Dynamic Joint Domain Adaptation Network for Motor Imagery Classification

Electroencephalogram (EEG) has been widely used in brain computer interface (BCI) due to its convenience and reliability. The EEG-based BCI applications are majorly limited by the time-consuming calibration procedure for discriminative feature representation classification. Existing EEG classification methods either heavily depend on handcrafted features or require adequate annotated samples at each session calibration. To address these issues, we propose a novel dynamic joint domain adaptation network based adversarial learning strategy learn domain-invariant representation, thus improve performance target leveraging useful information from source session. Specifically, explore global discriminator align marginal distribution across domains, local reduce conditional discrepancy between sub-domains via conditioning deep as well predicted labels classifier. In addition, further investigate factor adaptively estimate relative importance of alignment distributions. evaluate efficacy our method, extensive experiments conducted two public datasets, namely, Datasets IIa IIb Competition IV. experimental results demonstrate that proposed method achieves superior compared with state-of-the-art methods.