Towards a stochastic neural model for combinatorial optimization

Summary form only given, as follows. A stochastic neural model is proposed for the solution of hard combinatorial optimization problems, inspired by the Hopfield-Tank model and the stochastic search ability of simulated annealing. The authors start with a discrete-time algorithm for the simulation of a Hopfield-Tank network and modify it through the incorporation of probabilistic decisions which allow energy increases at each time step. The algorithm has been tested on the traveling salesman problem (TSP), and the results are encouraging. For these tests the authors used a formulation of TSP which ensures that every stable state of the network corresponds to a feasible TSP tour. From the simulation algorithm they then extract some possible foundations for the formalization of their model.<>