“Solving Stochastic Control and Mean Field Games using Deep Reinforcement Learning”
Venerdì 27 Gennaio 2023, ore 14:30 - Aula 2BC30 e Zoom - Athanasios Vasileiadis (Université Côte d'Azur)
Abstract
In recent years there have been big breakthroughs in applying machine learning for the solution of control problems primarily solving the curse of dimensionality. In this talk we will lay the foundations of Reinforcement Learning and apply it to stochastic control problems first and then Mean Field Games. On the theoretical side we will see some guarantees for the convergence of the approximations, as well as how to use noise to learn the Nash Equilibria of the MFG problem. On the practical side we will use Artificial Neural Networks to look for the optimal controls in feedback form and discuss results in higher dimensions.