Optimistic agents are asymptotically optimal

Peter Sunehag; Marcus Hutter

doi:10.1007/978-3-642-35101-3_2

Optimistic agents are asymptotically optimal

Peter Sunehag^*, Marcus Hutter

^*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference Paper › peer-review

9 Citations (Scopus)

Abstract

We use optimism to introduce generic asymptotically optimal reinforcement learning agents. They achieve, with an arbitrary finite or compact class of environments, asymptotically optimal behavior. Furthermore, in the finite deterministic case we provide finite error bounds.

Original language	English
Title of host publication	AI 2012
Subtitle of host publication	Advances in Artificial Intelligence - 25th Australasian Joint Conference, Proceedings
Pages	15-26
Number of pages	12
DOIs	https://doi.org/10.1007/978-3-642-35101-3_2
Publication status	Published - 2012
Event	25th Australasian Joint Conference on Artificial Intelligence, AI 2012 - Sydney, NSW, Australia Duration: 4 Dec 2012 → 7 Dec 2012

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	7691 LNAI
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	25th Australasian Joint Conference on Artificial Intelligence, AI 2012
Country/Territory	Australia
City	Sydney, NSW
Period	4/12/12 → 7/12/12

Access to Document

10.1007/978-3-642-35101-3_2

Cite this

Sunehag, P., & Hutter, M. (2012). Optimistic agents are asymptotically optimal. In AI 2012: Advances in Artificial Intelligence - 25th Australasian Joint Conference, Proceedings (pp. 15-26). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 7691 LNAI). https://doi.org/10.1007/978-3-642-35101-3_2

@inproceedings{f061f24b68ba4e519f37886809f83f5f,

title = "Optimistic agents are asymptotically optimal",

abstract = "We use optimism to introduce generic asymptotically optimal reinforcement learning agents. They achieve, with an arbitrary finite or compact class of environments, asymptotically optimal behavior. Furthermore, in the finite deterministic case we provide finite error bounds.",

keywords = "Optimality, Optimism, Reinforcement Learning",

author = "Peter Sunehag and Marcus Hutter",

year = "2012",

doi = "10.1007/978-3-642-35101-3\_2",

language = "English",

isbn = "9783642351006",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

pages = "15--26",

booktitle = "AI 2012",

note = "25th Australasian Joint Conference on Artificial Intelligence, AI 2012 ; Conference date: 04-12-2012 Through 07-12-2012",

}

Sunehag, P & Hutter, M 2012, Optimistic agents are asymptotically optimal. in AI 2012: Advances in Artificial Intelligence - 25th Australasian Joint Conference, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 7691 LNAI, pp. 15-26, 25th Australasian Joint Conference on Artificial Intelligence, AI 2012, Sydney, NSW, Australia, 4/12/12. https://doi.org/10.1007/978-3-642-35101-3_2

Optimistic agents are asymptotically optimal. / Sunehag, Peter; Hutter, Marcus.
AI 2012: Advances in Artificial Intelligence - 25th Australasian Joint Conference, Proceedings. 2012. p. 15-26 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 7691 LNAI).

Research output: Chapter in Book/Report/Conference proceeding › Conference Paper › peer-review

TY - GEN

T1 - Optimistic agents are asymptotically optimal

AU - Sunehag, Peter

AU - Hutter, Marcus

PY - 2012

Y1 - 2012

N2 - We use optimism to introduce generic asymptotically optimal reinforcement learning agents. They achieve, with an arbitrary finite or compact class of environments, asymptotically optimal behavior. Furthermore, in the finite deterministic case we provide finite error bounds.

AB - We use optimism to introduce generic asymptotically optimal reinforcement learning agents. They achieve, with an arbitrary finite or compact class of environments, asymptotically optimal behavior. Furthermore, in the finite deterministic case we provide finite error bounds.

KW - Optimality

KW - Optimism

KW - Reinforcement Learning

UR - https://www.scopus.com/pages/publications/84871399605

U2 - 10.1007/978-3-642-35101-3_2

DO - 10.1007/978-3-642-35101-3_2

M3 - Conference Paper

SN - 9783642351006

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 15

EP - 26

BT - AI 2012

T2 - 25th Australasian Joint Conference on Artificial Intelligence, AI 2012

Y2 - 4 December 2012 through 7 December 2012

ER -

Optimistic agents are asymptotically optimal

Abstract

Publication series

Conference

Access to Document

Other files and links

Fingerprint

Cite this