Investigating Trade-offs in Utility, Fairness and Differential Privacy in Neural Networks

To enable an ethical and legal use of machine learning algorithms, they must\nboth be fair and protect the privacy of those whose data are being used.\nHowever, implementing privacy and fairness constraints might come at the cost\nof utility (Jayaraman & Evans, 2019; Gong et al., 2020). This paper\ninvestigates the privacy-utility-fairness trade-off in neural networks by\ncomparing a Simple (S-NN), a Fair (F-NN), a Differentially Private (DP-NN), and\na Differentially Private and Fair Neural Network (DPF-NN) to evaluate\ndifferences in performance on metrics for privacy (epsilon, delta), fairness\n(risk difference), and utility (accuracy). In the scenario with the highest\nconsidered privacy guarantees (epsilon = 0.1, delta = 0.00001), the DPF-NN was\nfound to achieve better risk difference than all the other neural networks with\nonly a marginally lower accuracy than the S-NN and DP-NN. This model is\nconsidered fair as it achieved a risk difference below the strict (0.05) and\nlenient (0.1) thresholds. However, while the accuracy of the proposed model\nimproved on previous work from Xu, Yuan and Wu (2019), the risk difference was\nfound to be worse.\n