Multi-scale self-supervised representation learning with temporal alignment for multi-rate time series modeling

Jiawei Chen, Pengyu Song, Chunhui Zhao

Abstract: Highlights•A new framework is proposed for multi-rate time series modeling under label scarcity.•Our approach can explicitly characterize multi-scale temporal dynamics.•The temporal alignment mechanism can aggregate and refine the multi-rate feature representations.•Our proposed approach achieved state-of-the-art performance on five real-world datasets.