Backward explanations via redefinition of predicates

Léo Saulières (1, 2) , Martin Cooper (1, 2) , Florence Dupin de Saint-Cyr (1, 2, 3)

1 IRIT-ADRIA - Argumentation, Décision, Raisonnement, Incertitude et Apprentissage
2 UT3 - Université Toulouse III - Paul Sabatier
3 Lab-STICC_COMMEDIA - Equipe COgnition, Models and Machines for Engaging Digital Interactive Applications

History eXplanation based on Predicates (HXP), studies the behavior of a Reinforcement Learning (RL) agent in a sequence of agent's interactions with the environment (a history), through the prism of an arbitrary predicate. To this end, an action importance score is computed for each action in the history. The explanation consists in displaying the most important actions to the user. As the calculation of an action's importance is #W[1]-hard, it is necessary for long histories to approximate the scores, at the expense of their quality. We therefore propose a new HXP method, called Backward-HXP, to provide explanations for these histories without having to approximate scores. Experiments show the ability of B-HXP to summarise long histories.

Mots clés: Machine Learning; Reinforcement Learning; Explainability; Interpretability; History Explanation; Importance score

Domaines: Intelligence artificielle [cs.AI]; Apprentissage [cs.LG]

Léo Saulières, Martin Cooper, Florence Dupin de Saint-Cyr. Backward explanations via redefinition of predicates. 27th European Conference on Artifical Intelligence (ECAI 2024), European Association for Artificial Intelligence (EurAI); Spanish Artificial Intelligence Society (AEPIA), Oct 2024, Saint Jacques De Compostelle, Spain. à paraître. ⟨hal-04669413v2⟩