Empirical Risk Minimization under Random Censorship: Theory and Practice