An Improved Maximum Neural Network with Stochastic Dynamics Characteristic for Maximum Clique Problem

Gang Yang,Zheng Tang,Hongwei Dai

DOI: https://doi.org/10.1541/ieejeiss.128.94

2008-01-01

Abstract:Through analyzing the dynamics characteristic of maximum neural network with an added vertex, we find that the solution quality is mainly determined by the added vertex weights. In order to increase maximum neural network ability, a stochastic nonlinear self-feedback and flexible annealing strategy are embedded in maximum neural network, which makes the network more powerful to escape local minima and be independent of the initial values. Simultaneously, we present that solving ability of maximum neural network is dependence on problem. We introduce a new parameter into our network to improve the solving ability. The simulation in k random graph and some DIMACS clique instances in the second DIMACS challenge shows that our improved network is superior to other algorithms in light of the solution quality and CPU time.