Fast greedy algorithms for dictionary selection with generalized sparsity constraints

Neural Information Processing Systems 

In dictionary selection, several atoms are selected from finite candidates that successfully approximate given data points in the sparse representation.