MTSSL: Meta-Thresholding Semi-Supervised Learning

Abstract

A large body of Semi-supervised Learning~(SSL) algorithms encounter the threshold $\ddot{I}„$ to select pseudo-labels. The value of $\ddot{I}„$ across different SSL algorithms can vary depending on the learning perspective, yet they may achieve similar performance. It motivates us to establish a unified theoretical framework to explain the role of $\ddot{I}„$ in SSL. We statistically explained that the unsupervised loss is affected independently by correct and incorrect pseudo-labels, while $\ddot{I}„$ adjusts their numbers to balance the corresponding error term. This inherent trade-off indicates that SSL can reach the same loss with varying $\ddot{I}„$, precise optimal values of $\ddot{I}„$ during training may be unnecessary. With this, we treat $\ddot{I}„$ as an updatable parameter and optimize it via differentiation; the new policy is named \textbf{Meta-Thresholding Semi-Supervised Learning (MTSSL)}. Extensive experiments demonstrate the superior performance of MTSSL. We observe that the accuracy curves of SSL algorithms can overlap completely even when the values of $\ddot{I}„$ differ significantly, which supports our theoretical framework and indicates that the selection of $\ddot{I}„$ can be relaxed in the future design of SSL algorithms.