Autumn Semester 2020 takes place in a mixed form of online and classroom teaching.
Please read the published information on the individual courses carefully.

227-0559-00L  Seminar in Deep Reinforcement Learning

SemesterSpring Semester 2019
LecturersR. Wattenhofer, O. Richter
Periodicityyearly recurring course
Language of instructionEnglish
CommentNumber of participants limited to 24.


AbstractIn this seminar participating students present and discuss recent research papers in the area of deep reinforcement learning. The seminar starts with two introductory lessons introducing the basic concepts. Alongside the seminar a programming challenge is posed in which students can take part to improve their grade.
ObjectiveSince Google Deepmind presented the Deep Q-Network (DQN) algorithm in 2015 that could play Atari-2600 games at a superhuman level, the field of deep reinforcement learning gained a lot of traction. It sparked media attention with AlphaGo and AlphaZero and is one of the most prominent research areas. Yet many research papers in the area come from one of two sources: Google Deepmind or OpenAI. In this seminar we aim at giving the students an in depth view on the current advances in the area by discussing recent papers as well as discussing current issues and difficulties surrounding deep reinforcement learning.
ContentTwo introductory courses introducing Q-learning and policy gradient methods. Afterwards participating students present recent papers. For details see: www.disco.ethz.ch/courses.html
Lecture notesSlides of presentations will be made available.
LiteratureOpenAI course (https://spinningup.openai.com/en/latest/) plus selected papers.
The paper selection can be found on www.disco.ethz.ch/courses.html.
Prerequisites / NoticeIt is expected that student have prior knowledge and interest in machine and deep learning, for instance by having attended appropriate courses.