Boosting the Performance of Semi-Supervised Learning with Unsupervised Clustering

TL;DR

This paper introduces a semi-supervised learning method that intermittently ignores labels during training, using joint clustering and classification tasks with rotation-based self-supervision to improve performance on small labeled datasets.

Contribution

The proposed approach combines clustering with classification and rotation self-supervision, significantly enhancing semi-supervised learning results with minimal labels.

Findings

Achieved 92.6% accuracy on CIFAR-10 with only 4 labels per class

Improved results in extreme label scarcity scenarios (1-3 labels per class)

Features learned are more meaningful for data separation

Abstract

Recently, Semi-Supervised Learning (SSL) has shown much promise in leveraging unlabeled data while being provided with very few labels. In this paper, we show that ignoring the labels altogether for whole epochs intermittently during training can significantly improve performance in the small sample regime. More specifically, we propose to train a network on two tasks jointly. The primary classification task is exposed to both the unlabeled and the scarcely annotated data, whereas the secondary task seeks to cluster the data without any labels. As opposed to hand-crafted pretext tasks frequently used in self-supervision, our clustering phase utilizes the same classification network and head in an attempt to relax the primary task and propagate the information from the labels without overfitting them. On top of that, the self-supervised technique of classifying image rotations is incorporated during the unsupervised learning phase to stabilize training. We demonstrate our method's efficacy in boosting several state-of-the-art SSL algorithms, significantly improving their results and reducing running time in various standard semi-supervised benchmarks, including 92.6% accuracy on CIFAR-10 and 96.9% on SVHN, using only 4 labels per class in each task. We also notably improve the results in the extreme cases of 1,2 and 3 labels per class, and show that features learned by our model are more meaningful for separating the data.