Fig. 1.

Visualisation of the MS-LGS as outlined in algorithm 1. Each iteration is performed on their respective discretisation space, where the gradient $\nabla {𝒟}_i:=\nabla {𝒟}_i\left(f_i;g\right)$ is computed and the update is performed by the network Λ_θi. After each update an up-sampling by τ_i to the next finer space is performed until the final resolution S_N is achieved.