Cannot Advance Without Better User Studies
94. ICML 2025Position: Rethinking Explainable Machine Learning as Applied Statistics
95. ICML 2025Representation Shattering in Transformers: A Synthetic Study with Knowledge Editing
96. ICML 2025SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability
97. ICML 2025Scaling Sparse Feature Circuits For Studying In-Context Learning
98. ICML 2025Strategy Coopetition Explains the Emergence and Transience of In-Context Learning
99. ICML 2025Taming Knowledge Conflicts in Language Models
100. ICML 2025The Logical Implication Steering Method for Conditional Interventions on Transformer Generation

常见问题

2023–2026 年四大会议一共有多少篇机制解释论文？

共 311 篇，按年份分别是 2023 年 17 篇、2024 年 49 篇、2025 年 107 篇、2026 年 138 篇。

NeurIPS 2026 有多少篇机制解释论文？

NeurIPS 2026 有 16 篇机制解释论文，上一届是 34 篇。

ICLR 2026 有多少篇机制解释论文？

ICLR 2026 有 46 篇机制解释论文，上一届是 38 篇。

ICML 2026 有多少篇机制解释论文？

ICML 2026 有 74 篇机制解释论文，上一届是 33 篇。

CVPR 2026 有多少篇机制解释论文？

CVPR 2026 有 2 篇机制解释论文，上一届是 2 篇。

哪个会议的机制解释论文最多？

ICML，2023–2026 年共 119 篇。