Manish Shrimali

Central University of Rajasthan, India

Reservoir Computing: the bridge between Complex Dynamics and Artificial Intelligence

Abstract

Reservoir Computing is getting increasing attention as a promising architecture to perform computation with dynamical systems. It uses the dynamical system itself as a computing substrate. A reservoir computer consists of a dynamical reservoir to encode the input into a high-dimensional states and a linear readout to analyze those higher dimensional pattern and extracting the required output. Importantly, it requires very small training data sets, uses linear optimization, and thus requires minimal computing resources. Different kinds of dynamical systems can act as potential reservoir having a rich dynamics. According to recent findings in the field, various physical systems including mechanical, electronic, photonic, spintronics, chemical, biological systems has been proven to be successful in performing complex tasks. For all different reservoirs there are mainly two factors we need to take care of for its best performance in accordance with the given tasks: the method to feed the input to the reservoir and the optimal configuration of the reservoir. In recent work, it has been show that in a network of coupled nonlinear oscillators the occurrence of explosive transition helps to achieve that critical configuration of the reservoir.

Find Manish Shrimali