Reduced Products of Abstract Domains for Fairness Certification of Neural Networks

Abstract

We present Libra, an open-source abstract interpretation-based static analyzer for certifying fairness of ReLU neural network classifiers for tabular data. Libra combines a sound forward pre-analysis with an exact backward analysis that leverages the polyhedra abstract domain to provide definite fairness guarantees when possible, and to otherwise quantify and describe the biased input space regions. The analysis is configurable in terms of scalability and precision. We equipped Libra with new abstract domains to use in the pre-analysis, including a generic reduced product domain construction, as well as search heuristics to find the best analysis configuration. We additionally set up the backward analysis to allow further parallelization. Our experimental evaluation demonstrates the effectiveness of the approach on neural networks trained on a popular dataset in the fairness literature.

Publication
In Proc. 28th Static Analysis Symposium (SAS 2021)

The video presentation of the SAS2021 talk will be available soon!

Denis Mazzucato
Denis Mazzucato
Ph.D. Student

My research interests include formal methods of safety-critical software, specifically abstract interpretation applied to artificial intelligence

Caterina Urban
Caterina Urban
Research Scientist (Chargé de Recherche)

I am a member of the INRIA research team ANTIQUE (ANalise StaTIQUE). Previously, I was a postdoctoral researcher at the Chair of Programming Methodology …

Related