Sketch of the NF approach. The physical quantities are described by a neural network that takes the coordinates as input and outputs the physical quantities (temperature T, magnetic field B, velocity v, etc.) at that point (x, y). The output of the network is then used to compute the observables from the model using a radiative transfer (RT) module. In this work, the output of the network is the magnetic field vector and the RT module is the weak-field approximation. The synthetic observables are compared with the observations and the error is back-propagated through the network to obtain the optimal solution.