ITU

Learning Behavior Characterizations for Novelty Search

Research output: Conference Article in Proceeding or Book/Report chapterArticle in proceedingsResearchpeer-review

Standard

Learning Behavior Characterizations for Novelty Search. / Meyerson, Elliot; Lehman, Joel Anthony; Miikulainen, Risto.

Proceedings of the Genetic and Evolutionary Computation Conference 2016: GECCO '16. Association for Computing Machinery, 2016. p. 149-156.

Research output: Conference Article in Proceeding or Book/Report chapterArticle in proceedingsResearchpeer-review

Harvard

Meyerson, E, Lehman, JA & Miikulainen, R 2016, Learning Behavior Characterizations for Novelty Search. in Proceedings of the Genetic and Evolutionary Computation Conference 2016: GECCO '16. Association for Computing Machinery, pp. 149-156. https://doi.org/10.1145/2908812.2908929

APA

Meyerson, E., Lehman, J. A., & Miikulainen, R. (2016). Learning Behavior Characterizations for Novelty Search. In Proceedings of the Genetic and Evolutionary Computation Conference 2016: GECCO '16 (pp. 149-156). Association for Computing Machinery. https://doi.org/10.1145/2908812.2908929

Vancouver

Meyerson E, Lehman JA, Miikulainen R. Learning Behavior Characterizations for Novelty Search. In Proceedings of the Genetic and Evolutionary Computation Conference 2016: GECCO '16. Association for Computing Machinery. 2016. p. 149-156 https://doi.org/10.1145/2908812.2908929

Author

Meyerson, Elliot ; Lehman, Joel Anthony ; Miikulainen, Risto. / Learning Behavior Characterizations for Novelty Search. Proceedings of the Genetic and Evolutionary Computation Conference 2016: GECCO '16. Association for Computing Machinery, 2016. pp. 149-156

Bibtex

@inproceedings{bb00e79862314d2d80529d0dd6f5debb,
title = "Learning Behavior Characterizations for Novelty Search",
abstract = "Novelty search and related diversity-driven algorithms provide a promising approach to overcoming deception in complex domains. The behavior characterization (BC) is a critical choice in the application of such algorithms. The BC maps each evaluated individual to a behavior, i.e., some vector representation of what the individual is or does during evaluation. Search is then driven towards diversity in a metric space of these behaviors. BCs are built from hand-designed features that are limited by human expertise, or upon generic descriptors that cannot exploit domain nuance. The main contribution of this paper is an approach that addresses these shortcomings. Generic behaviors are recorded from evolution on several training tasks, and a new BC is learned from them that funnels evolution towards successful behaviors on any further tasks drawn from the domain. This approach is tested in increasingly complex simulated maze-solving domains, where it outperforms both hand-coded and generic BCs, in addition to outperforming objective-based search. The conclusion is that adaptive BCs can improve search in many-task domains with little human expertise.",
author = "Elliot Meyerson and Lehman, {Joel Anthony} and Risto Miikulainen",
year = "2016",
doi = "10.1145/2908812.2908929",
language = "English",
isbn = "978-1-4503-4206-3",
pages = "149--156",
booktitle = "Proceedings of the Genetic and Evolutionary Computation Conference 2016",
publisher = "Association for Computing Machinery",
address = "United States",

}

RIS

TY - GEN

T1 - Learning Behavior Characterizations for Novelty Search

AU - Meyerson, Elliot

AU - Lehman, Joel Anthony

AU - Miikulainen, Risto

PY - 2016

Y1 - 2016

N2 - Novelty search and related diversity-driven algorithms provide a promising approach to overcoming deception in complex domains. The behavior characterization (BC) is a critical choice in the application of such algorithms. The BC maps each evaluated individual to a behavior, i.e., some vector representation of what the individual is or does during evaluation. Search is then driven towards diversity in a metric space of these behaviors. BCs are built from hand-designed features that are limited by human expertise, or upon generic descriptors that cannot exploit domain nuance. The main contribution of this paper is an approach that addresses these shortcomings. Generic behaviors are recorded from evolution on several training tasks, and a new BC is learned from them that funnels evolution towards successful behaviors on any further tasks drawn from the domain. This approach is tested in increasingly complex simulated maze-solving domains, where it outperforms both hand-coded and generic BCs, in addition to outperforming objective-based search. The conclusion is that adaptive BCs can improve search in many-task domains with little human expertise.

AB - Novelty search and related diversity-driven algorithms provide a promising approach to overcoming deception in complex domains. The behavior characterization (BC) is a critical choice in the application of such algorithms. The BC maps each evaluated individual to a behavior, i.e., some vector representation of what the individual is or does during evaluation. Search is then driven towards diversity in a metric space of these behaviors. BCs are built from hand-designed features that are limited by human expertise, or upon generic descriptors that cannot exploit domain nuance. The main contribution of this paper is an approach that addresses these shortcomings. Generic behaviors are recorded from evolution on several training tasks, and a new BC is learned from them that funnels evolution towards successful behaviors on any further tasks drawn from the domain. This approach is tested in increasingly complex simulated maze-solving domains, where it outperforms both hand-coded and generic BCs, in addition to outperforming objective-based search. The conclusion is that adaptive BCs can improve search in many-task domains with little human expertise.

U2 - 10.1145/2908812.2908929

DO - 10.1145/2908812.2908929

M3 - Article in proceedings

SN - 978-1-4503-4206-3

SP - 149

EP - 156

BT - Proceedings of the Genetic and Evolutionary Computation Conference 2016

PB - Association for Computing Machinery

ER -

ID: 81058550