Showing papers presented at "Granular Computing in 1987"

Statistical heuristic search

Abstract: Heuristic search is a graph search procedure which uses heuristic information from sources outside the graph. But for many known algorithms, the computational complexity depends on the precision of the heuristic estimates, and for lack of global view in the search process the exponential explosion will be encountered when the node evaluation function estimated is not very precise. Based on the similarity between the statistical inference and heuristic search, we consider a heuristic search as a random sampling process, and treat evaluation functions as random variables. Once a searching direction is chosen, it's regarded as if making a statistical inference. By transferring the statistical inference techniques to the heuristic search, a new search method called statistical heuristic search algorithm, SA for short, is obtained. The procedure of SA is divided into two steps hierarchically. First it identifies quickly the most promising subpart (sub-tree) of a search graph by using some statistical inference method. The sub-trees which contain the goal with lower probability are rejected (pruned). The most promising one is selected. Second, it expands nodes within the selected sub-tree using some common heuristic search algorithm. These two steps are used alternately. The statistical heuristic search can be regarded as one of the applications of multi-granular computing.