Polynomial approximation based learning search

Abstract

In this paper, polynomial approximation method and theory are introduced into the research of learning search of artificial intelligence. By so doing, a learning search algorithm can, after sufficient number of problem-solving, construct a heuristic estimate function h(.) which uniformly approximates to the optimal estimate function h*(.) by arbitrary precision. One of these learning search algorithms, A-B/sub n/, is described and it is shown that, when the number of the previous problem-solving becomes large enough, the worst-case complexity of A-B/sub n/ can be reduced to O(poly(N)), where N is the length of the optimal solution path, poly(N) is a polynomial function of N.<>