Tech Report CS-90-14

On Irrelevance and Partial Assignments to Belief Networks

Solomon E. Shimony

October 1990

Abstract:

We evaluate current methods for finding best consistent explanations. The only known general probabilistic scheme for finding a consistent explanation, that of complete MAPs, suffers from the over-specification problem. We propose a partial solution by applying a probabilistic generalization to logical irrelevance. We evaluate const-based abduction in a probabilistic context by providing a probabilistic semantics for it in the case of partial assignments.

(complete text in pdf)