Time Series are Images: Vision Transformer for Irregularly Sampled Time Series

Zekun Li, Xifeng Yan

Keywords: irregularly sampled time series, vision transformer

TL;DR: We propose an approach that transforms time series data into line graph images and utilizes vision transformer to perform time series classification task.

Abstract: Irregularly sampled time series are often observed in medical applications. Highly customized models have been developed to tackle the irregularity. In this work, we propose a simple yet effective approach that transforms irregularly sampled time series into line graph images and adapts vision transformers to perform time series classification in a way similar to image classification. Our approach simplifies the model design without assuming prior knowledge. Despite its simplicity, we show that it is able to outperform state-of-the-art specialized algorithms on several popular healthcare and human activity datasets, especially in the challenging leave-sensors-out setting where a subset of variables are masked during testing. We hope this work could provide beneficial insight into leveraging fast-evolving computer vision techniques in the time series analysis domain.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Applications (eg, speech processing, computer vision, NLP)