Fig. 3From: Similarity-assisted variational autoencoder for nonlinear dimension reduction with application to single-cell RNA sequencing dataEmbeddings from MNIST dataset. a The UMAP in general separated the groups accurately except for two clusters that needed to be further separated. b The DGG formed almost perfect groups in the latent space, yet the gap between the two clusters (digits 4 and 9) was still slim. c The saVAE showed perfect groups in the latent space including digits 4 and 9Back to article page