All Labels Are Not Created Equal: Enhancing Semi-supervision via Label Grouping and Co-training
This work addresses a specific bottleneck in semi-supervised learning for computer vision tasks, offering improvements in accuracy and efficiency.
The paper tackles the problem of pseudo-labeling in semi-supervised learning by addressing the degradation in quality due to ignoring visual similarity among classes, and it proposes SemCo, which uses label grouping and co-training to achieve state-of-the-art performance, including a 5.6% accuracy improvement on Mini-ImageNet with 1000 labeled examples.
Pseudo-labeling is a key component in semi-supervised learning (SSL). It relies on iteratively using the model to generate artificial labels for the unlabeled data to train against. A common property among its various methods is that they only rely on the model's prediction to make labeling decisions without considering any prior knowledge about the visual similarity among the classes. In this paper, we demonstrate that this degrades the quality of pseudo-labeling as it poorly represents visually similar classes in the pool of pseudo-labeled data. We propose SemCo, a method which leverages label semantics and co-training to address this problem. We train two classifiers with two different views of the class labels: one classifier uses the one-hot view of the labels and disregards any potential similarity among the classes, while the other uses a distributed view of the labels and groups potentially similar classes together. We then co-train the two classifiers to learn based on their disagreements. We show that our method achieves state-of-the-art performance across various SSL tasks including 5.6% accuracy improvement on Mini-ImageNet dataset with 1000 labeled examples. We also show that our method requires smaller batch size and fewer training iterations to reach its best performance. We make our code available at https://github.com/islam-nassar/semco.