Evolving Efficient Connection for the Design of Artificial Neural Networks

  • Authors:
  • Min Shi;Haifeng Wu

  • Affiliations:
  • Department of Computer and Information Science, Norwegian University of Science and Technology, Norway;Department of Engineering Cybernetics, Norwegian University of Science and Technology, Norway

  • Venue:
  • ICANN '08 Proceedings of the 18th international conference on Artificial Neural Networks, Part II
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

Most of the recent neuroevolution (NE) approaches explore new network topologies based on a neuron-centered design principle. So far evolving connections has been poorly explored. In this paper, we propose a novel NE algorithm called Evolving Efficient Connections (EEC), where the connection weights and the connection paths of networks are evolved separately. We compare our new method with standard NE and several popular NE algorithms, SANE, ESP and NEAT. The experimental results indicate evolving connection weights along with connection paths can significantly enhance the performance of standard NE. Moreover the performances of cooperative coevolutionary algorithms are superior to non-cooperative evolutionary algorithms.