New Method Targets Representation Diversity in Self-Supervised Learning
A paper introduces Centroid Spectral Coupling, an approach for implicit alignment and representation diversity in self-supervised learning, addressing collapse prevention in non-contrastive methods.
Read the full story at arXiv stat.ML ↗
Timeline · 1 report
- 2026-11-18 05:00 · arXiv stat.ML
Centroid Spectral Coupling: Implicit Alignment and Representation Diversity in Self-Supervised Learning