Generalized Compressed Network Search

Rupesh Kumar Srivastava, Jürgen Schmidhuber, and Faustino Gomez

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.

LNCS 7491, p. 337 ff.

Full article in PDF | BibTeX

lncs@springer.com
© Springer-Verlag Berlin Heidelberg 2012