Scale-then-Privatize Adaptive Optimization (#88) Split PR 3 #109
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…izers (Ganesh et al., 2025)
debanganghosh08
changed the title
debanganghosh08
changed the title
debanganghosh08
changed the title
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This adds an example of Scale-then-Privatize adaptive optimization as detailed in Research paper, "On Design Principles for Private Adaptive Optimizers".
The Problem Addressed: Standard private adaptive optimizers add spherical noise to gradients before preconditioning, which creates a "noise floor" that destroys the benefits of adaptivity.
✔️ The Solution: I have implemented Algorithm 8 from the paper. Using the pre_clipping_transform hook in jax_privacy, gradients are rescaled into a "spherical" space before the sensitivity clip. The noised aggregate is then un-scaled, ensuring the effective noise distribution is aligned with the gradient geometry rather than being uniform.
✔️ Correctness Verification: I verified the implementation by monitoring the running_variance PyTree. The logs confirm that the preconditioner successfully departs from its initial state (1.0) to adapt to the data geometry, even in the presence of noise.