Recurrent Machine Learning and Computing with Nonlinear Optical Waves

We demonstrate that optical time-dynamics are equivalent to a recurrent neural network and that they can be trained for high-performance on complex classification tasks, paving the way for passive analog machine learning processors.