Fig. 4: Distances in the latent space are used to suggest similar images to user-specific images of interest.
In (A) sample heterodyne two-time correlations, C2, are plotted in the Uniform Manifold Approximation and Projection (UMAP) visualization of the latent space as large X’s. Nearest neighbors (calculated by Euclidean distance in the latent space) are shown as corresponding solid points. The UMAP visualization is cropped so that test data is easier to see. In (B) sample images are displayed alongside their nearest neighbors to evaluate similarity.
