Fig. 2

MultiMAP applied to synthetic data. a Data sampled from the 3D Swiss Roll (X¹) and a 2D rectangle (X²). b Shared embedding of both datasets produced by MultiMAP. Color indicates position along the manifold (a,b). c Left (X¹) and right (X²) halves of MNIST handwritten digit images with a 2 pixel wide shared region. Gaussian noise is added to the left half. UMAP projections of each half and the shared region. d Shared embedding of both MNIST halves (including Gaussian noise introduced for the left half) produced by MultiMAP. Each color is a different handwritten digit (0–9 as shown in the key). This illustrates that MultiMAP leverages both shared and unshared features to integrate multimodal datasets