Meta Pseudo Labels
This addresses the problem of improving semi-supervised learning efficiency and accuracy for computer vision tasks, representing a strong incremental advance over existing pseudo-labeling techniques.
The paper tackles semi-supervised learning by introducing Meta Pseudo Labels, which adapts a teacher network based on student feedback to generate better pseudo labels, achieving a new state-of-the-art top-1 accuracy of 90.2% on ImageNet, a 1.6% improvement over previous methods.
We present Meta Pseudo Labels, a semi-supervised learning method that achieves a new state-of-the-art top-1 accuracy of 90.2% on ImageNet, which is 1.6% better than the existing state-of-the-art. Like Pseudo Labels, Meta Pseudo Labels has a teacher network to generate pseudo labels on unlabeled data to teach a student network. However, unlike Pseudo Labels where the teacher is fixed, the teacher in Meta Pseudo Labels is constantly adapted by the feedback of the student's performance on the labeled dataset. As a result, the teacher generates better pseudo labels to teach the student. Our code will be available at https://github.com/google-research/google-research/tree/master/meta_pseudo_labels.