论文信息 - Generalized compressed network search

Generalized compressed network search

This paper presents initial results of Generalized Compressed Network Search (GCNS), a method for automatically identifying the important frequencies for neural networks encoded as a set of Fourier-type coefficients (i.e. "compressed" networks). GCNS achieves better compression than our previous approach, and promises better generalization capabilities. 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.

Jürgen Schmidhuber | Faustino J. Gomez | Rupesh Kumar Srivastava

[1] Kalyanmoy Deb,et al. Messy Genetic Algorithms: Motivation, Analysis, and First Results , 1989, Complex Syst..

[2] Tamar Flash,et al. Dynamic model of the octopus arm. I. Biomechanics of the octopus reaching movement. , 2005, Journal of neurophysiology.

[3] Jürgen Schmidhuber,et al. Compressed Network Complexity Search , 2012, PPSN.

[4] Jürgen Schmidhuber,et al. Evolving neural networks in compressed weight space , 2010, GECCO '10.

[5] Hiroaki Kitano,et al. Designing Neural Networks Using Genetic Algorithms with Graph Generation System , 1990, Complex Syst..

[6] Dario Floreano,et al. Neuroevolution with Analog Genetic Encoding , 2006, PPSN.

[7] Jürgen Schmidhuber,et al. Searching for Minimal Neural Networks in Fourier Space , 2010, AGI 2010.

[8] Corso Elvezia,et al. Discovering Neural Nets with Low Kolmogorov Complexity and High Generalization Capability , 1997 .

[9] Kalyanmoy Deb,et al. RapidAccurate Optimization of Difficult Problems Using Fast Messy Genetic Algorithms , 1993, ICGA.