A Differentially Private Weighted Empirical Risk Minimization Procedure and its Application to Outcome Weighted Learning