A non-discriminatory approach to ethical deep learning

TL;DR

This paper introduces NDR, a regularization method that reduces discrimination in neural networks by hiding sensitive features, achieving fairer models with minimal performance impact.

Contribution

The paper presents a novel non-discriminatory regularization strategy (NDR) that prevents neural networks from using discriminatory features during training.

Findings

NDR effectively reduces discriminatory bias in neural networks.

NDR maintains model performance with minimal computational overhead.

Experiments demonstrate NDR's ability to produce fairer models across tasks.

Abstract

Artificial neural networks perform state-of-the-art in an ever-growing number of tasks, nowadays they are used to solve an incredibly large variety of tasks. However, typical training strategies do not take into account lawful, ethical and discriminatory potential issues the trained ANN models could incur in. In this work we propose NDR, a non-discriminatory regularization strategy to prevent the ANN model to solve the target task using some discriminatory features like, for example, the ethnicity in an image classification task for human faces. In particular, a part of the ANN model is trained to hide the discriminatory information such that the rest of the network focuses in learning the given learning task. Our experiments show that NDR can be exploited to achieve non-discriminatory models with both minimal computational overhead and performance loss.