石井信
Chaotic Potts Spin Model for
Combinatorial Optimization Problems
Abstract:In the analog Hopfield network and the mean field theory model of the Boltzmann machine, there occur bifurcations of solutions according to the symmetry of the energy function. This also holds in the Potts neural network. In this report, we investigate the bifurcation processes of the Potts mean field theory equation applied to traveling salesman problems and show some limitations of the annealing procedure. As an alternative approach, we propose a nonequilibrium version of the Potts neural network model, which is called Chaotic Potts Spin (CPS). We show experimental results for a comparison with the mean field annealing and the Potts mean field annealing. We also describe a modified algorithm in which a heuristic method is employed.