Stochastic Optimization of Nonlinear Nanophotonic Media for Artificial Neural Inference

We show that optical waves passing through a nanophotonic medium can perform artificial neural computing. Complex information, such as an image, is encoded in the wave front of an input light. The medium continuously transforms the wave front to realize highly sophisticated computing tasks such as image recognition. At the output, the optical energy is concentrated to well defined locations, which for example can be interpreted as the identity of the object in the image. These computing media can be as small as tens of wavelengths and offer ultra-high computing density. They exploit sub-wavelength linear and nonlinear scatterers to realize complex input-output mapping far beyond the capabilities of traditional nanophotonic devices. All structural degrees of freedom can be used as training weights, forming a vast parameter space with strong expressive power.