Abstract
A simple coherent-optical implementation of a Rectifying Linear Unit (ReLu) as an interferometric phase-sensitive bidiectional switch along with convolutional layers using lenslet arrays and thick Fourier holograms allows the efficient implemention of Deep Neural Networks (DNN) in a self-aligning multilayer adaptive-holographic optical neural network.
© 2017 Optical Society of America
PDF ArticleMore Like This
Kelvin Wagner and Demetri Psaltis
TuA4 Optical Computing (IP) 1987
Dmitry S. Raupov, Oleg O. Myakinin, Ivan A. Bratchenko, and Valery P. Zakharov
JTu2A.4 Frontiers in Optics (FiO) 2017
Hao Wang, Meng Lyu, Ni Chen, and Guohai Situ
DW2F.2 Digital Holography and Three-Dimensional Imaging (DH) 2018