Domain Adaptation for Reinforcement Learning on the Atari

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

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.
Original languageEnglish
Title of host publicationProceedings of the 2019 International Conference on Autonomous Agents and Multi-Agent Systems
PublisherACM
Pages1859-1861
ISBN (Electronic)978-1-4503-6309-9
ISBN (Print)978-1-4503-6309-9
Publication statusPublished - 8 May 2019
Event18th International Conference on Autonomous Agents and MultiAgent Systems - Montreal, Canada
Duration: 13 May 201917 May 2019

Conference

Conference18th International Conference on Autonomous Agents and MultiAgent Systems
CountryCanada
CityMontreal
Period13/05/1917/05/19

Fingerprint

Reinforcement learning
Learning systems
Experiments

Bibliographical note

© 2019 International Foundation for Autonomous Agents and
Multiagent Systems (www.ifaamas.org).

Cite this

Carr, T., Chli, M., & Vogiatzis, G. (2019). Domain Adaptation for Reinforcement Learning on the Atari. In Proceedings of the 2019 International Conference on Autonomous Agents and Multi-Agent Systems (pp. 1859-1861). ACM.
Carr, Thomas ; Chli, Maria ; Vogiatzis, George. / Domain Adaptation for Reinforcement Learning on the Atari. Proceedings of the 2019 International Conference on Autonomous Agents and Multi-Agent Systems. ACM, 2019. pp. 1859-1861
@inproceedings{02c3c8197a6a4ce49d1fe85bfc44ad3b,
title = "Domain Adaptation for Reinforcement Learning on the Atari",
abstract = "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.",
author = "Thomas Carr and Maria Chli and George Vogiatzis",
note = "{\circledC} 2019 International Foundation for Autonomous Agents and Multiagent Systems (www.ifaamas.org).",
year = "2019",
month = "5",
day = "8",
language = "English",
isbn = "978-1-4503-6309-9",
pages = "1859--1861",
booktitle = "Proceedings of the 2019 International Conference on Autonomous Agents and Multi-Agent Systems",
publisher = "ACM",
address = "United States",

}

Carr, T, Chli, M & Vogiatzis, G 2019, Domain Adaptation for Reinforcement Learning on the Atari. in Proceedings of the 2019 International Conference on Autonomous Agents and Multi-Agent Systems. ACM, pp. 1859-1861, 18th International Conference on Autonomous Agents and MultiAgent Systems, Montreal, Canada, 13/05/19.

Domain Adaptation for Reinforcement Learning on the Atari. / Carr, Thomas; Chli, Maria; Vogiatzis, George.

Proceedings of the 2019 International Conference on Autonomous Agents and Multi-Agent Systems. ACM, 2019. p. 1859-1861.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - Domain Adaptation for Reinforcement Learning on the Atari

AU - Carr, Thomas

AU - Chli, Maria

AU - Vogiatzis, George

N1 - © 2019 International Foundation for Autonomous Agents and Multiagent Systems (www.ifaamas.org).

PY - 2019/5/8

Y1 - 2019/5/8

N2 - 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.

AB - 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.

UR - https://dl.acm.org/citation.cfm?id=3331943

M3 - Conference contribution

SN - 978-1-4503-6309-9

SP - 1859

EP - 1861

BT - Proceedings of the 2019 International Conference on Autonomous Agents and Multi-Agent Systems

PB - ACM

ER -

Carr T, Chli M, Vogiatzis G. Domain Adaptation for Reinforcement Learning on the Atari. In Proceedings of the 2019 International Conference on Autonomous Agents and Multi-Agent Systems. ACM. 2019. p. 1859-1861