Weighted defeasible knowledge bases and a multipreference semantics for a deep neural network model

This work aims to provide a theoretical link between defeasible reasoning and deep neural networks, which could be significant for researchers working on explainable AI and knowledge representation.

This paper explores the relationship between multipreferential semantics for defeasible reasoning in knowledge representation and deep neural networks. It considers weighted knowledge bases for description logics under a concept-wise multipreference semantics, extending it to fuzzy interpretations to provide a preferential interpretation of Multilayer Perceptrons.