You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi, sorry for the late reply.
You may use a different loss for KFAC and a different one for the gradients. For the KFAC loss, you should pick the probability distribution of your model's output while the loss may be any loss.
In the training loop from the MNIST example it may look like:
Can your code be used to compute the KFAC (Fisher) matrix using a different loss than the loss we take the gradient of?
If so, how?
thanks
The text was updated successfully, but these errors were encountered: