Journal
IEEE ACCESS
Volume 11, Issue -, Pages 118595-118605Publisher
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/ACCESS.2023.3327015
Keywords
Unsupervised representation learning; local binary pattern; global and local features; multivariate time series
Ask authors/readers for more resources
This study proposes an unsupervised representation learning model for multivariate time series by comprehensively considering the global and local information of the data. It introduces a specially designed local binary pattern method for multivariate time series to improve the representation performance, and also presents a novel unsupervised approach for learning multivariate time series representations.
Representation learning of multivariate time series is a crucial and complex task that offers valuable insights for numerous applications, including time series classification, trend analysis, and regression. Unsupervised learning approaches are often favored in practical scenarios due to the limited availability of labeled data. However, most existing studies focus more on the global information of time series and ignore the local information, especially the representation learning based on the self-attention mechanism. This affects representation performance and may lead to the failure of downstream tasks. This study proposed an unsupervised representation learning model for multivariate time series by comprehensively considering multivariate time series data's global and local information. Specifically, a specially designed local binary pattern (LBP) method for multivariate time series (multivariate LBP) is introduced to the self-attention mechanism to improve the representation performance of modeling in terms of local information. Additionally, we propose a novel unsupervised approach for learning multivariate time series representations. The experimental results demonstrate significant advantages of our model over other representation learning methods and can be well applied in various downstream tasks.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available