Capturing Sparsity in High Dimensional Data