Asymmetrical Training for Fully-Integrated Photonic Neural Network Processors
We introduce the asymmetrical training method for training encapsulated deep photonic neural networks with optimized procedures; we show its performance enhancement from in-silico backpropagation and its potential for aiding next-generation fully-integrated photonic neural network processors.
Authors
- Minjia Chen (ORCID: https://orcid.org/0000-0001-5361-8179)
- Ting Yan (ORCID: https://orcid.org/0000-0002-8919-7131)
- Richard V. Penty (ORCID: https://orcid.org/0000-0003-4605-1455)
- Jie Ma (ORCID: https://orcid.org/0000-0002-8875-4063)
- Qixiang Cheng (ORCID: https://orcid.org/0000-0001-8671-2102)
- Yizhi Wang (ORCID: https://orcid.org/0000-0002-9232-7523)
- Ziyao Zhang (ORCID: https://orcid.org/0009-0009-8377-6173)
- Chunhui Yao (ORCID: https://orcid.org/0000-0002-8524-0833)
Institutions
- University of Cambridge (GB)
- Xuzhou Construction Machinery Group (China) (CN)
Publication Details
- Published
- 2025-06-29
- DOI
- https://doi.org/10.23919/oecc/psc62146.2025.11111165
- Primary Topic
- Neural Networks and Reservoir Computing
- Type
- article
- Field-Weighted Citation Impact
- 0.00