Generalized compressed network search

Rupesh Kumar Srivastava; Jürgen Schmidhuber; Faustino Gomez

doi:10.1007/978-3-642-32937-1_34

Generalized compressed network search

Rupesh Kumar Srivastava, Jürgen Schmidhuber, Faustino Gomez

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

1 Scopus citations

Abstract

This paper presents initial results of Generalized Compressed Network Search (GCNS), a method for automatically identifying the important frequencies for neural networks encoded as Fourier-type coefficients (i.e. "compressed" networks [7]). GCNS is a general search procedure in this coefficient space - both the number of frequencies and their value are automatically determined by employing the use of variable-length chromosomes, inspired by messy genetic algorithms. The method achieves better compression than our previous approach, and promises improved generalization for evolved controllers. Results for a high-dimensional Octopus arm control problem show that a high fitness 3680-weight network can be encoded using less than 10 coefficients using the frequencies identified by GCNS. © 2012 Springer-Verlag.

Original language	English (US)
Title of host publication	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Pages	337-346
Number of pages	10
DOIs	https://doi.org/10.1007/978-3-642-32937-1_34
State	Published - Sep 24 2012
Externally published	Yes

ASJC Scopus subject areas

Theoretical Computer Science
General Computer Science

Access to Document

10.1007/978-3-642-32937-1_34

Cite this

@inproceedings{27fdf82a88fa469e9377363d0d35c654,

title = "Generalized compressed network search",

abstract = "This paper presents initial results of Generalized Compressed Network Search (GCNS), a method for automatically identifying the important frequencies for neural networks encoded as Fourier-type coefficients (i.e. {"}compressed{"} networks [7]). GCNS is a general search procedure in this coefficient space - both the number of frequencies and their value are automatically determined by employing the use of variable-length chromosomes, inspired by messy genetic algorithms. The method achieves better compression than our previous approach, and promises improved generalization for evolved controllers. Results for a high-dimensional Octopus arm control problem show that a high fitness 3680-weight network can be encoded using less than 10 coefficients using the frequencies identified by GCNS. {\textcopyright} 2012 Springer-Verlag.",

author = "Srivastava, {Rupesh Kumar} and J{\"u}rgen Schmidhuber and Faustino Gomez",

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

year = "2012",

month = sep,

day = "24",

doi = "10.1007/978-3-642-32937-1_34",

language = "English (US)",

isbn = "9783642329364",

pages = "337--346",

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

}

TY - GEN

T1 - Generalized compressed network search

AU - Srivastava, Rupesh Kumar

AU - Schmidhuber, Jürgen

AU - Gomez, Faustino

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

PY - 2012/9/24

Y1 - 2012/9/24

N2 - This paper presents initial results of Generalized Compressed Network Search (GCNS), a method for automatically identifying the important frequencies for neural networks encoded as Fourier-type coefficients (i.e. "compressed" networks [7]). GCNS is a general search procedure in this coefficient space - both the number of frequencies and their value are automatically determined by employing the use of variable-length chromosomes, inspired by messy genetic algorithms. The method achieves better compression than our previous approach, and promises improved generalization for evolved controllers. Results for a high-dimensional Octopus arm control problem show that a high fitness 3680-weight network can be encoded using less than 10 coefficients using the frequencies identified by GCNS. © 2012 Springer-Verlag.

AB - This paper presents initial results of Generalized Compressed Network Search (GCNS), a method for automatically identifying the important frequencies for neural networks encoded as Fourier-type coefficients (i.e. "compressed" networks [7]). GCNS is a general search procedure in this coefficient space - both the number of frequencies and their value are automatically determined by employing the use of variable-length chromosomes, inspired by messy genetic algorithms. The method achieves better compression than our previous approach, and promises improved generalization for evolved controllers. Results for a high-dimensional Octopus arm control problem show that a high fitness 3680-weight network can be encoded using less than 10 coefficients using the frequencies identified by GCNS. © 2012 Springer-Verlag.

UR - http://link.springer.com/10.1007/978-3-642-32937-1_34

UR - http://www.scopus.com/inward/record.url?scp=84866412462&partnerID=8YFLogxK

U2 - 10.1007/978-3-642-32937-1_34

DO - 10.1007/978-3-642-32937-1_34

M3 - Conference contribution

SN - 9783642329364

SP - 337

EP - 346

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

ER -

Generalized compressed network search

Abstract

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this