Rigorous results from Machine Learning for Theoretical Physics

Rigorous results from Machine Learning for Theoretical Physics
Fabian Ruehle, Northeastern University
Fabian Ruehle
Date and time: Tue, Feb 20, 2024 - 2:30pm
Location: LGRT 1033
Category: ACFI Seminar
Abstract:

Machine learning techniques are increasingly powerful, but they are also often black-box and make stochastic rather than exact predictions, making them less attractive for theoretical physicists and mathematicians. In this talk, I summarize recent ideas and developments to overcome this problem.

I will start with a short introduction to neural networks, which form the backbone of modern machine learning applications, and subsequently introduce a field of machine learning called reinforcement learning. Using examples from physics and mathematics, I will illustrate how ML techniques can be used to generate (and hopefully subsequently prove) new conjectures, and to obtain rigorous, verifiably correct results.