Dr. Fabian Ruehle Seminar on November 06, 2025

Title: Machine Learning for Theoretical and Mathematical Physics Abstract: I will discuss two ways to obtain rigorous results from machine learning, which is needed if we want to apply it to areas like theoretical physics or pure mathematics. I will first introduce neural networks and explain their shortcomings for rigorous applications. Based on this, I will discuss two possible avenues: The first is a new, more interpretable architecture called Kolmogorov-Arnold networks. The other is to formulate the mathematical problem as a (single-player) game and solve it with Reinforcement Learning, a branch of machine learning. I will illustrate both techniques with examples from pure mathematics and theoretical physics.