Efficient Counterexample-Guided Fairness Verification and Repair of Neural Networks Using Satisfiability Modulo Convex Programming

Arya Fayyazi, Yifeng Xiao, Pierluigi Nuzzo, Massoud Pedram

Keywords: Fairness Verification, Neural Network Repair, Counterexample-Guided Synthesis

TL;DR: This paper proposes an efficient framework for verifying and repairing neural network fairness via sensitivity-guided neuron updates, ensuring provable convergence in finite steps with minimal impact on accuracy.

Abstract: Ensuring fairness is essential for ethical decisionmaking in various domains. Informally, a neural network is considered fair if and only if it treats similar individuals similarly in a given task. We introduce FaVeR (Fairness Verification and Repair), a framework for efficiently verifying and repairing pre-trained neural networks with respect to individual fairness properties. FaVeR ensures fairness via iterative search of high-sensitivity neurons and backward adjustment of their weights, guided by counterexamples generated from fairness verification using satisfiability modulo convex programming. By addressing fairness at the neuron level, FaVeR minimizes the impact of neural network repair on the overall performance. Experimental evaluations on common fairness datasets show that FaVeR achieves a 100% fairness repair rate across all models, with accuracy reduction of less than 2.27%. Moreover, its significantly lower average runtime makes it suitable for practical applications.

Paper Type: Previously Published Paper

Venue For Previously Published Paper: IJCAI 2025

Submission Number: 17