Rapid Configuration of Optical Processor Based on Reconfigurable Multiport Interferometer

We demonstrate an optimization method for universal multiport interferometers to perform rapid configuration reconstruction. This method uses deep learning to model the relationship between the transmission characteristics and the configuration parameters of the component of the reconfigurable multiport interferometers, making it resistant to manufacturing flaws of the interferometer. Our demonstration provides a potential way to implement fast and accurate reconfiguration functions for optical processors.