How do I compute the gradient of a convolutional neural network with respect to a given input image?

I would like to do a gradient ascent of a CNN-based classification network with respect to the input image in order to maximize the score of a certain class. Is there any functionality to do so? The layer template for defining new deep learning layers has a "backward" function; it would be great to be able to call such a function for the built-in layers, too.

Answer 1

[Sai Bhargav Avula]

Hi,

The dlgradient can be used for computing the gradient.