CoMatch: Semi-supervised Learning with Contrastive Graph Regularization
By Junnan Li, Caiming Xiong, Steven C. H. Hoi. IEEE International Conference on Computer Vision, 2020
352 citations recorded by the source corpus.
Semi-supervised learning has been an effective paradigm for leveraging unlabeled data to reduce the reliance on labeled data. We propose CoMatch, a new semi-supervised learning method that unifies dominant approaches and addresses their limitations. CoMatch jointly learns two representations of the training data, their class probabilities and low-dimensional embeddings. The two representations interact with each other to jointly evolve. The embeddings impose a smoothness constraint on the class probabilities to improve the pseudo-labels, whereas the pseudo-labels regularize the structure of the embeddings through graph-based contrastive learning. CoMatch achieves state-of-the-art performance on multiple datasets. It achieves substantial accuracy improvements on the label-scarce CIFAR-10 and STL-10. On ImageNet with 1% labels, CoMatch achieves a top-1 accuracy of 66.0%, outperforming FixMatch [32] by 12.6%. Furthermore, CoMatch achieves better representation learning performance on downstream tasks, outperforming both supervised learning and self-supervised learning. Code and pre-trained models are available at https://github.com/salesforce/CoMatch/.
Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition
View Paper: https://www.semanticscholar.org/paper/669cbe118d868941172dbebfcaca9a479301d23b
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Taghi M. Khoshgoftaar
- Wayne Zhang
- Avital Oliver
- Geoff French
- Connor Shorten
- Borko Furht
- Lei Qi
- Lihe Yang
In Computer Vision and Pattern Recognition
- Wayne Zhang
- Avital Oliver
- Taghi M. Khoshgoftaar
- Tim Salimans
- Connor Shorten
- Yinghuan Shi
- Lei Qi
- Lihe Yang
Citations in the corpus
- Milking CowMask for Semi-Supervised Image Classification (2020, methodology): “CoMatch (Li et al., 2020) combines consistency regularization with self-supervised contrastive learning.”
- Revisiting Weak-to-Strong Consistency in Semi-Supervised Semantic Segmentation (2022, methodology): “Two main branches of methodology are proposed to tackle the issue, namely entropy minimization [22, 37, 51, 53, 64, 80] and consistency regularization [5,6,20,30,36,38,48,54,58,63].”
- Text Data Augmentation for Deep Learning (2021, background): “Li et al. [49] similarly construct an embedding graph to enforce consistency between predictions of strong and weakly augmented data.”