This repository contains the code used for generating and analyzing the data in the study "Image sensing with multilayer, nonlinear optical neural networks". We constructed a physical setup of a multi-layer optical neural network (ONN) with two linear layers and an element-wise nonlinear layer in-between. The linear layers were each implemented with a fully optical matrix-vector multiplier, and the optical-to-optical nonlinear activations were implemented with an image intensifier. This repository contains the code used for training optical-neural-network encoders for the experimental setup and the code for analyzing the experimental data produced by the setup.
A quick guide on the contents of the repository:
- Folder 'Data_Collection_Example' and 'Data_Extraction_Example' contain example scripts for instrument control and data collection through the multilayer optical neural network.
- Other folders are organized according to the figures of the main text of the paper, each containing the data and the code required to reproduce the plots shown in the main text and associated supplementary figures. Note that some customized dataset used in this study can only be downloaded from Zenodo: https://doi.org/10.5281/zenodo.6888985, due to the size limit of files that are uploadable to GitHub.
- Each folder contains a README.txt file that explains the role of each file in the folder in more detail.
- For an exmaple of training an optical-neural-network enocder with a digital backend for classification, one can refer to 'Figure_2bcd/QuickDraw_train_onn_encoder.ipynb' as an introduction and 'Figure_4/Full_EBI_Cell_PLOTS.ipynb' for examples with more complex neural-network models.
If you use any of our compiled datasets and code in your research, please consider citing the following paper:
T. Wang*, M.M. Sohoni*, L.G. Wright, M.M. Stein, S.-Y. Ma, T. Onodera, M. Anderson and P.L. McMahon. "Image sensing with multilayer, nonlinear optical neural networks." arXiv:2207.14293 (2022)
The code in this repository is released under the following license:
Creative Commons Attribution 4.0 International
A copy of this license is given in this repository as License.txt.