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