Been reading about machine learning lately. Asked myself, if it would be possible with 3DGS. Tried my luck and voila! It's not fast (obviously), but it turned out better than I expected! It's still "learning", but it can already detect most good-written digits:
I'll let it "learn" a bit more, then I'll upload it with the code so you can have fun with it as well. It was just planned as a prototype anyway. Currently this network has two hidden units and a 10 nodes output layer. I think this will be enough for this task. Also, thanks to MNIST for the handwritten digit database.