SimCLR: L2 Normalization of Projections
Beginner
The problem, from first principles
Before the contrastive loss, SimCLR's projection head output is normalized to unit length. That turns the dot product into a cosine similarity, so the loss depends only on direction.
From theory to code
Implement l2_normalize(z), returning the vector scaled to unit length.
Constraints
- A zero vector is returned unchanged.
Hints
Hint 1
Divide by the norm, guarding the zero case.
Keywords
research-papersunsupervisedcontrastiveself-supervised
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.