A SimSiam-based Generalized Model Training Technique for Classification of ECG from Heterogeneous Devices.

In this study, contrastive learning-based SimSiam network were applied to utilize data from different acquisition environments and devices. The contrastive learning tends to make the data of the same label be located at a close distance in the feature space. The network was trained using a small amount of data from a wearable ECG device together with a large amount of resting ECGs gathered from hospitals. The experiment shows that even the small amount of data from heterogeneous device can effectively improve the ECG classification performance of related devices.