Structured Learning With Scale Mixture Priors