Towards training fault tolerant and noise immune diffractive optical neural engines
Source
Optics Infobase Conference Papers
Date Issued
2021-01-01
Author(s)
Panda, Soumyashree S.
Abstract
We report a novel robust training regimen for Diffractive Optical Networks that uses gradient based regularization terms in the training objective. Enhanced fault tolerance and noise immunity has been observed with models trained with this method.