A computational model for causal and diagnostic reasoning in inference systems 论文

摘要

This paper introduces a representation of evidential relationships which permits updating of belief in two simultaneous modes: causal (i. e. top-down) and diagnostic (i.e. bottom-up). It extends the hierarchical tree representation by allowing multiple causes to a given manifestation. We develop an updating scheme that obeys the axioms of probability, is computationally efficient, and is compatible with experts reasoning. The belief parameters of each variable are defined and updated by those of its neighbors in such a way that the impact of each new evidence propagates and settles through the network in a single pass.