Reinforcement learning in a large-scale photonic recurrent neural network

Abstract

Photonic Neural Network implementations have been gaining considerable\nattention as a potentially disruptive future technology. Demonstrating learning\nin large scale neural networks is essential to establish photonic machine\nlearning substrates as viable information processing systems. Realizing\nphotonic Neural Networks with numerous nonlinear nodes in a fully parallel and\nefficient learning hardware was lacking so far. We demonstrate a network of up\nto 2500 diffractively coupled photonic nodes, forming a large scale Recurrent\nNeural Network. Using a Digital Micro Mirror Device, we realize reinforcement\nlearning. Our scheme is fully parallel, and the passive weights maximize energy\nefficiency and bandwidth. The computational output efficiently converges and we\nachieve very good performance.\n