Design and Analysis of Competition-Based Neural Networks
Tao Li
Department of Computer Science, Concordia University,
1455 de Maisonneuve Blvd. W., Montreal, Quebec H3G 1M8, Canada
Yun Peng
Department of Computer Science, University of Maryland Baltimore County,
Baltimore, MD 21228, USA
Abstract
A simple and systematic approach to the design of neural networks for combinatorial optimization is presented in this paper. This approach is based on competition among the neurons of a network. Our approach relies on the use of simple heuristics in network design. The activation rule governing neuron behavior is derived by breaking down the global constraints for a problem into local constraints for individual neurons. The local constraints can be expressed as simple heuristic rules for network design. Using this approach, high performance neural networks for optimization can be readily designed and the global properties can be easily verified. Simulation results indicate that this approach can outperform some sequential heuristic algorithms and simulated annealing algorithms.
