Approaching Machine Learning Fairness through Adversarial Network