Characterization of a New Restart Strategy for Randomized Backtrack Search

We propose an improved restart strategy for randomized backtrack search, and evaluate its performance by comparing to other heuristic and stochastic search techniques for solving random problems and a tight real-world resource allocation problem. The restart strategy proposed by Gomes et al. [1] requires the specification of a cutoff value determined from an overall profile of the cost of search for solving the problem. When no such profile is known, the cutoff value is chosen by trial-and-error. The Randomization and Geometric Restart (RGR) proposed by Walsh does not rely on a cost profile but determines the cutoff value as a function of a constant parameter and the number of variables in the problem [2]. Unlike these strategies, which have fixed restart schedules, our technique (RDGR) dynamically adapts the value of the cutoff parameter to the results of the search process. Our experiments investigate the behavior of these techniques using the cumulative distribution of the solutions, over different run-time durations, values of the cutoff, and problem types (i.e., a real-world resource allocation problem and randomly-generated binary constraint satisfaction problems). We show that distinguishing between solvable and over-constrained problem instances in our real-world case-study yields new insights on the relative performance of the search techniques tested. We propose to use this characterization as a basis for building new strategies of cooperative, hybrid search.