Measuring and optimizing behavioral complexity for evolutionary reinforcement learning

Faustino J. Gomez; Julian Togelius; Juergen Schmidhuber

doi:10.1007/978-3-642-04277-5_77

Measuring and optimizing behavioral complexity for evolutionary reinforcement learning

Faustino J. Gomez, Julian Togelius, Juergen Schmidhuber

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

Model complexity is key concern to any artificial learning system due its critical impact on generalization. However, EC research has only focused phenotype structural complexity for static problems. For sequential decision tasks, phenotypes that are very similar in structure, can produce radically different behaviors, and the trade-off between fitness and complexity in this context is not clear. In this paper, behavioral complexity is measured explicitly using compression, and used as a separate objective to be optimized (not as an additional regularization term in a scalar fitness), in order to study this trade-off directly.

Original language	English (US)
Title of host publication	Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings
Pages	765-774
Number of pages	10
Edition	PART 2
DOIs	https://doi.org/10.1007/978-3-642-04277-5_77
State	Published - 2009
Event	19th International Conference on Artificial Neural Networks, ICANN 2009 - Limassol, Cyprus Duration: Sep 14 2009 → Sep 17 2009

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Number	PART 2
Volume	5769 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Other

Other	19th International Conference on Artificial Neural Networks, ICANN 2009
Country/Territory	Cyprus
City	Limassol
Period	9/14/09 → 9/17/09

ASJC Scopus subject areas

Theoretical Computer Science
General Computer Science

Access to Document

10.1007/978-3-642-04277-5_77

Cite this

Gomez, F. J., Togelius, J., & Schmidhuber, J. (2009). Measuring and optimizing behavioral complexity for evolutionary reinforcement learning. In Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings (PART 2 ed., pp. 765-774). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 5769 LNCS, No. PART 2). https://doi.org/10.1007/978-3-642-04277-5_77

Measuring and optimizing behavioral complexity for evolutionary reinforcement learning. / Gomez, Faustino J.; Togelius, Julian; Schmidhuber, Juergen.
Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings. PART 2. ed. 2009. p. 765-774 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 5769 LNCS, No. PART 2).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Gomez, FJ, Togelius, J & Schmidhuber, J 2009, Measuring and optimizing behavioral complexity for evolutionary reinforcement learning. in Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings. PART 2 edn, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), no. PART 2, vol. 5769 LNCS, pp. 765-774, 19th International Conference on Artificial Neural Networks, ICANN 2009, Limassol, Cyprus, 9/14/09. https://doi.org/10.1007/978-3-642-04277-5_77

Gomez FJ, Togelius J, Schmidhuber J. Measuring and optimizing behavioral complexity for evolutionary reinforcement learning. In Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings. PART 2 ed. 2009. p. 765-774. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 2). doi: 10.1007/978-3-642-04277-5_77

Gomez, Faustino J. ; Togelius, Julian ; Schmidhuber, Juergen. / Measuring and optimizing behavioral complexity for evolutionary reinforcement learning. Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings. PART 2. ed. 2009. pp. 765-774 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 2).

@inproceedings{a4b55d8ef89b403c84cba1426616ef31,

title = "Measuring and optimizing behavioral complexity for evolutionary reinforcement learning",

abstract = "Model complexity is key concern to any artificial learning system due its critical impact on generalization. However, EC research has only focused phenotype structural complexity for static problems. For sequential decision tasks, phenotypes that are very similar in structure, can produce radically different behaviors, and the trade-off between fitness and complexity in this context is not clear. In this paper, behavioral complexity is measured explicitly using compression, and used as a separate objective to be optimized (not as an additional regularization term in a scalar fitness), in order to study this trade-off directly.",

author = "Gomez, {Faustino J.} and Julian Togelius and Juergen Schmidhuber",

year = "2009",

doi = "10.1007/978-3-642-04277-5_77",

language = "English (US)",

isbn = "3642042767",

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

number = "PART 2",

pages = "765--774",

booktitle = "Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings",

edition = "PART 2",

note = "19th International Conference on Artificial Neural Networks, ICANN 2009 ; Conference date: 14-09-2009 Through 17-09-2009",

}

TY - GEN

T1 - Measuring and optimizing behavioral complexity for evolutionary reinforcement learning

AU - Gomez, Faustino J.

AU - Togelius, Julian

AU - Schmidhuber, Juergen

PY - 2009

Y1 - 2009

N2 - Model complexity is key concern to any artificial learning system due its critical impact on generalization. However, EC research has only focused phenotype structural complexity for static problems. For sequential decision tasks, phenotypes that are very similar in structure, can produce radically different behaviors, and the trade-off between fitness and complexity in this context is not clear. In this paper, behavioral complexity is measured explicitly using compression, and used as a separate objective to be optimized (not as an additional regularization term in a scalar fitness), in order to study this trade-off directly.

AB - Model complexity is key concern to any artificial learning system due its critical impact on generalization. However, EC research has only focused phenotype structural complexity for static problems. For sequential decision tasks, phenotypes that are very similar in structure, can produce radically different behaviors, and the trade-off between fitness and complexity in this context is not clear. In this paper, behavioral complexity is measured explicitly using compression, and used as a separate objective to be optimized (not as an additional regularization term in a scalar fitness), in order to study this trade-off directly.

UR - http://www.scopus.com/inward/record.url?scp=70450227694&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=70450227694&partnerID=8YFLogxK

U2 - 10.1007/978-3-642-04277-5_77

DO - 10.1007/978-3-642-04277-5_77

M3 - Conference contribution

AN - SCOPUS:70450227694

SN - 3642042767

SN - 9783642042768

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

SP - 765

EP - 774

BT - Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings

T2 - 19th International Conference on Artificial Neural Networks, ICANN 2009

Y2 - 14 September 2009 through 17 September 2009

ER -

Measuring and optimizing behavioral complexity for evolutionary reinforcement learning

Abstract

Publication series

Other

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this