A Generic Neural Network Approach For Constraint Satisfaction Problems

Abstract

The Constraint Satisfaction Problem (CSP) is a mathematical abstraction of the problems in many AI application domains. In many of such applications timely response by a CSP solver is so crucial that to achieve it, the user may be willing to sacrifice completeness to a certain extent. This paper describes a neural network approach for solving CSPs which aims at providing prompt responses. The effectiveness of this model, which is called GENET, in solving CSPs with binary constraints is demonstrated by a simulator. Although the completeness is not guaranteed, as in the case of most of the existing stochastic search techniques, solutions have been found by the GENET simulator in all of our randomly generated problems tested so far. Since the neural network model lends itself to the VLSI implementation of parallel processing architectures, the limited number of cycles required by GENET to find the solutions for the tested problems gives hope for solving large CSPs in a fraction of the time required by conventional methods.