摘要
(Received February 2022; April 2022; accepted 2022; published August 2022) In contrast to software simulations of neural networks, hardware implementations have limited or no tunability. While such networks promise great improvements in terms of speed and energy efficiency, their performance is limited by the difficulty of applying efficient training. We propose and realize experimen-tally an optical system where highly efficient backpropagation training can be applied through an array of highly nonlinear, nontunable nodes. The system includes exciton-polariton nodes realizing nonlinear acti-vation functions. We demonstrate high classification accuracy in the MNIST handwritten digit benchmark in a single-hidden-layer system.
学科与关键词
关键词
EXTREME LEARNING-MACHINE
- Energy-Efficient Neural Network Inference with Microcavity Exciton PolaritonsMatuszewski, M.*; Opala, A.; Mirek, R.; Furman, M.; Krol, M.; Tyszka, K.; Liew, T. C. H.; Ballarini, D.; Sanvitto, D.; Szczytko, J.; Pietka, B.Physical Review Applied, 2021, 16(2): 024045.被引 16
- Harnessing exciton-polaritons for digital computing, neuromorphic computing, and optimization [Invited]Opala, Andrzej; Matuszewski, Michal*The Loyola University Chicago International Law Review, 2023, 13(9): 2674-2689.被引 7
- Fully forward mode training for optical neural networksXue, Zhiwei; Zhou, Tiankuang; Xu, Zhihao; Yu, Shaoliang; Dai, Qionghai*; Fang, Lu*Nature, 2024, 632(8024).被引 2
