Enhancing Self-Training Methods

TL;DR

This paper proposes multiple enhancements to self-training methods in semi-supervised learning to reduce confirmation bias, improve pseudo-label accuracy, and extend applicability to open set unlabeled data, demonstrating consistent performance gains.

Contribution

The paper introduces novel improvements to the self-training pipeline that mitigate confirmation bias and extend its use to open set data, outperforming existing methods.

Findings

Performance gains over existing self-training methods

Effective mitigation of confirmation bias

Successful extension to open set unlabeled data

Abstract

Semi-supervised learning approaches train on small sets of labeled data along with large sets of unlabeled data. Self-training is a semi-supervised teacher-student approach that often suffers from the problem of "confirmation bias" that occurs when the student model repeatedly overfits to incorrect pseudo-labels given by the teacher model for the unlabeled data. This bias impedes improvements in pseudo-label accuracy across self-training iterations, leading to unwanted saturation in model performance after just a few iterations. In this work, we describe multiple enhancements to improve the self-training pipeline to mitigate the effect of confirmation bias. We evaluate our enhancements over multiple datasets showing performance gains over existing self-training design choices. Finally, we also study the extendability of our enhanced approach to Open Set unlabeled data (containing classes not seen in labeled data).