Improving Genet and Egenet by New Variable Ordering Strategies

Constraint satisfaction problems (CSPs) naturally occur in a number of important industrial applications such as planning and scheduling defeating many algorithmic search methods. GENET and it extended model, EGENET, are probabilistic neural networks which had some remarkable success in solving some hard instances of CSPs such as a set of hard graph coloring problems. Both GENET or EGENET does not employ any variable ordering strategy in its computation to guide the search. In this paper, we proposed several new variable ordering strategies for improving GENET and EGENET. We compared the efficiency of their improved versions using the conventional or new variable ordering strategies against the original GENET and EGENET on a number of randomly generated and hard instances of binary CSPs. The improved versions with variable ordering heuristics compares favorably with the original versions on these binary CSPs. Our work shed lights on several directions for future exploration.