Domain adaptation for reinforcement learning on the Atari


Deep Reinforcement learning is a powerful machine learning paradigm that has had significant success across a wide range of control problems. This success often requires long training times to achieve. Observing that many problems share similarities, it is likely that much of the training done could be redundant if knowledge could be efficiently and appropriately shared across tasks. In this paper we demonstrate a novel adversarial domain adaptation approach to transfer state knowledge between domains and tasks on the Atari game suite. We show how this approach can successfully transfer across very different visual domains of the Atari platform. We focus on semantically related games that involve returning a ball with the user controlled agent. Our experiments demonstrate that our method reduces the number of samples required to successfully train an agent to play an Atari game.

Divisions: College of Engineering & Physical Sciences > School of Informatics and Digital Engineering > Computer Science
College of Engineering & Physical Sciences
College of Engineering & Physical Sciences > Systems analytics research institute (SARI)
Additional Information: © 2019 International Foundation for Autonomous Agents and Multiagent Systems (
Event Title: 18th International Conference on Autonomous Agents and Multiagent Systems, AAMAS 2019
Event Type: Other
Event Dates: 2019-05-13 - 2019-05-17
Uncontrolled Keywords: Deep learning,Domain adaptation,Reinforcement learning,Artificial Intelligence,Software,Control and Systems Engineering
ISBN: 978-1-4503-6309-9, 9781510892002
Full Text Link: https://arxiv.o ... /abs/1812.07452
Related URLs: http://www.scop ... tnerID=8YFLogxK (Scopus URL)
https://dl.acm. ... 3306127.3331943 (Publisher URL)
PURE Output Type: Conference contribution
Published Date: 2019-05-08
Accepted Date: 2019-05-01
Authors: Carr, Thomas
Chli, Maria (ORCID Profile 0000-0002-2840-4475)
Vogiatzis, George (ORCID Profile 0000-0002-3226-0603)



Version: Accepted Version

| Preview

Export / Share Citation


Additional statistics for this record