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 311 篇，第 3 / 7 页

101. ICML 2026RouterInterp: Understanding Superposed Specialisation in Mixture of Experts Routing
102. ICML 2026Safety Anchor: Defending Harmful Fine-tuning via Geometric Bottlenecks
103. ICML 2026Seeing to Generalize: How Visual Data Corrects Binding Shortcuts
104. ICML 2026Shared Lexical Task Representations Explain Behavioral Variability In LLMs
105. ICML 2026Singular Vectors of Attention Heads Align with Features
106. ICML 2026Spurious Rewards Paradox: Mechanistically Understanding How RLVR Activates Memorization Shortcuts in LLMs
107. ICML 2026SVD as a Fast Interpretability Method for Transformers
108. ICML 2026Symmetries in language statistics shape the geometry of model representations
109. ICML 2026Syntax vs. Semantics: How Transformers Learn Deep Dependencies
110. ICML 2026Temporal Context Reinstatement Drives Episodic-Like Order Memory in Long-Context Language Models
111. ICML 2026Textual Supervision Enhances Geospatial Representations in Vision-Language Models
112. ICML 2026The ACUTE Protocol: Operationalizing Language Model Activations for Better Calibration, Utility, and Trust
113. ICML 2026The Cylindrical Representation Hypothesis for Language Model Steering
114. ICML 2026The Geometry of Representational Failures in Vision Language Models
115. ICML 2026The Mechanistic Emergence of Symbol Grounding in Language Models
116. ICML 2026The Truth Lies Somewhere in the Middle (of the Generated Tokens)
117. ICML 2026Tracing the Persona Circuit: How Large Language Models Encode and Express Character Traits
118. ICML 2026Translation Heads: Disentangling meaning from language in LLM-based machine translation
119. ICML 2026Uncovering Grounding IDs: How External Cues Shape Multi-Modal Binding
120. ICML 2026Weights to Code: Extracting Interpretable Algorithms from the Discrete Transformer
121. ICML 2026WET: Mitigating World-Conditioned Knowledge Conflicts via World Entropy Tethering
122. ICML 2026What Makes Effective Supervision in Latent Chain-of-Thought: An Information-Theoretic Analysis
123. NeurIPS 2026A Mechanistic Interpretability Study of an Astronomical Foundation Model
124. NeurIPS 2026A Unifying Perspective on Language Model Interpretability
125. NeurIPS 2026Assign and Add: A Mechanistic Study of Compositional Arithmetic
126. NeurIPS 2026Circuit-Level Knowledge Distillation for Large Language Models
127. NeurIPS 2026Discovering Mechanistic Models of Neural Activity: System Identification in an in Silico Zebrafish
128. NeurIPS 2026Geometry-Adaptive Explainer for Faithful Dictionary-Based Interpretability under Distribution Shift
129. NeurIPS 2026How Do Language Models Understand Tables? A Mechanistic Analysis of Cell Location
130. NeurIPS 2026How do Small Transformer Models Learn Hard Math Tasks?
131. NeurIPS 2026Instruction Anchor: Dissecting the Mechanistic Dynamics of Modality Arbitration
132. NeurIPS 2026OceanCBM: A Concept Bottleneck Model for Mechanistic Interpretability in Ocean Forecasting
133. NeurIPS 2026On the Recall Scaling Laws in Mamba: A Theoretical and Mechanistic Study via Hashing
134. NeurIPS 2026Patching Up Circuit Learning on Images
135. NeurIPS 2026Predictive Concept Decoders: Training Scalable End-to-End Interpretability Assistants
136. NeurIPS 2026Structural Support Certificates for Mechanistic Hypothesis Selection
137. NeurIPS 2026The parameters in weight-sparse transformers are interpretable
138. NeurIPS 2026The Unembedding Bottleneck: A Mechanistic Analysis of Single Digit Counting in LLMs
139. CVPR 2025Dissecting and Mitigating Diffusion Bias via Mechanistic Interpretability
140. CVPR 2025Towards Fine-Grained Interpretability: Counterfactual Explanations for Misclassification with Saliency Partition
141. ICLR 2025Adaptive Transformer Programs: Bridging the Gap Between Performance and Interpretability in Transformers
142. ICLR 2025Arithmetic Without Algorithms: Language Models Solve Math with a Bag of Heuristics
143. ICLR 2025Attention as a Hypernetwork
144. ICLR 2025Bilinear MLPs enable weight-based mechanistic interpretability
145. ICLR 2025Controllable Context Sensitivity and the Knob Behind It
146. ICLR 2025Density estimation with LLMs: a geometric investigation of in-context learning trajectories
147. ICLR 2025Differential learning kinetics govern the transition from memorization to generalization during in-context learning
148. ICLR 2025Differentiation and Specialization of Attention Heads via the Refined Local Learning Coefficient
149. ICLR 2025Do I Know This Entity? Knowledge Awareness and Hallucinations in Language Models
150. ICLR 2025Eliminating Position Bias of Language Models: A Mechanistic Approach

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。