Variational autoencoder used to create a K = 5 dimensional latent space given the 2 × 55 continuous BP and RP spectra points. Each coloured rectangle represents a layer of the ANN, with the size of the output vector given at the top. The encoder starts with the (standardised) observations denoted in the leftmost white rectangle, and ends with the K = 5-dimensional latent space denoted in green, representing the position μ and its (spherical) uncertainty σ² of the observed data in the latent space. The decoder starts by sampling a K=5-dimensional value z from a Gaussian distribution with mean μ and the diagonal covariance matrix σ², and ends with the reconstructed 2 × 55 XP spectra values. Some layers use ‘layer normalisation’, which is denoted by LN. The activation function of the neurons in a particular layer is noted at the bottom: GELU, LIN (linear), or ReLU.