Shaped Supercontinuum as a Neural Network Computing Element

Optical computing is being developed as a supplement or even potential replacement for semiconductor electronic computing. Neural networks computing may be particularly compatible with optics as it involves performing many parallel, nonlinear operations on a set of inputs to calculate a set of outputs [1]. This is like supercontinuum generation, where almost any subtle adjustment of the seed pulse can significantly change the pulse spectral or temporal shape. Here, we incorporate a shaped supercontinuum source into a neural network, performing standard classification and autoencoding tasks by measuring shaped supercontinuum spectra.