| Abstract | This repository provides Python code accompanying the tutorial paper on inverse photonic design using neural networks, as described in the associated manuscript.
This tutorial is structured into clear steps corresponding to the sections of the paper, covering:
Forward Model: Training a neural network model to predict TE coupling coefficients from geometry parameters.
Simple forward model (N_MODELS = 1, AUGMENT_DATA = False)
Simple forward model with data augmentation (N_MODELS = 1, AUGMENT_DATA = True)
Ensemble forward model (N_MODELS > 1, AUGMENT_DATA = True recommended)
Inverse Model: Predicting geometry parameters from specified TE coupling coefficients.
Simple inverse neural network (without tandem)
Tandem inverse neural network (with pre-trained forward network)
The provided Python script allows you to reproduce these examples directly, using simple flags and parameters. Additionally, the folder ./Simulations_setup contains example code and needed scripts to run Ansys Lumerical simulations for data generation. |
|---|
