New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Optimize `torch.func` full/diag GGN/Fisher-MC #149

Open

wiseodd opened this issue Mar 9, 2024 · 0 comments

Labels

Milestone

Collaborator

wiseodd commented Mar 9, 2024

Currently, we compute the Jacobians explicitly. We can improve this by using VJPs.

Reference for full GGN: https://github.com/f-dangel/curvlinops/blob/5852711aedf2728bc609fabfa95eac00da1beb63/curvlinops/examples/functorch.py#L72-L138

Not a high priority since KFAC is usually used and (diag/full) EF implementations are already efficient.

wiseodd added the enhancement label

runame added this to the 0.1 milestone

wiseodd removed this from the 0.2 milestone

wiseodd added this to the 0.3 milestone

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment