Qualitativeprobabilistic reasoningin a Bayesiannetworkoften reveals tradeoffs: relationships that are ambiguousdue to competingqualitative influences. Wepresent twotechniquesthat combinequalitative and numericprobabilistic reasoningto resolve such tradeoffs, inferring the qualitative relationship betweennodesin a Bayesiannetwork.Thefirst approach incrementallymarginalizesnodesin network,andthe secondincrementallyrefines the state spacesof randomvariables. Bothprovidesystematicmethodsfor tradeoffresolutionat potentiallylowercomputational cost than applicationof purely numericmethods.
Chao-Lin Liu, Michael P. Wellman