Suboptimum solutions obtained by the Hopfield-Tank neural network algorithm

Abstract

The neural network method of Hopfield and Tank claims to be able to find nearly-optimum solutions for discrete optimization problems, e.g. the travelling salesman problem. In the present paper, an example is given which shows that the Hopfield-Tank algorithm systematically prefers certain solutions even if the energy values of these solutions are clearly higher than the energy of the global minimum. © 1991 Springer-Verlag.