Probabilistic Verification of Fairness Properties via Concentration

TL;DR

This paper introduces VeriFair, a scalable probabilistic verification algorithm for fairness in machine learning, leveraging adaptive concentration inequalities to handle large models with strong correctness guarantees.

Contribution

It presents a novel scalable algorithm using adaptive concentration inequalities for probabilistic fairness verification, capable of verifying large neural networks.

Findings

Successfully verified a deep recurrent neural network over five orders of magnitude larger than previous models.

Provides probabilistic correctness guarantees with extremely low error probabilities.

Demonstrates scalability and effectiveness of the VeriFair tool for fairness verification.

Abstract

As machine learning systems are increasingly used to make real world legal and financial decisions, it is of paramount importance that we develop algorithms to verify that these systems do not discriminate against minorities. We design a scalable algorithm for verifying fairness specifications. Our algorithm obtains strong correctness guarantees based on adaptive concentration inequalities; such inequalities enable our algorithm to adaptively take samples until it has enough data to make a decision. We implement our algorithm in a tool called VeriFair, and show that it scales to large machine learning models, including a deep recurrent neural network that is more than five orders of magnitude larger than the largest previously-verified neural network. While our technique only gives probabilistic guarantees due to the use of random samples, we show that we can choose the probability of error to be extremely small.