Regularization Strategies and Empirical Bayesian Learning for MKL