This is an example of the inverse-design algorithm, based on the gradient ascent/descent. Here it is used to optimise phase profiles (masks) of the multi-plane light conversion system in order to achieve efficient mode sorting/transforming. It is possible to adjust $\alpha$, $\beta$, $\gamma$ parameters of the specifically tailored objective function to improve the parameter of a particular interest, such as average cross-talk, average efficiency or average localised fidelity, for the specific application of the MPLC system.

This example code performs sorting of up to 55 random speckle 'modes' into spatially separated output channels. The main file 'Gradient ascent - speckle mode sorter.ipynb' contains introductory remarks and all the detailed comments along with the code. 'utils.py' contains all the additional functions that are called from the main code, and 'Speckle_sorter_bases.mat' contains pre-generated input-output bases for the sorter.