deepnanodesign's Introduction

DeepNanoDesign - training a bi-directional neural network for the design of nano-photonics structures

DeepNanoDesign is a software library for training deep neural networks for the design and retrieval of nano-photonic structures.

The raw dataset is available

You can download and use our raw dataset (generated by comsol). It can be found under the name "raw dataset.rar". In addition, the pre-processed version of the dataset is also available under the "inverseDataset" folder.

If you find our dataset useful, please consider citing both papers below.

Run an experiment

Training a network:

Set-up your experiment in configuration.lua.
Run experiment:

th doall.lua

Running Genetics Algorithm:

th geneticsAlgorithm.lua

Models

You can choose between:

Training a bi-directional model that given two spectrums predicts a geometry and then predicts back the two spectrums of the predicted geometry.
Training an inverse network (GPN) that only predicts a geometry.
Training a direct network (SPN) that given a geometry predicts two spectrums.
Running Genetic Algorithm (GA) to design a geometry for a given spectra.

Citation

If you find the code or the data useful in your research, please consider citing both papers:

I. Malkiel, M. Mrejen, A. Nagler, U. Arieli, L. Wolf and H. Suchowski, "Plasmonic nanostructure design and characterization via Deep Learning", Light: Science & Applications 7 (1), 60


I. Malkiel, M. Mrejen, A. Nagler, U. Arieli, L. Wolf and H. Suchowski, "Deep learning for the design of nano-photonic structures", 2018 IEEE International Conference on Computational Photography (ICCP), Pittsburgh, PA, 2018, pp. 1-14.

You can find the papers here:

https://www.nature.com/articles/s41377-018-0060-7

https://ieeexplore.ieee.org/document/8368462/

Recommend Projects