Go to JKU Homepage
Institute for Machine Learning
What's that?

Institutes, schools, other departments, and programs create their own web content and menus.

To help you better navigate the site, see here where you are at the moment.

Special Topics Deep Reinforcement Learning (1VL)

Course no.: 365.101
Lecturer: Sepp Hochreiter, Jose Arjona, Michael Gillhofer
Times/locations: Mon 12:45-13:30, room K 224B
Mode: VL, 1h
Registration: KUSSS, opens an external URL in a new window

Motivation:

Reinforcement Learning is getting more and more relevance in the field of Machine Learning, playing a fundamental role in a wide range of areas such as autonomous driving, robotics or health-care. Classical reinforcement learning techniques combined with Deep Learning allow for complex systems able to perform highly sophisticated tasks, unthinkable only one decade ago.

This course will provide a broad view of the most important state-of-the-art methods and the core challenges that Reinforcement Learning is facing nowadays. The goal of this course is to assimilate the key ideas and understand upcoming contributions to the field.

The lectures are distributed in three main blocks: introduction, basic methods and advance methods. Additionally, a lecture about how to evaluate RL methods is introduced since it is a research topic itself. As a final lecture, an analysis of the latest major keystone in the field, AlphaStar is analysed and discussed.

Introduction

  • Introduction to RL
  • Deep Learning for Reinforcement Learning

Basic methods

  • Deep Q-Network based methods
  • Policy Gradient based methods
  • Reinforcement Learning and Control as Probabilistic Inference
  • Planing with Monte Carlo Tree Search

Advance methods

  • Exploration in Reinforcement Learning
  • Imitation and off-policy learning
  • Hierarchical Reinforcement Learning
  • Return decomposition for delayed rewards

Evaluation in Reinforcement Learning

A case study: Alphasta