How well do differential neural nets perform in casual inference ? There seems to be a pretty good library from sciml that claims to learn models from limited d

How well do differential neural nets perform in casual inference ? There seems to be a pretty good library from sciml that claims to learn models from limited data but wondering if they generalize well