Tunable Nonlinear Activation Functions for Optical Neural Networks

We introduce an electro-optic hardware platform for realizing optical nonlinearities and demonstrate that, as a nonlinear activation function, it can substantially improve the classification performance of optical neural networks.