Abstract
This paper utilises knowledge distillation to compress a convolutional neural network trained to learn the nonlinear Schrodinger equation. The teacher-taught student network has improved generalisation, quicker convergence, and fewer trainable parameters. The proposed network is 91.2% compressed with a mean square error comparable to the teacher.
© 2022 The Author(s)
PDF Article | Presentation VideoMore Like This
Sasipim Srivallapanondh, Pedro J. Freire, Bernhard Spinnler, Nelson Costa, Antonio Napoli, Sergei K. Turitsyn, and Jaroslaw E. Prilepsky
Th1F.7 Optical Fiber Communication Conference (OFC) 2023
Bixia Tang, Jianying Chen, Yue-Cai Huang, Yun Xue, and Weixing Zhou
T4A.82 Asia Communications and Photonics Conference (ACP) 2021
Lauri Salmela, Mathilde Hary, Mehdi Mahed, John M. Dudley, and Goëry Genty
NpM3G.2 Nonlinear Photonics (NP) 2022