Randomness in heuristics: An experimental investigation for the maximum satisfiability problem

Abstract

In this paper, a genetic algorithm and a tabu search are investigated for the maximum satisfiability problem. When the evolutionary algorithm is hybridized with the randomized procedure G-bit [14], better performance is achieved and it even outperforms the well known probabilistic procedure GSAT[25]. On the other hand, when the random noise strategy is introduced in the tabu search, the latter competes with GSAT with walk [27] independently of the length of the tabu list. The basic result we can argue from this study is that the robustness of a method seems to be bound to the degree of ‘randomness’ involved in it, but at the expense of the running time. According to the experiments, GSAT and the genetic algorithm are more powerful than tabu search in its simplest form because they incorporate more ‘randomness’. GSAT with random walk is even more interesting than simple GSAT for the same reason. Also heuristic methods and local search bocome more efficient when a random strategy such as a noise is introduced to deviate the search from its usual rules.