Post-processing Counterexample-guided Fairness Guarantees in Neural Networks

Kiarash Mohammadi, Aishwarya Sivaraman, Golnoosh Farnadi

Keywords: fairness, neural networks, verification

TL;DR: An approach to guarantee individual fairness in neural networks at prediction time using verification.

Abstract: There is an increasing interest in adopting high-capacity machine learning models such as deep neural networks to semi-automate human decisions. Hence, it is crucial that these models guarantee similar decisions for similar individuals. To ensure such a fair decision, it is necessary to construct tools capable of verifying and enforcing fairness constraints. In this work, we propose methods to guarantee fairness of a neural network via verification using mixed-integer programming. We show, that it is possible to guarantee individual fair prediction without intervening in the model, efficiently and with little to no loss in accuracy.