Does Training Lead to the Formation of Modules in Threshold Networks?

Delphine Nicolay; Roli Andrea; Timoteo Carletti

Does Training Lead to the Formation of Modules in Threshold Networks?

Delphine Nicolay, Roli Andrea, Timoteo Carletti

Namur Institute for Complex Systems

Research output: Contribution in Book/Catalog/Report/Conference proceeding › Conference contribution

31 Downloads (Pure)

Abstract

This paper addresses the question to determine the necessary
conditions for the emergence of modules in the framework of artificial
evolution. In particular, threshold networks are trained as controllers
for robots able to perform two different tasks at the same time. It is
shown that modules do not emerge under a wide set of conditions in our
experimental framework. This finding supports the hypothesis that the
emergence of modularity indeed depends upon the algorithm used for
artificial evolution and the characteristics of the tasks.

Original language	English
Title of host publication	Proceedings of ECCS 2014
Subtitle of host publication	European Conference on Complex Systems
Publisher	Springer
Pages	181-192
Number of pages	12
ISBN (Electronic)	978-3-319-29228-1
ISBN (Print)	978-3-319-29226-7
Publication status	Published - 1 May 2016
Event	eccs'14 - Lucca, Italy Duration: 22 Sept 2014 → 26 Sept 2014

Publication series

Name	Springer Proceedings in Complexity
Publisher	Springer

Scientific committee

Scientific committee	eccs'14
Country/Territory	Italy
City	Lucca
Period	22/09/14 → 26/09/14

Keywords

neural networks
learning
genetic algorithm
modularity
evolutionary robotics

Access to Document

NicolayEtAlFinal published version, 369 KB

Cite this

@inproceedings{8e1ebee40c7647a0a129e5e95d1c96e4,

title = "Does Training Lead to the Formation of Modules in Threshold Networks?",

abstract = "This paper addresses the question to determine the necessaryconditions for the emergence of modules in the framework of artificialevolution. In particular, threshold networks are trained as controllersfor robots able to perform two different tasks at the same time. It isshown that modules do not emerge under a wide set of conditions in ourexperimental framework. This finding supports the hypothesis that theemergence of modularity indeed depends upon the algorithm used forartificial evolution and the characteristics of the tasks.",

keywords = "neural networks, learning, genetic algorithm, modularity, evolutionary robotics",

author = "Delphine Nicolay and Roli Andrea and Timoteo Carletti",

year = "2016",

month = may,

day = "1",

language = "English",

isbn = "978-3-319-29226-7",

series = "Springer Proceedings in Complexity",

publisher = "Springer",

pages = "181--192",