Deep Semi-Supervised Learning for Time Series Classification

TL;DR

This paper explores adapting deep semi-supervised learning models from image recognition to time series classification, demonstrating significant performance improvements especially with limited labeled data.

Contribution

It introduces necessary model adaptations and evaluates the transferability of state-of-the-art semi-supervised models to time series data, filling a research gap.

Findings

Semi-supervised models outperform supervised and self-supervised methods.

Significant gains with very few labeled samples.

Extensive evaluation on large public datasets.

Abstract

While Semi-supervised learning has gained much attention in computer vision on image data, yet limited research exists on its applicability in the time series domain. In this work, we investigate the transferability of state-of-the-art deep semi-supervised models from image to time series classification. We discuss the necessary model adaptations, in particular an appropriate model backbone architecture and the use of tailored data augmentation strategies. Based on these adaptations, we explore the potential of deep semi-supervised learning in the context of time series classification by evaluating our methods on large public time series classification problems with varying amounts of labelled samples. We perform extensive comparisons under a decidedly realistic and appropriate evaluation scheme with a unified reimplementation of all algorithms considered, which is yet lacking in the field. We find that these transferred semi-supervised models show significant performance gains over strong supervised, semi-supervised and self-supervised alternatives, especially for scenarios with very few labelled samples.