[non-record track] Asymmetric Squared Unit (ASQU): learning per-channel asymmetric activations#1035
Open
andrewmouldon wants to merge 5 commits intoopenai:mainfrom
Open
[non-record track] Asymmetric Squared Unit (ASQU): learning per-channel asymmetric activations#1035andrewmouldon wants to merge 5 commits intoopenai:mainfrom
andrewmouldon wants to merge 5 commits intoopenai:mainfrom
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
This PR introduces ASQU (Asymmetric Squared Unit), a per-channel activation that allows each feature dimension to learn its own asymmetric response.
Instead of using a single shared activation function across all channels, ASQU learns a per-channel scaling for the negative branch of a squared activation.
Motivation
Activation functions are typically shared across all channels, either fixed (ReLU, GELU) or globally parameterized (e.g. PReLU).
ASQU relaxes this assumption by allowing each channel to specialize its activation behavior with minimal parameter overhead.
Results (10k steps, same setup)
ASQU provides a consistent improvement over both ReLU² and fixed-slope asymmetric activations.
Notes on Evaluation Setting
ASQU introduces minimal computational overhead and is straightforward to integrate, making it well-suited for both fixed-step and time-constrained settings.
In this work, we evaluate under a fixed 10k step budget to isolate architectural effects from slight potential differences in data exposure. This allows for cleaner comparisons when studying small changes such as activation functions.
See README for further details and discussion.