dc.contributor.author |
Hernandez-Leal, P
|
|
dc.contributor.author |
Rosman, Benjamin S
|
|
dc.contributor.author |
Taylor, ME
|
|
dc.contributor.author |
Sucar, LE
|
|
dc.contributor.author |
de Cote, EM
|
|
dc.date.accessioned |
2016-07-20T10:59:56Z |
|
dc.date.available |
2016-07-20T10:59:56Z |
|
dc.date.issued |
2016-02 |
|
dc.identifier.citation |
Hernandez-Leal, P. Rosman, B.S. Taylor, M.E. Sucar, L.E. de Cote, E.M. 2016. A bayesian approach for learning and tracking switching, non-stationary opponents. In: Autonomous Agents and Multiagent Systems, 9-13 May 2016, Singapore |
en_US |
dc.identifier.isbn |
978-1-4503-4239-1 |
|
dc.identifier.uri |
http://dl.acm.org/citation.cfm?id=2937137
|
|
dc.identifier.uri |
http://hdl.handle.net/10204/8650
|
|
dc.description |
Autonomous Agents and Multiagent Systems, 9-13 May 2016, Singapore. Due to copyright restrictions, the attached PDF file only contains the abstract of the full text item. For access to the full text item, please consult the publisher's website. |
en_US |
dc.description.abstract |
In many situations, agents are required to use a set of strategies (behaviors) and switch among them during the course of an interaction. This work focuses on the problem of recognizing the strategy used by an agent within a small number of interactions. We propose using a Bayesian framework to address this problem. Bayesian policy reuse (BPR) has been empirically shown to be efficient at correctly detecting the best policy to use from a library in sequential decision tasks. In this paper we extend BPR to adversarial settings, in particular, to opponents that switch from one stationary strategy to another. Our proposed extension enables learning new models in an online fashion when the learning agent detects that the current policies are not performing optimally. Experiments presented in repeated games show that our approach is capable of efficiently detecting opponent strategies and reacting quickly to behavior switches, thereby yielding better performance than state-of-the-art approaches in terms of average rewards. |
en_US |
dc.language.iso |
en |
en_US |
dc.publisher |
ACM |
en_US |
dc.relation.ispartofseries |
Workflow;16651 |
|
dc.subject |
Policy reuse |
en_US |
dc.subject |
Non-stationary opponents |
en_US |
dc.subject |
Repeated games |
en_US |
dc.title |
A bayesian approach for learning and tracking switching, non-stationary opponents |
en_US |
dc.type |
Conference Presentation |
en_US |
dc.identifier.apacitation |
Hernandez-Leal, P., Rosman, B. S., Taylor, M., Sucar, L., & de Cote, E. (2016). A bayesian approach for learning and tracking switching, non-stationary opponents. ACM. http://hdl.handle.net/10204/8650 |
en_ZA |
dc.identifier.chicagocitation |
Hernandez-Leal, P, Benjamin S Rosman, ME Taylor, LE Sucar, and EM de Cote. "A bayesian approach for learning and tracking switching, non-stationary opponents." (2016): http://hdl.handle.net/10204/8650 |
en_ZA |
dc.identifier.vancouvercitation |
Hernandez-Leal P, Rosman BS, Taylor M, Sucar L, de Cote E, A bayesian approach for learning and tracking switching, non-stationary opponents; ACM; 2016. http://hdl.handle.net/10204/8650 . |
en_ZA |
dc.identifier.ris |
TY - Conference Presentation
AU - Hernandez-Leal, P
AU - Rosman, Benjamin S
AU - Taylor, ME
AU - Sucar, LE
AU - de Cote, EM
AB - In many situations, agents are required to use a set of strategies (behaviors) and switch among them during the course of an interaction. This work focuses on the problem of recognizing the strategy used by an agent within a small number of interactions. We propose using a Bayesian framework to address this problem. Bayesian policy reuse (BPR) has been empirically shown to be efficient at correctly detecting the best policy to use from a library in sequential decision tasks. In this paper we extend BPR to adversarial settings, in particular, to opponents that switch from one stationary strategy to another. Our proposed extension enables learning new models in an online fashion when the learning agent detects that the current policies are not performing optimally. Experiments presented in repeated games show that our approach is capable of efficiently detecting opponent strategies and reacting quickly to behavior switches, thereby yielding better performance than state-of-the-art approaches in terms of average rewards.
DA - 2016-02
DB - ResearchSpace
DP - CSIR
KW - Policy reuse
KW - Non-stationary opponents
KW - Repeated games
LK - https://researchspace.csir.co.za
PY - 2016
SM - 978-1-4503-4239-1
T1 - A bayesian approach for learning and tracking switching, non-stationary opponents
TI - A bayesian approach for learning and tracking switching, non-stationary opponents
UR - http://hdl.handle.net/10204/8650
ER -
|
en_ZA |