Mean-field approximation with neural network

Abstract

Mean-field approximation is a powerful method for finding minimum points of cost or energy functions. The method has similarities to Boltzmann machines, as both methods are based on simulated annealing in order to avoid local minimum. Mean-field approximation is a deterministic method that uses the results of spin-glass theory. In this paper the solution of a large size constraint satisfaction problem is described. In the radio link frequency assignment problem frequencies from a given set should be assigned to numerous radio links such that the assignments should satisfy predefined constraints. The paper contains the description of the applied method and the results of the simulations.