A Maximal Correlation Approach to Imposing Fairness in Machine Learning