Projections, Learning, and Sparsity for Efficient Data Processing