Hi friends, I’m trying to figure out the way to calculate the norm of Jacobian between layers of a deep CNN, e.g. the jacobian between the 15th and 18th layer output of VGG19. The problem is these Jacobians are very high dimensional, normal methods like autograd.jacobian will most certainly throw an OOM error. Since I don’t need the exact full Jacobian but it’s Frobenius norm, what is the best way to make it work?
Any help would be much appreciated!