Full network architecture used in the paper. The network is composed of inception blocks (purple, rounded rectangles), show in Figure B.2, max pooling layers (turquoise, trapeziums), linear layers (red, rectangles), a flatten layer (yellow, rectangle), and an information-ordered bottleneck layer (yellow, rounded rectangle). The output dimensions of each layer is represented underneath the name of the layer, the arrows represent the direction of data flow and if there is a ×2 next to the arrow, then the layer is repeated twice. The input is the total mass maps and optionally X-ray maps of galaxy clusters and the latent space is where we calculate the loss functions from equations 1 and 3 as well as our feature analysis.