Fair Universal Representations using Adversarial Models