MDLR: A Multi-Task Disentangled Learning Representations for unsupervised time series domain adaptation

Yu Liu, Duantengchuan Li, Jian Wang, Bing Li, Bo Hang

Abstract: Highlights•Introduces an innovative theoretical framework for Unsupervised Time Series Domain Adaptation (UTSDA) that uses causal inference to improve domain adaptation.•Proposes MDLR, a unique method for learning disentangled representations that effectively isolate trend and seasonal domain-invariant features in time series data.•Features an innovative dual-tower architecture in the MDLR framework that facilitates simultaneous learning of both long-term and short-term time series patterns.•Demonstrates MDLR’s superior performance on UTSDA tasks, significantly outperforming current state-of-the-art methods on real-world datasets.