Comment by whinvik

I think a lot of machine learning applied to computational science problems is basically function approximation. However a lot of computational science problems actually are => here's a good basis function to approximate functions, now use it to solve an ODE/PDE. It seems machine learning is eternally stuck in the former.