Starting backward from any convolutional layer

autograd
1

Hi
The backward function starts the gradient operation from the output respect to the input.
How can I compute gradient from specific convolution layer in specific layer, instead from output, respect to input?Is there any solution in Pytorch or should I create my own backward function?

Thanks you in advance.

1 Like
2

I have encountered the same problem. Have you solved the problem? Can you teach me some experience