Diagram of the IAE architecture for the simple case with n_A = 2 anchor points 𝒜₁ and 𝒜₂, shown in blue and orange. In black, s_i is an example member of the training set. The encoder E learns how to reduce the dimensions of s_i (n_E energy channels) into two parameters, {θ₁ ,θ₂}, such that in the latent space, these correspond to the weight of each encoded anchor point for E(s_i) in order to be linearly interpolated between them. The decoder D learns how to return ŝ_i close to the input s_i from {θ₁ ,θ₂}.