4 ms·

You're right - a LOT of work goes into this (eg GANs) because these large nNets just start "remembering" the training data.