Reconfigurable Integrated Photonic Unitary Neural Networks With Phase Encoding Enabled by In-Situ Training
Photonic neural networks are emerging as promising computing platforms for artificial intelligence (AI). Particularly, integrated photonic unitary neural networks (IPUNNs) are capable of mitigating gradient vanishing/explosion problems when deeper neural networks are constructed. Furthermore, their...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2024-01-01
|
| Series: | IEEE Photonics Journal |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10663838/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832590403370483712 |
|---|---|
| author | Shengjie Tang Cheng Chen Qi Qin Xiaoping Liu |
| author_facet | Shengjie Tang Cheng Chen Qi Qin Xiaoping Liu |
| author_sort | Shengjie Tang |
| collection | DOAJ |
| description | Photonic neural networks are emerging as promising computing platforms for artificial intelligence (AI). Particularly, integrated photonic unitary neural networks (IPUNNs) are capable of mitigating gradient vanishing/explosion problems when deeper neural networks are constructed. Furthermore, their optical implementations are also much simpler compared to non-unitary counterparts. Meanwhile, real-valued datasets still dominate AI research and the encoding strategy is critical for IPUNNs' performances. However, there are few studies to compare different encoding strategies of IPUNNs to represent these real-valued datasets and their impacts on IPUNNs' performances. Here, in the scope of encoding strategies for real-valued features, we first compare different schemes, such as phase, amplitude and hybrid encoding using numerical simulations, with benchmarks of decision boundary and image recognition tasks. These encoding strategies of IPUNNs are also compared to non-unitary real-valued neural networks (RVNNs) with trainable biases for the same benchmarks. The results suggest that phase encoding outperforms amplitude and hybrid encoding, and exhibits comparable performances to non-unitary RVNNs. To verify the numerical results, a 10×10 IPUNN chip is designed and fabricated. The phase encoding is chosen to be implemented because of its superior performances in numerical studies. We reconfigure the IPUNN chip to perform decision boundary and image recognition tasks by on-chip in-situ training. The experimental results match the simulations well. Our work provides insights for implementing reconfigurable IPUNNs in AI computing. |
| format | Article |
| id | doaj-art-801d49189c9f43d09625f1f20ff5e4bc |
| institution | Kabale University |
| issn | 1943-0655 |
| language | English |
| publishDate | 2024-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Photonics Journal |
| spelling | doaj-art-801d49189c9f43d09625f1f20ff5e4bc2025-01-24T00:00:46ZengIEEEIEEE Photonics Journal1943-06552024-01-0116511110.1109/JPHOT.2024.345389810663838Reconfigurable Integrated Photonic Unitary Neural Networks With Phase Encoding Enabled by In-Situ TrainingShengjie Tang0https://orcid.org/0000-0003-3272-2924Cheng Chen1https://orcid.org/0009-0006-9334-9106Qi Qin2https://orcid.org/0009-0005-5068-8629Xiaoping Liu3https://orcid.org/0000-0001-7955-2067School of Physical Science and Technology, ShanghaiTech University, Shanghai, ChinaSchool of Physical Science and Technology, ShanghaiTech University, Shanghai, ChinaCollege of Physics and Optoelectronic Engineering, Shenzhen University, Shenzhen, ChinaSchool of Physical Science and Technology, ShanghaiTech University, Shanghai, ChinaPhotonic neural networks are emerging as promising computing platforms for artificial intelligence (AI). Particularly, integrated photonic unitary neural networks (IPUNNs) are capable of mitigating gradient vanishing/explosion problems when deeper neural networks are constructed. Furthermore, their optical implementations are also much simpler compared to non-unitary counterparts. Meanwhile, real-valued datasets still dominate AI research and the encoding strategy is critical for IPUNNs' performances. However, there are few studies to compare different encoding strategies of IPUNNs to represent these real-valued datasets and their impacts on IPUNNs' performances. Here, in the scope of encoding strategies for real-valued features, we first compare different schemes, such as phase, amplitude and hybrid encoding using numerical simulations, with benchmarks of decision boundary and image recognition tasks. These encoding strategies of IPUNNs are also compared to non-unitary real-valued neural networks (RVNNs) with trainable biases for the same benchmarks. The results suggest that phase encoding outperforms amplitude and hybrid encoding, and exhibits comparable performances to non-unitary RVNNs. To verify the numerical results, a 10×10 IPUNN chip is designed and fabricated. The phase encoding is chosen to be implemented because of its superior performances in numerical studies. We reconfigure the IPUNN chip to perform decision boundary and image recognition tasks by on-chip in-situ training. The experimental results match the simulations well. Our work provides insights for implementing reconfigurable IPUNNs in AI computing.https://ieeexplore.ieee.org/document/10663838/In-situ trainingphase encodingphotonic unitary neural networkreconfigurable |
| spellingShingle | Shengjie Tang Cheng Chen Qi Qin Xiaoping Liu Reconfigurable Integrated Photonic Unitary Neural Networks With Phase Encoding Enabled by In-Situ Training IEEE Photonics Journal In-situ training phase encoding photonic unitary neural network reconfigurable |
| title | Reconfigurable Integrated Photonic Unitary Neural Networks With Phase Encoding Enabled by In-Situ Training |
| title_full | Reconfigurable Integrated Photonic Unitary Neural Networks With Phase Encoding Enabled by In-Situ Training |
| title_fullStr | Reconfigurable Integrated Photonic Unitary Neural Networks With Phase Encoding Enabled by In-Situ Training |
| title_full_unstemmed | Reconfigurable Integrated Photonic Unitary Neural Networks With Phase Encoding Enabled by In-Situ Training |
| title_short | Reconfigurable Integrated Photonic Unitary Neural Networks With Phase Encoding Enabled by In-Situ Training |
| title_sort | reconfigurable integrated photonic unitary neural networks with phase encoding enabled by in situ training |
| topic | In-situ training phase encoding photonic unitary neural network reconfigurable |
| url | https://ieeexplore.ieee.org/document/10663838/ |
| work_keys_str_mv | AT shengjietang reconfigurableintegratedphotonicunitaryneuralnetworkswithphaseencodingenabledbyinsitutraining AT chengchen reconfigurableintegratedphotonicunitaryneuralnetworkswithphaseencodingenabledbyinsitutraining AT qiqin reconfigurableintegratedphotonicunitaryneuralnetworkswithphaseencodingenabledbyinsitutraining AT xiaopingliu reconfigurableintegratedphotonicunitaryneuralnetworkswithphaseencodingenabledbyinsitutraining |