Posters
Accepted poster presentations, including 23 spotlight presentations, showcasing recent research in mechanistic interpretability.
Machine Learning/CORE Unranked
Mech Interp Workshop ICML 2026
Mechanistic Interpretability Workshop at ICML 2026
The Mechanistic Interpretability Workshop at ICML 2026 brings together researchers to advance the understanding of neural network internals—weights and activations—to bridge the gap between model performance and interpretability. Building on prior workshops at ICML 2024 and NeurIPS 2025, this edition focuses on recent advances, shared understanding, and future directions in mechanistic interpretability.
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Research areas in scope
01
Mechanistic Interpretability
Analysis of weights and activationsUnderstanding model behavior and underlying computationsTools and techniques for interpretabilityDetecting adversarial or deceptive behaviorReliability and predictability of neural networks
Policies worth checking twice
- Call for papers is closed
- Accepted posters are now online
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026