Random Matrices: l1 Concentration and Dictionary Learning with Few Samples