Resolution of inconsistent revision problems in Markov networks

Abstract

The item planning system at Volkswagen uses Markov networks in its core, with revision of item demands being one of the most important operations. The represented dependencies and given stipulations of the application are highly complex and make the revision problems prone to inconsistencies. We present an approach that solves inconsistent revision problems by fair adaptations of the conflicting probability assignments and fits neatly into the knowledge-based setting of Markov networks. © 2013 Springer-Verlag.