Sebastian Rosengren: Neural networks - Reinforcement learning

Abstract

Reinforcement learning is, as many things are in machine learning, both the problem and the solution.  

The problem: How should an agent act in an uncertain environment? The Solution: Optimal. 

In this lecture we will introduce the underlying theory of reinforcement learning: Markov Decision Processes, Bellman equations, and Value Iteration. 

We will also provide an introduction to more modern approaches such as Q-learning, and Deep Q-learning. In the end we will play some tic-tac-toe against our own creation.

How to register

We wish to confine the number of participants to members of academic and professional communities, so that we introduced a registration procedure. In order to get registered, please send an email to otryakhin@math.su.se .