Geometric Contrastive Learning

Yeskendir Koishekenov, Sharvaree P. Vadgama, Riccardo Valperga, Erik J. Bekkers

ICCVW 2023 pp. 206-215

doi:10.1109/ICCVW60793.2023.00028 /iccvw/2023/koishekenov2023iccvw-geometric/

Abstract

Contrastive learning has been a long-standing research area due to its versatility and importance in learning representations. Recent works have shown improved results if the learned representations are constrained to be on a hypersphere. However, this prior geometric constraint is not fully utilized during training. In this work, we propose making use of geodesic distances on the hypersphere to learn contrasts between representations. Through empirical results, we show that this contrastive learning approach improves downstream tasks across different contrastive learning frameworks. We show that having geometric inductive priors perform even better in contrastive learning if used along with other correct geometric information.

ICCVW Semantic Scholar

Cite

Text

Koishekenov et al. "Geometric Contrastive Learning." IEEE/CVF International Conference on Computer Vision Workshops, 2023. doi:10.1109/ICCVW60793.2023.00028

Markdown

[Koishekenov et al. "Geometric Contrastive Learning." IEEE/CVF International Conference on Computer Vision Workshops, 2023.](https://mlanthology.org/iccvw/2023/koishekenov2023iccvw-geometric/) doi:10.1109/ICCVW60793.2023.00028

BibTeX

@inproceedings{koishekenov2023iccvw-geometric,
  title     = {{Geometric Contrastive Learning}},
  author    = {Koishekenov, Yeskendir and Vadgama, Sharvaree P. and Valperga, Riccardo and Bekkers, Erik J.},
  booktitle = {IEEE/CVF International Conference on Computer Vision Workshops},
  year      = {2023},
  pages     = {206-215},
  doi       = {10.1109/ICCVW60793.2023.00028},
  url       = {https://mlanthology.org/iccvw/2023/koishekenov2023iccvw-geometric/}
}