Towards Rigorous Interpretations: a Formalisation of Feature Attribution

                Darius Afchar 1 2 Romain Hennequin 1 Vincent Guigue 2

             Abstract                 controlling and disentangling every aspect of a model pre-
                                  diction, ranging from how the data is inherently structured,
   Feature attribution is often loosely presented        to what priors are induced by a model architecture, to what
   as the process of selecting a subset of rel ...