Learning with nonlinear optical waves

Davide Pierangeli


We propose photonic neural networks embedding nonlinear waves as a computing reservoir. We demonstrate learning of large datasets and Boolean logic, discussing the existence of a critical nonlinearity for learning transition. The scheme is experimentally demonstrated using spatial dispersive shock waves for performing classification and regression tasks. The nonlinear optical device is easy-to-train and reaches accuracies comparable to digital reservoir machines.

Related paper: https://arxiv.org/abs/1912.07044