Abstract
We present the initial experiments of a project to train large-scale artificial neural networks (ANNs) using a reconfigurable optical processing system. The system is built around a Calomel (mercurous chloride) acousto-optic (AO) matrix-vector multiplier[1]. The matrix of weights are displayed on a two-dimensional LCD (liquid crystal display) panel and the input vector is encoded in the acoustic wave of the AO unit. This system has been investigated in order to determine the processing limits due to (i) the combination of individual components and to (ii) the fundamental limits imposed by the nature of this optical processing technique.
© 1998 IEEE
PDF ArticleMore Like This
Kilian Müller, Julien Launay, Iacopo Poli, Matthew Filipovich, Alessandro Capelli, Daniel Hesslow, Igor Carron, Laurent Daudet, Florent Krzakala, and Sylvain Gigan
jsiii_3_3 European Quantum Electronics Conference (EQEC) 2023
Saumil Bandyopadhyay, Alexander Sludds, Stefan Krastanov, Ryan Hamerly, Nicholas Harris, Darius Bunandar, Matthew Streshinsky, Michael Hochberg, and Dirk Englund
SM2P.2 CLEO: Science and Innovations (CLEO:S&I) 2023
James Spall, Xianxin Guo, and A. I. Lvovsky
FTu6D.2 Frontiers in Optics (FiO) 2022