Skip to content

Conversation

@mcourteaux
Copy link
Contributor

See discussion in #7681.

This patch simplifies the regularization. The most important takeaway is in the comment of the changed code, and a thought I had: #7681 (comment) :

The only thing I'm not sure about is that I changed it to regularize towards the identity function, instead of "towards the average gain in the cell". The problem I have with "the average gain" is that you also need samples to compute the average gain, which is exactly the problem being solved.

@jiawen @abadams I don't know if you want to validate the result by running the experiment setup of your paper? I guess that will be a more formal test.

@mcourteaux
Copy link
Contributor Author

@steven-johnson steven-johnson requested review from abadams and jiawen July 17, 2023 17:26
@abadams
Copy link
Member

abadams commented Jul 17, 2023

Your argument about the original regularization makes sense to me. I'd appreciate @jiawen 's take on it though.

@steven-johnson
Copy link
Contributor

ASAN failures are unrelated

@steven-johnson
Copy link
Contributor

Ready to merge?

@steven-johnson steven-johnson merged commit ce16f91 into halide:main Jul 24, 2023
@mcourteaux
Copy link
Contributor Author

🥳

ardier pushed a commit to ardier/Halide-mutation that referenced this pull request Mar 3, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants