Abstract

PatternAttribution is a recent method, introduced in the vision domain, thatexplains classifications of deep neural networks. We demonstrate that it alsogenerates meaningful interpretations in the language domain.

 

Quick Read (beta)