Efficient Optimisation of Physical Reservoir Computers using only a Delayed Input

Enrico Picco,Lina Jaurigue,Kathy Lüdge,Serge Massar
2024-01-26
Abstract:We present an experimental validation of a recently proposed optimization technique for reservoir computing, using an optoelectronic setup. Reservoir computing is a robust framework for signal processing applications, and the development of efficient optimization approaches remains a key challenge. The technique we address leverages solely a delayed version of the input signal to identify the optimal operational region of the reservoir, simplifying the traditionally time-consuming task of hyperparameter tuning. We verify the effectiveness of this approach on different benchmark tasks and reservoir operating conditions.
Emerging Technologies,Artificial Intelligence,Neural and Evolutionary Computing,Optics
What problem does this paper attempt to address?