Exploiting non-monotonic influences in qualitative belief networks

Abstract

In A qualitative belief network, dependences between variables are indicated by qual-itative signs These signs serve to model monotoniecprobabilistic relationships only: non-monotonic relationships between variables are modelled as lack of information. In this paper, we propose to include information about non-monotonic probabilistic influences between variables explicitly in a qualitative belief network We show that this information can be exploited in probabilistic inference to forestall unnecessarily weak results.