Experimental Demonstration of a Photonic Extreme Learning Machine with an Array of Microresonators

The training process of a feed-forward neural network is typically power- and time-consuming because it requires optimization of the output response through a gradient descent algorithm. An alternative to these approaches is the extreme learning machine, which is a feed-forward neural network composed of a single hidden layer in which training occurs only in the readout. Here, we propose and experimentally validate an extreme learning machine architecture based on an array of 18 silicon microresonators. We provide a proof-of-concept demonstration of the network by solving the nonlinear logic operation XOR, the iris flower classification, and banknote authentication.