PhD Courses in Denmark

Welcome to Reinforcement Learning

Description:

An intelligent system is expected to generate policies autonomously to achieve a goal, which is mostly to maximize a given reward function. Reinforcement learning is a set of methods in machine learning that can produce such policies. To learn optimal actions in an environment that is not fully comprehensible to itself, an intelligent system can use reinforcement algorithms to leverage its experience to figure out optimal policies. Nowadays, reinforcement learning techniques are successfully applied in various engineering fields, including robotics (DeepMind’s walking robot) and computer-games (AlphaGo and TD-Gammon).

Developed independently from reinforcement learning, dynamic programming and related stochastic optimisation is a set of algorithms that generate policies assuming that the environment is fully comprehensible to the intelligent system. Therefore, dynamic programming provides an essential base to learn reinforcement learning. The course aims at building a fundamental understanding of both methods based on their relations to each other and on their applications to similar problems.

The course consists of the following topics:

Markov decision processes, dynamic programming for infinite time and stopping time, reinforcement learning, safe learning and verification tools for reinforcement learning.

Prerequisites: Basic knowledge of mathematics: calculus and probability

Learning objectives:

- General Introduction to machine learning herein Reinforcement Learning

- Markov Decision Processes and Dynamic Programming

- Reinforcement Learning with Temporal-Difference Learning

- Policy prediction with approximation

- Verification tool UPPAAL for model-based reinforcement learning

Key literature: Richard S. Sutton, Andrew G. Barto, Reinforcement Learning

For additional information, updates, and registration, please refer to AAU PhDMoodle via the link provided on the right side of this page.