Individual Fairness in Bayesian Neural Networks

TL;DR

This paper investigates individual fairness in Bayesian neural networks, proposing a framework for systematic estimation and demonstrating that BNNs tend to be more fair than deterministic models across various benchmarks.

Contribution

It introduces a novel framework for estimating individual fairness in BNNs and extends gradient-based attacks to be fairness-aware, with empirical evidence showing BNNs are more fair than deterministic models.

Findings

BNNs are more individually fair than deterministic models.

The proposed framework effectively estimates $oldsymbol{ ext{ε-δ}}$-IF.

Fairness-aware attacks can be systematically designed for BNNs.

Abstract

We study Individual Fairness (IF) for Bayesian neural networks (BNNs). Specifically, we consider the $\epsilon$-$\delta$-individual fairness notion, which requires that, for any pair of input points that are $\epsilon$-similar according to a given similarity metrics, the output of the BNN is within a given tolerance $\delta>0.$ We leverage bounds on statistical sampling over the input space and the relationship between adversarial robustness and individual fairness to derive a framework for the systematic estimation of $\epsilon$-$\delta$-IF, designing Fair-FGSM and Fair-PGD as global,fairness-aware extensions to gradient-based attacks for BNNs. We empirically study IF of a variety of approximately inferred BNNs with different architectures on fairness benchmarks, and compare against deterministic models learnt using frequentist techniques. Interestingly, we find that BNNs trained by means of approximate Bayesian inference consistently tend to be markedly more individually fair than their deterministic counterparts.