Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results

TL;DR

This paper introduces the Mean Teacher method, which uses weight averaging for consistency targets in semi-supervised learning, leading to improved accuracy and efficiency over previous methods like Temporal Ensembling.

Contribution

The paper proposes the Mean Teacher approach, a novel semi-supervised learning technique that averages model weights instead of predictions, enhancing performance and reducing label requirements.

Findings

Achieved 4.35% error on SVHN with 250 labels.

Improved CIFAR-10 accuracy from 10.55% to 6.28%.

Reduced ImageNet error from 35.24% to 9.11% with fewer labels.

Abstract

The recently proposed Temporal Ensembling has achieved state-of-the-art results in several semi-supervised learning benchmarks. It maintains an exponential moving average of label predictions on each training example, and penalizes predictions that are inconsistent with this target. However, because the targets change only once per epoch, Temporal Ensembling becomes unwieldy when learning large datasets. To overcome this problem, we propose Mean Teacher, a method that averages model weights instead of label predictions. As an additional benefit, Mean Teacher improves test accuracy and enables training with fewer labels than Temporal Ensembling. Without changing the network architecture, Mean Teacher achieves an error rate of 4.35% on SVHN with 250 labels, outperforming Temporal Ensembling trained with 1000 labels. We also show that a good network architecture is crucial to performance. Combining Mean Teacher and Residual Networks, we improve the state of the art on CIFAR-10 with 4000 labels from 10.55% to 6.28%, and on ImageNet 2012 with 10% of the labels from 35.24% to 9.11%.