Foundations of Interpretability
📅 Monday, 31 August 2026 → Friday, 4 September 2026 in 45 days
📍
Institute for Pure and Applied Mathematics (IPAM), UCLA, Los Angeles, 
United States
Research workshop on mathematical and human-centred foundations for interpreting modern AI systems.
Foundations of Interpretability is a week-long workshop at the Institute for Pure and Applied Mathematics (IPAM) at UCLA in Los Angeles, running from 31 August to 4 September 2026. It sets out to establish rigorous mathematical and theoretical foundations for understanding why and how neural networks work, spanning mechanistic interpretability and theoretical analysis.
The workshop deliberately draws together researchers from mathematics, theoretical computer science, physics, machine learning and human–computer interaction, and includes a poster session for participants. The organising committee is Been Kim (Google DeepMind), Marina Meila (University of Waterloo), Alexander Oldenziel (Coefficient Giving) and Terence Tao (UCLA).
The event is held in person at IPAM, which runs an application and registration process for its workshops; prospective attendees should apply via the official page and confirm any funding or fee details there. Given the mathematically demanding theme, it is best suited to researchers and graduate students already engaged with interpretability or the theory of deep learning rather than newcomers to the area.
