Abstract
Artificial neural networks (ANNs) have emerged as powerful tools, which for example accelerate research on COVID-19. They solve challenging tasks by taking advantage of their self-learning abilities and nonlinearity [1, 2]. However, the freedom gained through the complexity of ANNs leads to computational intense model training, eventually consuming a lot of energy [3]. One possible solution to overcome the current issues could be by transferring ANNs into the optical domain. For instance, Shen et al. have demonstrated the rich capabilities of optical neural networks (ONNs) [4]. A key element of an ANN and ONN is its nonlinear activation function, however, only a few all-optical approaches are published. In addition, those published are frequency insensitive, limiting the maximum data bandwidth throughput.
© 2023 IEEE
PDF ArticleMore Like This
Yu Zhang, Hong Deng, Emadreza Soltanian, Jing Zhang, Gunther Roelkens, and Wim Bogaerts
ck_13_2 The European Conference on Lasers and Electro-Optics (CLEO/Europe) 2023
Yichun Li, Zizhou Wei, Han Wang, Haozhe Shou, Fufei Pang, and Liang Zhang
JTh2A.135 CLEO: Applications and Technology (CLEO:A&T) 2023
Laura Blázquez Martínez, Philipp Wiedemann, Andreas Geilen, Changlong Zhu, and Birgit Stiller
ef_7_3 European Quantum Electronics Conference (EQEC) 2023