On The Impact of Machine Learning Randomness on Group Fairness

TL;DR

This paper investigates how randomness in neural network training affects group fairness measures, identifying data order stochasticity as a key factor and proposing a simple method to improve fairness consistency.

Contribution

It reveals the primary source of variance in fairness measures and demonstrates a straightforward data ordering technique to enhance fairness stability.

Findings

Variance in fairness is due to stochastic training dynamics on under-represented groups.

Data order during training significantly influences group fairness outcomes.

Simple data order adjustments can improve fairness with minimal impact on overall accuracy.

Abstract

Statistical measures for group fairness in machine learning reflect the gap in performance of algorithms across different groups. These measures, however, exhibit a high variance between different training instances, which makes them unreliable for empirical evaluation of fairness. What causes this high variance? We investigate the impact on group fairness of different sources of randomness in training neural networks. We show that the variance in group fairness measures is rooted in the high volatility of the learning process on under-represented groups. Further, we recognize the dominant source of randomness as the stochasticity of data order during training. Based on these findings, we show how one can control group-level accuracy (i.e., model fairness), with high efficiency and negligible impact on the model's overall performance, by simply changing the data order for a single epoch.