A discrete stochastic neural network algorithm for constraint satisfaction problems

An artificial neural network algorithm for solving constraint-satisfaction problems (CSPs) is described. The algorithm is based on the discrete Hopfield network but differs from it primarily in that auxiliary networks (guards) are asymmetrically coupled to the main network to enforce certain types of constraints. Although the presence of asymmetric connections implies that the network may not converge, it is found that, for certain classes of problems, the network often quickly converges to find satisfying solutions when they exist. The network. referred to as the guarded discrete stochastic network, can run efficiently on serial machines and can find solutions to very large problems; among the CSPs studied is the N-queen problem, for which solutions for N up to 1024 (>16⁶ neurons) have been found. The authors describe the behavior of the network several types of CSPs and discuss the heuristics that are implicit in its operation