Fix decomon normalization layers #169
Merged
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
Unit, Layer and Group normalization are not linear. For now we remove linear attribute from the decomon counterparts and do not implement yet their ibp/affine bounds, nor test them.
- layer_pos and layer_neg must not take into account the bias (already
added in forward_ibp_propagate())
=> center and add_moving_mean
default to False in BatchNormalizationKernelConstraint
=> we update the docstring for layer_neg and layer_pos in DecomonLayer to
make it explicit
- if scale = False, we set layer_neg to 0 (as gamma = None but should be
considered to 1. so only layer_pos has to be considered)
- in tests, when randomizing layer weights, we need to be careful not to
put a negative number for BatchNormalization.moving_variance
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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.
No description provided.