Abstract

Weighted knowledge bases for description logics with typicality have beenrecently considered under a "concept-wise" multipreference semantics (in boththe two-valued and fuzzy case), as the basis of a logical semantics ofMultiLayer Perceptrons (MLPs). In this paper we consider weighted conditionalALC knowledge bases with typicality in the finitely many-valued case, throughthree different semantic constructions. For the boolean fragment LC of ALC weexploit ASP and "asprin" for reasoning with the concept-wise multipreferenceentailment under a phi-coherent semantics, suitable to characterize thestationary states of MLPs. As a proof of concept, we experiment the proposedapproach for checking properties of trained MLPs.

 

Quick Read (beta)