Interpretability research in neuroscience and AI has independently arrived at a variety of methodological principles for investigating the mechanisms and representations underlying intelligent behavior. While these lines of inquiry have developed in parallel, the pursuit of similar goals has led to the emergence of conceptually analogous methods in both fields, with each emphasizing causality. In this workshop, we will examine the recently developed causal interventions and interpretability methods that facilitate the pursuit of mechanistic understanding of complex cognition.
Our goal is to build a shared understanding by translating and integrating the strongest ideas from each field, emphasizing their complementary perspectives.
National Institutes of Health
Brown University
| Start | End | Title | Speaker(s) |
|---|---|---|---|
| 9:30 AM | 9:45 AM | Introduction | Ciana Deveau & Jake Russin |
| 9:45 AM | 10:15 AM | Is Interpretability Neuroscience With Better Access? Stimulation methods, representations, and the future of mental health understanding | Mark Histed |
| 10:15 AM | 10:45 AM | Revealing Convergent Algorithms in Brains and Machines using Mechanistic Interpretability | Michael Lepori |
| 10:45 AM | 11:00 AM | Coffee break | |
| 11:00 AM | 11:30 AM | Understanding neural mechanisms by studying the dynamics of the representational geometry | Stefano Fusi |
| 11:30 AM | 12:00 PM | Shortcomings of Interpretability Tools for Understanding Long-Context Reasoning in Language Models | Jack Merullo |
| 12:00 PM | 3:15 PM | Lunch | |
| 3:15 PM | 3:45 PM | Surfing symmetries in overparameterized neural networks: Consequences for representational comparisons between biological and artificial neural networks and the brain | Andrew Saxe |
| 3:45 PM | 4:15 PM | Back to Bayes: Formalizing Mechanistic Interpretability in the Language of Cognitive Science | Ekdeep Lubana (virtual) |
| 4:15 PM | 4:30 PM | Coffee break | |
| 4:30 PM | 5:00 PM | The Compositional Geometry of Open-ended Reasoning | Ida Momennejad (virtual) |
| 5:00 PM | 5:30 PM | Bidirectional Lessons Between Interpretability and Neuroscience | Grace Lindsay (virtual) |
| 5:30 PM | 6:00 PM | Panel |