Photonic Neural Networks with Random Projection Kernel Optimization

Optical computing offers faster processing, lower power use, and parallel computations. Neural networks find promise in optical computing due to dense connectivity and noise resistance, with applications making use of chaos in photonic neural networks [1]. As a prominent technique, linear optical neural networks for large datasets use random projection, reducing high-dimensional data dimensionality [2]. Optical platforms with this method are explored [3], where the challenge of frequent optical-electronic connections is circumvented with numeric designs of scattering environments, but only in the THz band [4].