Spoken digit recognition using optoelectronic reservoir computing based on semiconductor laser with optical feedback

Reservoir Computing (RC) is an Artificial Neural Network (ANNs) that simulates human brain behavior and thinking. As a new type of ANNs, RC does not require training for its reservoir layer, only for the output weights. This simple training method enables RC to be implemented in a physical way. This article successfully builds a time-delayed optical feedback RC system based on semiconductor lasers using the nonlinear characteristics of optics. The system adopts the idea of time division multiplexing, uses the delay loop of optical feedback to construct a reservoir, and sets up a large number of virtual nodes in the feedback loop to replace traditional reservoir nodes. Finally, we successfully implemented spoken digit recognition tasks through this RC system.