Munich AI Lecture – Solvable High-Dimensional Models of Attention
The first Munich AI Lecture of 2026 is dedicated to the theoretical analysis of attention layers, which form the heart of modern machine learning architectures.
Prof. Lenka Zdeborová focuses on how such systems generalize from data and what principles determine their learning behavior on sequences of tokens. Using analyzable high dimensional models, learning and generalization performances are characterized in closed form. The talk focuses on supervised learning scenarios that enable precise theoretical predictions and provide mechanistic insights into representational learning of attention based architectures, and it concludes with an outlook towards theoretical models for self supervised and generative training with attention.
Event language
English
Target audience
Researchers, students, scientists from machine learning, AI, mathematics, physics, and theoretical computer science
About the speaker
Prof. Lenka Zdeborová is a professor of physics and computer science at the École Polytechnique Fédérale de Lausanne and leads the Statistical Physics of Computation Laboratory. Her research combines methods of statistical physics with questions from machine learning, inference, and optimization. She has received awards including the CNRS Bronze Medal, the Irène Joliot Curie Prize, and ERC Grants.