In this talk I will introduce the concept of reservoir computing as a recently proposed, highly efficient bio-inspired approach for processing time dependent data, and describe our experimental implementation of this concept in an optoelectronic circuit. The basic scheme of reservoir computing consists of a non linear recurrent dynamical system, coupled to a single input layer and a single output layer; only the linear output layer is trained to produce the desired output as a response to an input time series. One of the main advantages of this technique is that the training procedure is very simple and straightforward; furthermore, it allows a great amount of freedom when choosing the dynamical system to be used as a reservoir. By choosing a simple dynamical system, composed by a nonlinear node and a delay line, and implemented by an optoelectronic circuit, we are able to create an experimental reservoir computer. The reservoir computer has been tested on several tasks, such as system modelization, nonlinear channel equalization and speech recognition, with results comparable to the ones obtained by state-of-the-art software reservoirs; we believe it represents a first step towards an all-optical, high-performance nonconventional computer.
Reservoir Computing, Optoelectronic circuits, Nonlinear channel equalization, Speech recognition
Yvan Paquot and Francois Duport and Anteo Smerieri and Joni Dambre and Benjamin Schrauwen and Marc Haelterman and Serge Massar. (2011)
Optoelectronic Reservoir Computing,
Scientific Reports, .
submitted; preprint available at http://arxiv.org/abs/1111.7219v1.