BLS: A learning search algorithm with Bayesian learning

In learning search, it is assumed that a search algorithm has adequate opportunities to traverse the problem state space, and to sense and learn the topological (or probability) knowledge about the state space. In this paper, a new search algorithm based on Bayesian learning (algorithm BLS) is proposed, which finds the solution path for any given state pair (s, t) in the state space efficiently. The algorithm depicts the search as a process of pattern recognition for the current path. During its depth-first search, the BLS classifies the current path, depending on the probability distribution functions acquired in previous Bayesian learning, and prunes it if necessary. It is proved that the mean complexity of BLS grows linearly with the length of the solution path after sufficient training.