 In the EEG-based motor imagery classification task, the device and subject problems can lead to a shift in the time-related data distribution, which can negatively impact the classification accuracy. This paper proposes a multi-subdomain adaptation network, MSDAN, which addresses this issue by simultaneously minimizing the adaptation and classification losses in multiple subdomains. The proposed method was tested on the BCI Competition 3 IVA dataset and demonstrated improved performance over existing methods. This article was authored by each N, Rui Yang, Ming Jia Huang, and others.