Co-evolution of Fitness Predictors and Deep Neural Networks

Deep neural networks proved to be a very useful and powerful tool with many applications. In order to achieve good learning results, the network architecture has, however, to be carefully designed, which requires a lot of experience and knowledge. Using an evolutionary process to develop new network...

Full description

Saved in:
Bibliographic Details
Published inParallel Processing and Applied Mathematics Vol. 10777; pp. 555 - 564
Main Authors Funika, Włodzimierz, Koperek, Paweł
Format Book Chapter
LanguageEnglish
Published Switzerland Springer International Publishing AG 2018
Springer International Publishing
SeriesLecture Notes in Computer Science
Subjects
Co-evolution
Deep neural networks
Evolutionary programming
Fitness predictors
Genetic programming
Neural networks
Online AccessGet full text
ISBN9783319780238
3319780239
ISSN0302-9743
1611-3349
DOI10.1007/978-3-319-78024-5_48

Cover

More Information
Summary:Deep neural networks proved to be a very useful and powerful tool with many applications. In order to achieve good learning results, the network architecture has, however, to be carefully designed, which requires a lot of experience and knowledge. Using an evolutionary process to develop new network topologies can facilitate this process. The limiting factor is the speed of evaluation of a single specimen (a single network architecture), which includes learning based on a large dataset. In this paper we propose a new approach which uses subsets of the original training set to approximate the fitness. We describe a co-evolutionary algorithm and discuss its key elements. Finally we draw conclusions from experiments and outline plans for future work.
ISBN:9783319780238
3319780239
ISSN:0302-9743
1611-3349
DOI:10.1007/978-3-319-78024-5_48