Distributed Learning without Distress: Privacy-Preserving Empirical Risk Minimization