Beam Search with Dynamic Pruning for Artificial Intelligence Hard Problems

The "Beam" search algorithm is derived from the classical Artificial Intelligence discipline and it searches under the strategy of "Space-states-operators", guided by heuristics. Although “Beam” is better than the “Breadth” search strategy, it suffers of some drawbacks. In this paper we propose and describe two innovative dynamic pruning variants of the "Beam" search algorithm. In the experiments that we performed with our two variants we obtained that they were consistently more efficient than the original algorithm, maintaining the solution quality. In particular we observed very significant reductions in the number of states required to reach the problem solution, and consequently reducing the response time.