Affiliations:
Institute of Naval Weapon, Polish Naval Academy, Smidowicza, Gdynia, Poland
Correspondence:
[*]
Corresponding author. Tomasz Praczyk, Institute of Naval Weapon, Polish Naval Academy, Smidowicza 69, 81-116, Gdynia, Poland. Tel.: +48 261 262 703; Fax: +48 261 262 503; E-mail: t.praczyk@amw.gdynia.pl.
Abstract: The paper presents Cooperative Co–Evolutionary Neural Networks (CCENN), that is, a new method for evolving modular artificial neural networks (MANN). In CCENN, individual module–networks evolve in separate populations and to form a complete MANN each population delegates a single module. Modules collaborating within the same artificial neural network (ANN) are not connected and they work like networks in an ensemble–based approach, i.e. output of a complete ANN is determined based on a negotiation process between the module–networks. A module with the greatest negotiation strength is allowed to set one of the outputs of the entire ANN, to fix all the outputs, the modules negotiate many times. To test performance of CCENN, it was used to evolve neuro–controllers for a team of underwater vehicles whose common goal was to capture other vehicle behaving by a deterministic strategy (predator–prey problem). The experiments were carried out in simulation whereas their results were used to compare CCENN with two other neuro–evolutionary methods designed for building monolithic ANNs.
Keywords: Evolutionary modular neural networks, underwater vehicles, predator–prey problem
