I have developed a regression CNN model using Pytorch, and I am interested in finding the sensitivities of the output y with respect to the input x. Please note that I am interested in finding the sensitivity of the output with respect to the input rather than finding the sensitivity of the loss function with respect to the input. Is it possible to do that? If yes, any guidance would be much appreciated.
Also, one clarification, you normally don’t compute the sensitivity of the loss function with respect to the input, but rather, to the model’s parameters. Hope, this helps