Preserving Fairness in AI under Domain Shift

Abstract

Existing algorithms for ensuring fairness in AI use a single-shot training strategy, where an AI model is trained on an annotated training dataset with sensitive attributes and then fielded for utilization. This training strategy is effective in problems with stationary distributions, where both the training and testing data are drawn from the same distribution. However, it is vulnerable with respect to distributional shifts in the input space that may occur after the initial training phase. As a result, the time-dependent nature of data can introduce biases and performance degradation into the model predictions, even if the model is initially fair. Model retraining from scratch using a new annotated dataset is a naive solution that is expensive and time-consuming. We develop an algorithm to adapt a fair model to remain fair and generalizable under domain shift using solely new unannotated data points. We recast this learning setting as an unsupervised domain adaptation (UDA) problem. Our algorithm is based on updating the model such that the internal representation of data remains unbiased despite distributional shifts in the input space. We provide empirical validation on three common fairness datasets to show that the challenge exists in practical setting and to demonstrate the effectiveness of our algorithm.