Certifying Fairness of Probabilistic Circuits

Nikil Roashan Selvam, Guy Van den Broeck, YooJung Choi

Abstract: With the increased use of machine learning systems for decision making, questions about their fairness properties start to take center stage. A recently introduced notion of fairness asks whether the model exhibits a \textit{discrimination pattern}, in which an individual---characterized by (partial) feature observations---receives vastly different decisions merely by disclosing some sensitive attributes. Existing work on checking the presence of such patterns is limited to naive Bayes classifiers, which make strong independence assumptions. This paper proposes an algorithm to search for discrimination patterns in a more general class of probabilistic models---probabilistic circuits. If a model is not fair, it may be useful to quickly find discrimination patterns and distill them for better interpretability. As such, we also propose a sampling-based approach to more efficiently mine discrimination patterns, and introduce new classes of discrimination patterns: minimal, maximal, and Pareto optimal.

Community Implementations: [ 2 code implementations](https://www.catalyzex.com/paper/arxiv:2212.02474/code)