Prior work has shown that features which appear to be biologically plausible as well as empirically useful can be found by sparse coding with a prior such as a Laplacian (L1 ) that promotes sparsity.
