Distance Correlation GAN: Fair Tabular Data Generation with Generative Adversarial Networks

With the growing impact of artificial intelligence, the topic of fairness in AI has received increasing attention for valid reasons. In this paper, we propose a generative adversarial network for fair tabular data generation. The model is a WGAN, where the generator is enforcing fairness by penalizing distance correlation between protected attribute and target attribute. We compare our results with another state-of-the-art generative adversarial network for fair tabular data generation and a preprocessing repairment method on four datasets, and show that our model is able to produce synthetic data, such that training a classifier on it results in a fair classifier, beating the other two methods. This makes the model suitable for applications that concern with fairness and preserving privacy.