English

$\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs

Computer Vision and Pattern Recognition 2024-09-13 v2 Machine Learning

Abstract

Learning good representations involves capturing the diverse ways in which data samples relate. Contrastive loss - an objective matching related samples - underlies methods from self-supervised to multimodal learning. Contrastive losses, however, can be viewed more broadly as modifying a similarity graph to indicate how samples should relate in the embedding space. This view reveals a shortcoming in contrastive learning: the similarity graph is binary, as only one sample is the related positive sample. Crucially, similarities \textit{across} samples are ignored. Based on this observation, we revise the standard contrastive loss to explicitly encode how a sample relates to others. We experiment with this new objective, called X\mathbb{X}-Sample Contrastive, to train vision models based on similarities in class or text caption descriptions. Our study spans three scales: ImageNet-1k with 1 million, CC3M with 3 million, and CC12M with 12 million samples. The representations learned via our objective outperform both contrastive self-supervised and vision-language models trained on the same data across a range of tasks. When training on CC12M, we outperform CLIP by 0.6%0.6\% on both ImageNet and ImageNet Real. Our objective appears to work particularly well in lower-data regimes, with gains over CLIP of 16.8%16.8\% on ImageNet and 18.1%18.1\% on ImageNet Real when training with CC3M. Finally, our objective seems to encourage the model to learn representations that separate objects from their attributes and backgrounds, with gains of 3.33.3-5.65.6\% over CLIP on ImageNet9. We hope the proposed solution takes a small step towards developing richer learning objectives for understanding sample relations in foundation models.

Keywords

Cite

@article{arxiv.2407.18134,
  title  = {$\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs},
  author = {Vlad Sobal and Mark Ibrahim and Randall Balestriero and Vivien Cabannes and Diane Bouchacourt and Pietro Astolfi and Kyunghyun Cho and Yann LeCun},
  journal= {arXiv preprint arXiv:2407.18134},
  year   = {2024}
}
R2 v1 2026-06-28T17:53:39.615Z