Fig. 2

We use a synthetic Gaussian distributed dataset to demonstrate our method. In panel A, we generate three clusters of gaussian distributed samples. Their clusters are distinct, hence simple decision boundaries can separate them clearly. This example corresponds to the AD vs NC classification task. In panel B, we also generate three clusters which are not distinctive from one another. In fact, the E-MCI and L-MCI clusters are much less distinct than the samples in panel B. In panel C, we use OT to map the source domain samples onto the target domain. In the last panel (D), we use our proposed method adopting OT to map target samples onto the source domain by utilizing sample labels