Improve numerical stability of batch normalization on CPUs #3113
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
…` fail by introducing non-zero mean
…_normalize_conv` by computing variances using Welford's algorithm
|
Note that the CI error is addressed separately in PR #3112. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Problem: Calculating mean and variance in the CPU implementation of batch normalization is prone to losing some numerical precision (when the data is not zero mean, for example).
Solution: Use Welford's algorithm that is numerically more stable.
Note that the CUDA implementation should already be doing this (or something similar at least), so this change should make the CPU implementation better match the CUDA one.