Fig. 4.

The architecture of the proposed deep learning framework for joint batch effect removal and classification. The source batch ${\mathbf{X}}_1$ and the target batch ${\mathbf{X}}_2$ are processed through the same calibrator $\mathit{C}$, such that both batches are compactly distributed in the latent space. The source batch supervises the training of the discriminator $\mathit{D}$, which then predicts the labels for the target batch in testing. Two reconstructors, ${\mathit{R}}_1$ and ${\mathit{R}}_2$, are used to ensure that the input data can be fully recovered from latent encoding