A wavelet-based encoding for neuroevolution

Sjoerd Van Steenkiste; Jan Koutník; Kurt Driessens; Jürgen Schmidhuber

doi:10.1145/2908812.2908905

A wavelet-based encoding for neuroevolution

Sjoerd Van Steenkiste, Jan Koutník, Kurt Driessens, Jürgen Schmidhuber

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

12 Scopus citations

Abstract

A new indirect scheme for encoding neural network connection weights as sets of wavelet-domain coefficients is proposed in this paper. It exploits spatial regularities in the weight-space to reduce the gene-space dimension by considering the low-frequency wavelet coefficients only. The wavelet-based encoding builds on top of a frequency-domain encoding, but unlike when using a Fourier-type transform, it offers gene locality while preserving continuity of the genotype-phenotype mapping. We argue that this added property allows for more efficient evolutionary search and demonstrate this on the octopus-arm control task, where superior solutions were found in fewer generations. The scalability of the wavelet-based encoding is shown by evolving networks with many parameters to control game-playing agents in the Arcade Learning Environment.

Original language	English (US)
Title of host publication	GECCO 2016 - Proceedings of the 2016 Genetic and Evolutionary Computation Conference
Publisher	Association for Computing Machinery, [email protected]
Pages	517-524
Number of pages	8
ISBN (Print)	9781450342063
DOIs	https://doi.org/10.1145/2908812.2908905
State	Published - Jul 20 2016
Externally published	Yes

Access to Document

10.1145/2908812.2908905

Cite this

@inproceedings{fd372ea121544fad995d1474f36b60ee,

title = "A wavelet-based encoding for neuroevolution",

abstract = "A new indirect scheme for encoding neural network connection weights as sets of wavelet-domain coefficients is proposed in this paper. It exploits spatial regularities in the weight-space to reduce the gene-space dimension by considering the low-frequency wavelet coefficients only. The wavelet-based encoding builds on top of a frequency-domain encoding, but unlike when using a Fourier-type transform, it offers gene locality while preserving continuity of the genotype-phenotype mapping. We argue that this added property allows for more efficient evolutionary search and demonstrate this on the octopus-arm control task, where superior solutions were found in fewer generations. The scalability of the wavelet-based encoding is shown by evolving networks with many parameters to control game-playing agents in the Arcade Learning Environment.",

author = "{Van Steenkiste}, Sjoerd and Jan Koutn{\'i}k and Kurt Driessens and J{\"u}rgen Schmidhuber",

note = "Generated from Scopus record by KAUST IRTS on 2022-09-14",

year = "2016",

month = jul,

day = "20",

doi = "10.1145/2908812.2908905",

language = "English (US)",

isbn = "9781450342063",

pages = "517--524",

booktitle = "GECCO 2016 - Proceedings of the 2016 Genetic and Evolutionary Computation Conference",

publisher = "Association for Computing Machinery, [email protected]",

}

TY - GEN

T1 - A wavelet-based encoding for neuroevolution

AU - Van Steenkiste, Sjoerd

AU - Koutník, Jan

AU - Driessens, Kurt

AU - Schmidhuber, Jürgen

N1 - Generated from Scopus record by KAUST IRTS on 2022-09-14

PY - 2016/7/20

Y1 - 2016/7/20

N2 - A new indirect scheme for encoding neural network connection weights as sets of wavelet-domain coefficients is proposed in this paper. It exploits spatial regularities in the weight-space to reduce the gene-space dimension by considering the low-frequency wavelet coefficients only. The wavelet-based encoding builds on top of a frequency-domain encoding, but unlike when using a Fourier-type transform, it offers gene locality while preserving continuity of the genotype-phenotype mapping. We argue that this added property allows for more efficient evolutionary search and demonstrate this on the octopus-arm control task, where superior solutions were found in fewer generations. The scalability of the wavelet-based encoding is shown by evolving networks with many parameters to control game-playing agents in the Arcade Learning Environment.

AB - A new indirect scheme for encoding neural network connection weights as sets of wavelet-domain coefficients is proposed in this paper. It exploits spatial regularities in the weight-space to reduce the gene-space dimension by considering the low-frequency wavelet coefficients only. The wavelet-based encoding builds on top of a frequency-domain encoding, but unlike when using a Fourier-type transform, it offers gene locality while preserving continuity of the genotype-phenotype mapping. We argue that this added property allows for more efficient evolutionary search and demonstrate this on the octopus-arm control task, where superior solutions were found in fewer generations. The scalability of the wavelet-based encoding is shown by evolving networks with many parameters to control game-playing agents in the Arcade Learning Environment.

UR - https://dl.acm.org/doi/10.1145/2908812.2908905

UR - http://www.scopus.com/inward/record.url?scp=84985991452&partnerID=8YFLogxK

U2 - 10.1145/2908812.2908905

DO - 10.1145/2908812.2908905

M3 - Conference contribution

SN - 9781450342063

SP - 517

EP - 524

BT - GECCO 2016 - Proceedings of the 2016 Genetic and Evolutionary Computation Conference

PB - Association for Computing Machinery, [email protected]

ER -

A wavelet-based encoding for neuroevolution

Abstract

Access to Document

Other files and links

Fingerprint

Cite this