[feat] add CosFace paper's LMCL to MEVO (#916)

* [feat] add CosFace paper's LMCL to MEVO - added baseline algorithm to the reference kernel - added MEVO version of LMCL - added unit test to verify it is correct with respect to the reference as well as its memory usage * updated changelog Co-authored-by: Min Xu <min.xu.public@gmail.com>
facebookresearch · Jan 28, 2022 · 89e1ae5 · 89e1ae5
1 parent 8ba649e
commit 89e1ae5
Show file tree

Hide file tree

Showing 3 changed files with 214 additions and 40 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -7,6 +7,12 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [0.4.6] - TBD
 
+### Added
+- CosFace's LMCL is added to MEVO. This is a loss function that is suitable
+  for large number of prediction target classes. It added normalization,
+  class separation margins and feature vector scaling to the standard
+  output projection + cross-entropy loss. MEVO supported this with its
+  memory saving techniques so that peak GPU memory is much reduced. [#916]
 
 ## [0.4.5] - 2022-01-14