We show how and why to use a financially meaningful differential regularization method when pricing options by Monte Carlo simulation, be that in polynomial regression or neural network context.