Seonglae ChoIt is good to separate out hard zeros for interpretability. Thisis because sparsity causes a dirac delta in density at zero.
But Is sparsity actually a good proxy for interpretability? is still open question
Linear model or tree-based model are considered to have intrinsic interpretability
Interpretable Sparse Codings
Regularization forest
VAE
helpful for insects
Brain sprase coding
odd cell dense
Sparse
Superposition
Place Cells
