Information processing using optoelectronic delayed systems: influence of an additional delay

Abstract

[eng] In this MSc thesis, we study the performance of a delayed optoelectronic reservoir computer. We focus our atention on the system’s optimization through the addition of a second temporal delay. We find that the addition of this delay has a positive influence for some benchmark tasks while it remains neutral in other cases. In particular, we evaluate linear and nonlinear memory capacities as well as nonlinear time series prediction of a chaotic Mackey-Glass system and NARMA (Nonlinear Auto Regressive Moving Average Models) systems. First, we carry out a theoretical analysis of the reservoir computer, and precisely of the delay-based RC. The Chapter 2 is then devoted to the numerical implementation of the delayed optoelectronic oscillator. We simulate the addition of a second delay and study its effects through the benchmark tasks. Finally, in Chapter 3 we evaluate experimentally an optoelectronic hardware-based reservoir computer with one and two delays. During this experimental work, besides carrying out the different benchmark tasks, we address the reservoir computer’s characterization and look for the best set of parameters.