Abstract
Optical implementations of analogue artificial neural networks are susceptible to the inevitable fabrication and environment noise. Here we show how robustness of such networks can be enchanced by the noise injection during the training stage.
© 2021 IEEE
PDF ArticleMore Like This
Xavier Porte, Louis Andreoli, Nadezhda Semenova, Vladimir Semenov, Maxime Jacquot, Laurent Larger, and Daniel Brunner
jsi_p_10 European Quantum Electronics Conference (EQEC) 2019
Lei Liu, Xiaomin Liu, Zhiqun Zhai, Yiwen Wu, Hexun Jiang, Lilin Yi, Weisheng Hu, and Qunbi Zhuge
T2B.4 Optoelectronics and Communications Conference (OECC) 2021
Bin Shi, Bitao Pan, Nicola Calabretta, and Ripalta Stabile
Tu3B.4 Photonics in Switching and Computing (PS) 2021