Sample codes for training of Voronoi-tessellation-assisted convolutional neural network.
The present framework achieves robust global field reconstructions based on neural networks for moving and arbitrary number of sensors.
Kai Fukami (UCLA), Romit Maulik (ANL), Nesar Ramachandra (ANL), Koji Fukagata (Keio), and Kunihiko Taira (UCLA), "Global field reconstruction from sparse sensors with Voronoi tessellation-assisted deep learning," Nature Machine Intelligence, 3, 945-951, preprint: arXiv:2101.00554, 2021
Author: Kai Fukami (UCLA)
This repository contains
- Voronoi-CNN-cy.py (Example 1)
- Voronoi-CNN-NOAA.py (Example 2)
- Voronoi-CNN-ch2Dxysec.py (Example 3)
Authors provide no guarantees for this code. Use as-is and for academic research use only; no commercial use allowed without permission. The code is written for educational clarity and not for speed. Training data for each example can be downloaded from Google Drive. Links of GD are provided in each sample code.
Sample training data sets used in the present study are available as follows:
Example 1 (two-dimensional cylinder wake):, Example 2 (NOAA sea surface temperature):, Example 3 (turbulent channel flow):).
- Python 3.x
- keras
- tensorflow
- sklearn
- numpy
- pandas