Sep 17-29, 2019
Room 306, No. 5 Science Building
No registration fee. Apply online
Franz Clemens Heitzinger, Vienna University of Technology, Vienna, Austria
| Date | Time | Lecture |
|---|---|---|
| Sep 17, 2019 Tuesday | 18:00 - 20:00 | Lecture 1-2 |
| Sep 18, 2019 Wednesday | 18:00 - 21:00 | Lecture 3-5 |
| Sep 24, 2019 Tuesday | 18:00 - 21:00 | Lecture 6-8 |
| Sep 25, 2019 Wednesday | 18:00 - 20:00 | Lecture 9-10 |
Reinforcement learning is a field of machine learning. It is concerned with finding optimal strategies for agents that interact with stochastic environments. The agent can choose its actions, and the environment puts the agent into a new state and yields a reward. The objective of the agent is to maximize its future rewards.
Reinforcement learning can be viewed as a generalization of dynamic programming or optimal control. Due to its general setting, it has many applications in the sciences and in engineering, for example, autonomous driving, playing chess, playing Go, playing computer games, understanding how the brain works, and supply chains.
In this course, the basic concepts and the mathematical framework are explained first. The fundamental algorithms and their building blocks are discussed. Then convergence proofs for the algorithms are given, including multiple proofs based on different ideas in some cases. Relationships to partial differential equations (Hamilton-Jacobi-Bellman equations) are presented as well, and an overview of recent research is given. At the end of the course, the most important algorithms and their convergence theory will have been covered.
Lecture notes
Chapter_neural_networks