Maximum Entropy and Variable Strength Defaults

Rachel A. Bourne, Simon Parsons

Abstract: A new algorithm for computing the maximum entropy ranking over models is presented. The algorithm handles arbitrary sets of propositional defaults with associated strength assignments and succeeds whenever the set satisfies a robustness condition. Failure of this condition implies the problem may not be sufficiently specified for a unique solution to exist. This work extends the applicability of the maximum entropy approach detailed in [Goldszmidt et al., 1993]) and clarifies the assumptions on which the method is based.