Failure-Driven Refinement Search with Local Repair-Based Heuristics for Constraint Satisfaction Problems

There has been substantial recent interest in two new families of search techniques. One family consists of systematic approaches, which use the idea of dependency directed backtracking (DDB) and dynamic backtracking (DB) as an antidote for the inefficiencies of chronological backtracking. Based on backtracking, systematic approaches are able to produce an optimal assignment, when no time limit is imposed. The other contains non-systematic methods such as Min-Conflict Repair Heuristic or Selman’s GSAT. Based on local improvement mechanisms, these methods cann’t guarantee optimality, but may produce better quality assignments in a limited time. The main attractions are their reactivity and applicability to optimization problems. Our aim in this paper is to describe a new search procedure that combines the benefits of both of the earlier approaches. To achieve this goal, we describe a complete ’ hybrid method called failure driven search control with min-conflict repair (FDB-MC) for solving constraint satisfaction problems. The failure explanation-backmarking process controls the search space of this combined approach, a method of learning constraints during search (at each failure point) that may be used to avoid the repeated traversing of failed path in a search tree. The approach is based on repair-based algorithms, and can be proved to find an optimal solution