Replies: 1 comment 1 reply
-
Interesting, are you interested in contributing? |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Interesting, are you interested in contributing? |
Beta Was this translation helpful? Give feedback.
-
🚀 Feature
This paper proposes a new normalization technique. Where the nodes in a group are normalized independently and the group assignment is learnable.
It also proposes two metrics to measure over smoothing.
Motivation
The results look good in comparison to no norm(NN), pair-norm(PN) and batch norm(BN). But the paper doesn't compare results of other normalizations techniques.
Further the metrics for over smoothing might be useful to analyze gcn's.
Additional Context
torch-geometric has a few normalization layers, so it might not be too useful to add this. But i feel the over smoothing metrics could be a good addition. Do let me know your thoughts on this.
Beta Was this translation helpful? Give feedback.
All reactions