Reservoir Computing using VCSEL Polarization Dynamics
Jeremy Vatin, Damien Rontani, Marc Sciamanna · 2019
The huge amount of data generated everyday motivates the development of new approaches for information processing. Machine-learning is one of these approaches: it explores the possibility to solve problems without having its explicit formulation. In this field, reservoir computing allows to easily implement a neural network with physical components since the training of the network is performed only at the readout layer and typically from linear regression methods [1]. A readout layer is trained using a linear regression [1]. Several all-optical architectures have been developed for reservoir computing, e.g. using fiber-based optic [2] and optical integrated components [3].