From Contrastive to Abductive Explanations and Back Again

Alexey Ignatiev (1) , Nina Narodytska (2) , Nicholas Asher (3, 4) , Joao Marques-Silva (5)

1 Monash University [Melbourne]
2 VMware
3 IRIT-MELODI - MEthodes et ingénierie des Langues, des Ontologies et du DIscours
4 CNRS - Centre National de la Recherche Scientifique
5 IRIT-ADRIA - Argumentation, Décision, Raisonnement, Incertitude et Apprentissage

Explanations of Machine Learning (ML) models often address a Open image in new window question. Such explanations can be related with selecting feature-value pairs which are sufficient for the prediction. Recent work has investigated explanations that address a Open image in new window question, i.e. finding a change of feature values that guarantee a change of prediction. Given their goals, these two forms of explaining predictions of ML models appear to be mostly unrelated. However, this paper demonstrates otherwise, and establishes a rigorous formal relationship between Open image in new window and Open image in new window explanations. Concretely, the paper proves that, for any given instance, Open image in new window explanations are minimal hitting sets of Open image in new window explanations and vice-versa. Furthermore, the paper devises novel algorithms for extracting and enumerating both forms of explanations.

Domaines: Apprentissage [cs.LG]

Alexey Ignatiev, Nina Narodytska, Nicholas Asher, Joao Marques-Silva. From Contrastive to Abductive Explanations and Back Again. International Conference of the Italian Association for Artificial Intelligence (AIxIA 2020), Italian Association for Artificial Intelligence, Nov 2020, Virtual, Italy. pp.335-355, ⟨10.1007/978-3-030-77091-4_21⟩. ⟨hal-03317618⟩