A Tale of Fairness Revisited: Beyond Adversarial Learning for Deep Neural Network Fairness

TL;DR

This paper offers a theoretical analysis of adversarial training methods for fairness in deep neural networks, highlighting the inherent tradeoffs and proposing mechanisms to mitigate performance sacrifices.

Contribution

It advances understanding of fairness in deep learning by analyzing tradeoffs and introducing mechanisms to improve fairness without severely impacting accuracy.

Findings

Identifies persistent tradeoff between fairness and performance.

Provides theoretical insights into adversarial training for fairness.

Explores mechanisms to offset fairness-performance tradeoff.

Abstract

Motivated by the need for fair algorithmic decision making in the age of automation and artificially-intelligent technology, this technical report provides a theoretical insight into adversarial training for fairness in deep learning. We build upon previous work in adversarial fairness, show the persistent tradeoff between fair predictions and model performance, and explore further mechanisms that help in offsetting this tradeoff.