A case-based filter for diagnostic belief networks

Abstract

Special-case algorithms for Bayesian belief networks are designed to alleviate the computational burden of problem solving. These algorithms provide a case base for storing solutions for a small number of situations that are likely to be en- countered during problem solving. This case base is employed as a lter for belief-network inference: for a problem under consideration, the network at hand is consulted only if the case base does not provide a solution for the problem. We present a new algorithm that further extends on the basic idea of special-case al- gorithms by exploiting knowledge about the way diagnostic problem solving with a belief network is shaped.