Contrastive Learning with Hard Negative Samples

We consider the question: how can you sample good negative examples for contrastive learning? We argue that, as with metric learning, learning contrastive representations benefits from hard negative samples (i.e., points that are difficult to distinguish from an anchor point). The key challenge toward using hard negatives is that contrastive methods must remain unsupervised, making it infeasible to adopt existing negative sampling strategies that use label information. In response, we develop a new class of unsupervised methods for selecting hard negative samples where the user can control the amount of hardness. A limiting case of this sampling results in a representation that tightly clusters each class, and pushes different classes as far apart as possible. The proposed method improves downstream performance across multiple modalities, requires only few additional lines of code to implement, and introduces no computational overhead.

Contrastive Learning with Hard Negative Samples [paper]
Joshua Robinson, Ching-Yao Chuang, Suvrit Sra, and Stefanie Jegelka

Citation

If you find this repo useful for your research, please consider citing the paper

@article{robinson2020hard,
  title={Contrastive Learning with Hard Negative Samples},
  author={Robinson, Joshua and Chuang, Ching-Yao, and Sra, Suvrit and Jegelka, Stefanie},
  journal={International Conference on Learning Representations},
  year={2021}
}

For any questions, please contact Josh Robinson ([email protected]).

Acknowledgements

Part of this code is inspired by leftthomas/SimCLR, by chingyaoc/DCL, and by fanyun-sun/InfoGraph.