Optoelectronic Reservoir Computing Using a Mixed Digital-Analog Hardware Implementation

Abstract

Optoelectronic systems have proven to be an attractive platform for the realization of hardware-based Reservoir Computing (RC). These unconventional computers can perform nonlinear prediction of chaotic timeseries and generate arbitrary input/output functions, after training. One of the main advantages of a delay-based optoelectronic reservoir is that it only requires a single nonlinear hardware node and a feedback loop. To implement the RC scheme experimentally, we use photonic and electronic components and an FPGA to drive the system. The resulting optoelectronic RC is a hybrid analog and digital system with great versatility. We show that this set-up can perform a chaotic timeseries prediction task, in which the output is computed online, with low prediction errors. Ultimately, this testbed system will allow to test several strategies to improve the prediction performance, including the addition of multiple delays and the tailoring of the input mask for noise mitigation.