年間 12 号発行
ISSN 印刷: 1064-2315
ISSN オンライン: 2163-9337
Indexed in
Coevolving Feedforward Neural Networks
要約
An evolutionary algorithm of determining the architecture of feedforward neural networks and their training is proposed, based on the coevolutionary models of cooperation and competition with using of clustering algorithms for partitioning the main problem of neural network synthesis into subtasks which are to be solved in certain sub-populations. The proposed algorithm implements an environment that is conducive to cooperation and competition of populations in which every individual is a feedforward neural network, and the totality of the populations is responsible for the final solution of the set problem. The simulation results confirm the effectiveness of the proposed method of feedforward neural networks synthesis.