Using forward hooks to modify the output of the neural network

1

I have a simple neural network and I want to modify the value of the output layer for a given input ‘I’ to the input layer.

Pytorch documentation has this tutorial which says -

"
We’ve inspected the weights and the gradients. But how about inspecting / modifying the output and grad_output of a layer?

We introduce hooks for this purpose.
"
The tutorial just says this can be done but does not show how to actually do it. Can someone please provide some code that shows how to change/set the outputs?

1 Like
2

http://pytorch.org/docs/master/nn.html#torch.nn.Module.register_forward_hook

3

tl;tr: torch >=v1.2.0

Hey @dakshanand,
I just ran into the same issue. In case anybody else is wondering: Update torch to >=1.2.0. Before this version, hooks were read-only, now they can modify the output.
Best