FlexMatch: Boosting Semi-Supervised Learning with Curriculum Pseudo Labeling
This addresses a bottleneck in semi-supervised learning for researchers and practitioners by improving performance and efficiency, especially in data-scarce scenarios, though it is incremental as it builds on existing methods like FixMatch.
The paper tackles the issue of using a fixed threshold for pseudo-label selection in semi-supervised learning, which fails to account for varying class learning difficulties, by proposing Curriculum Pseudo Labeling (CPL) to dynamically adjust thresholds based on learning status. It results in FlexMatch, which reduces error rates by 13.96% on CIFAR-100 and 18.96% on STL-10 with only 4 labels per class and speeds up convergence by using 1/5 of the training time compared to FixMatch.
The recently proposed FixMatch achieved state-of-the-art results on most semi-supervised learning (SSL) benchmarks. However, like other modern SSL algorithms, FixMatch uses a pre-defined constant threshold for all classes to select unlabeled data that contribute to the training, thus failing to consider different learning status and learning difficulties of different classes. To address this issue, we propose Curriculum Pseudo Labeling (CPL), a curriculum learning approach to leverage unlabeled data according to the model's learning status. The core of CPL is to flexibly adjust thresholds for different classes at each time step to let pass informative unlabeled data and their pseudo labels. CPL does not introduce additional parameters or computations (forward or backward propagation). We apply CPL to FixMatch and call our improved algorithm FlexMatch. FlexMatch achieves state-of-the-art performance on a variety of SSL benchmarks, with especially strong performances when the labeled data are extremely limited or when the task is challenging. For example, FlexMatch achieves 13.96% and 18.96% error rate reduction over FixMatch on CIFAR-100 and STL-10 datasets respectively, when there are only 4 labels per class. CPL also significantly boosts the convergence speed, e.g., FlexMatch can use only 1/5 training time of FixMatch to achieve even better performance. Furthermore, we show that CPL can be easily adapted to other SSL algorithms and remarkably improve their performances. We open-source our code at https://github.com/TorchSSL/TorchSSL.