SemPPL: Predicting pseudo-labels for better contrastive representations
Abstract
Learning from large amounts of unsupervised data and a small amount of supervision is an important open problem in computer vision. We propose a new semi-supervised learning method, Semantic Positives via Pseudo-Labels (SemPPL), that combines labelled and unlabelled data to learn informative representations. Our method extends self-supervised contrastive learning -- where representations are shaped by distinguishing whether two samples represent the same underlying datum (positives) or not (negatives) -- with a novel approach to selecting positives. To enrich the set of positives, we leverage the few existing ground-truth labels to predict the missing ones through a -nearest neighbours classifier by using the learned embeddings of the labelled data. We thus extend the set of positives with datapoints having the same pseudo-label and call these semantic positives. We jointly learn the representation and predict bootstrapped pseudo-labels. This creates a reinforcing cycle. Strong initial representations enable better pseudo-label predictions which then improve the selection of semantic positives and lead to even better representations. SemPPL outperforms competing semi-supervised methods setting new state-of-the-art performance of and top- accuracy when using a ResNet- and training on and of labels on ImageNet, respectively. Furthermore, when using selective kernels, SemPPL significantly outperforms previous state-of-the-art achieving and top- accuracy on ImageNet with and labels, respectively, which improves absolute and over previous work. SemPPL also exhibits state-of-the-art performance over larger ResNet models as well as strong robustness, out-of-distribution and transfer performance. We release the checkpoints and the evaluation code at https://github.com/deepmind/semppl .
Cite
@article{arxiv.2301.05158,
title = {SemPPL: Predicting pseudo-labels for better contrastive representations},
author = {Matko Bošnjak and Pierre H. Richemond and Nenad Tomasev and Florian Strub and Jacob C. Walker and Felix Hill and Lars Holger Buesing and Razvan Pascanu and Charles Blundell and Jovana Mitrovic},
journal= {arXiv preprint arXiv:2301.05158},
year = {2024}
}
Comments
Published as a conference paper at ICLR 2023. For checkpoints and source code see https://github.com/google-deepmind/semppl