A Reconstruction Error Formulation for Semi-Supervised Multi-task and Multi-view Learning