A preferential interpretation of MultiLayer Perceptrons in a conditional logic with typicality
This work addresses the interpretability of neural networks for researchers in AI and knowledge representation, but it appears incremental as it builds on existing semantics and methods.
The paper tackles the problem of interpreting multilayer perceptrons (MLPs) by relating them to a multipreferential semantics for defeasible reasoning in knowledge representation, using a concept-wise approach to verify conditional properties of MLPs.
In this paper we investigate the relationships between a multipreferential semantics for defeasible reasoning in knowledge representation and a multilayer neural network model. Weighted knowledge bases for a simple description logic with typicality are considered under a (many-valued) ``concept-wise" multipreference semantics. The semantics is used to provide a preferential interpretation of MultiLayer Perceptrons (MLPs). A model checking and an entailment based approach are exploited in the verification of conditional properties of MLPs.