We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? # to your account
https://www.autodiff.org/Docs/euroad/Trond_Steihaug.pdf describes a bit how it would be done via jvps and hvps, which would mix in AD tricks.
The text was updated successfully, but these errors were encountered:
Revisiting this seems like Halley should never compute a Hessian in the first place.
It should be done with a HVP. The linear solve can also be done with a krylov method but that is a separate point.
Sorry, something went wrong.
Successfully merging a pull request may close this issue.
https://www.autodiff.org/Docs/euroad/Trond_Steihaug.pdf describes a bit how it would be done via jvps and hvps, which would mix in AD tricks.
The text was updated successfully, but these errors were encountered: