In brief
What the paper does.
We propose VarCon, reformulating supervised contrastive learning as variational inference over latent class variables. Achieves SOTA among contrastive learning methods: 79.36% Top-1 on ImageNet-1K with ResNet-50, with superior embedding space organization and few-shot robustness.
Cite
Cite the paper.
@inproceedings{wang2025varcon,
title={Variational Supervised Contrastive Learning},
author={Wang, Ziwen and Fan, Jiajun and Nguyen, Thao and Ji, Heng and Liu, Ge},
booktitle={Advances in Neural Information Processing Systems},
year={2025},
url={https://openreview.net/forum?id=uOOlHOq500}
}