On Training Derivative-Constrained Neural Networks

KaiChieh Lo, Daniel Huang

We refer to the setting where the (partial) derivatives of a neural network's (NN's) predictions with respect to its inputs are used as additional training signal as a derivative-c…