PyTorch code for Composing Partial Differential Equations with Physics-Aware Neural Networks

Last update: Dec 18, 2022

Related tags

Overview

FInite volume Neural Network (FINN)

This repository contains the PyTorch code for models, training, and testing, and Python code for data generation to conduct the experiments as reported in the work Composing Partial Differential Equations with Physics-Aware Neural Networks

If you find this repository helpful, please cite our work:

@article{karlbauer2021composing,
	author    = {Karlbauer, Matthias and Praditia, Timothy and Otte, Sebastian and Oladyshkin, Sergey and Nowak, Wolfgang and Butz, Martin V},
	title     = {Composing Partial Differential Equations with Physics-Aware Neural Networks},
	journal   = {arXiv preprint arXiv:2111.11798},
	year      = {2021},
}

Dependencies

We recommend setting up an (e.g. conda) environment with python 3.7 (i.e. conda create -n finn python=3.7). The required packages for data generation and model evaluation are

conda install -c anaconda numpy scipy
conda install -c pytorch pytorch==1.9.0
conda install -c jmcmurray json
conda install -c conda-forge matplotlib torchdiffeq jsmin

Models & Experiments

The code of the different pure machine learning models (TCN, ConvLSTM, DISTANA) and physics-aware models (PINN, PhyDNet, FINN) can be found in the models directory.

Each model directory contains a config.json file to specify model parameters, data, etc. Please modify the sections in the respective config.json files as detailed below (further information about data and model architectures is reported in the according data sections of the paper's appendices):

"training": {
	"t_stop": 150  // burger and allen-cahn 150, diff-sorp 400, diff-react 70
},

"validation": {
	"t_start": 150,  // burger and allen-cahn 150, diff-sorp 400, diff-react 70
	"t_stop": 200  // burger and allen-cahn 200, diff-sorp 500, diff-react 100
},

"data": {
	"type": "burger",  // "burger", "diffusion_sorption", "diffusion_reaction", "allen_cahn"
	"name": "data_ext",  // "data_train", "data_ext", "data_test"
}

"model": {
  	"name": "burger"  // "burger", "diff-sorp", "diff-react", "allen-cahn"
	"field_size": [49],  // burger and allen-cahn [49], diff-sorp [26], fhn [49, 49]
	... other settings to be specified according to the model architectures section in the paper's appendix
}

The actual models can be trained and tested by calling the according python train.py or python test.py scripts. Alternatively, python experiment.py can be used to either train or test n models (please consider the settings in the experiment.py script).

Data generation

The Python scripts to generate the burger, diffusion-sorption, diffusion-reaction, and allen-cahn data can be found in the data directory.

In each of the burger, diffusion_sorption, diffusion_reaction, and allen-cahn directories, a data_generation.py and simulator.py script can be found. The former is used to generate train, extrapolation (ext), or test data. For details about the according data generation settings of each dataset, please refer to the corresponding data sections in the paper's appendices.

You might also like...

Official implementation for the paper: "Multi-label Classification with Partial Annotations using Class-aware Selective Loss"

Multi-label Classification with Partial Annotations using Class-aware Selective Loss Paper | Pretrained models Official PyTorch Implementation Emanuel

99 Dec 27, 2022

Must-read Papers on Physics-Informed Neural Networks.

PINNpapers Contributed by IDRL lab. Introduction Physics-Informed Neural Network (PINN) has achieved great success in scientific computing since 2017.

330 Jan 7, 2023

Physics-informed convolutional-recurrent neural networks for solving spatiotemporal PDEs

PhyCRNet Physics-informed convolutional-recurrent neural networks for solving spatiotemporal PDEs Paper link: [ArXiv] By: Pu Ren, Chengping Rao, Yang

11 Aug 23, 2022

Physics-Aware Training (PAT) is a method to train real physical systems with backpropagation.

Physics-Aware Training (PAT) is a method to train real physical systems with backpropagation. It was introduced in Wright, Logan G. & Onodera, Tatsuhiro et al. (2021)1 to train Physical Neural Networks (PNNs) - neural networks whose building blocks are physical systems.

230 Jan 5, 2023

Pytorch Implementation of Interaction Networks for Learning about Objects, Relations and Physics

Interaction-Network-Pytorch Pytorch Implementraion of Interaction Networks for Learning about Objects, Relations and Physics. Interaction Network is a

117 Nov 5, 2022

IDRLnet, a Python toolbox for modeling and solving problems through Physics-Informed Neural Network (PINN) systematically.

IDRLnet IDRLnet is a machine learning library on top of PyTorch. Use IDRLnet if you need a machine learning library that solves both forward and inver

105 Dec 17, 2022

Releases(v1.0.0)

v1.0.0(Oct 28, 2022)

This release contains the PyTorch code for models, training, and testing, and Python code for data generation to conduct the experiments.
Source code(tar.gz)
Source code(zip)

PyTorch code for Composing Partial Differential Equations with Physics-Aware Neural Networks

Related tags

Overview

FInite volume Neural Network (FINN)

Dependencies

Models & Experiments

Data generation

You might also like...

Official implementation for the paper: "Multi-label Classification with Partial Annotations using Class-aware Selective Loss"

Must-read Papers on Physics-Informed Neural Networks.

Physics-informed convolutional-recurrent neural networks for solving spatiotemporal PDEs

Physics-Aware Training (PAT) is a method to train real physical systems with backpropagation.

Pytorch Implementation of Interaction Networks for Learning about Objects, Relations and Physics

IDRLnet, a Python toolbox for modeling and solving problems through Physics-Informed Neural Network (PINN) systematically.

PINN(s): Physics-Informed Neural Network(s) for von Karman vortex street

Implementation of "Scaled-YOLOv4: Scaling Cross Stage Partial Network" using PyTorch framwork.

Unofficial pytorch implementation of 'Image Inpainting for Irregular Holes Using Partial Convolutions'

Releases(v1.0.0)

v1.0.0(Oct 28, 2022)

Owner

Cognitive Modeling

AnimationKit: AI Upscaling & Interpolation using Real-ESRGAN+RIFE

Running AlphaFold2 (from ColabFold) in Azure Machine Learning

[CVPR 2022 Oral] TubeDETR: Spatio-Temporal Video Grounding with Transformers

High-Resolution Image Synthesis with Latent Diffusion Models

An end-to-end framework for mixed-integer optimization with data-driven learned constraints.

Official implementation of the paper Chunked Autoregressive GAN for Conditional Waveform Synthesis

Open source implementation of AceNAS: Learning to Rank Ace Neural Architectures with Weak Supervision of Weight Sharing

Re-implementation of 'Grokking: Generalization beyond overfitting on small algorithmic datasets'

Datasets and source code for our paper Webly Supervised Fine-Grained Recognition: Benchmark Datasets and An Approach

Updated for TTS(CE) = Also Known as TTN V3. The code requires the first server to be 'ttn' protocol.

MILK: Machine Learning Toolkit

Differentiable Wavetable Synthesis

Conjugated Discrete Distributions for Distributional Reinforcement Learning (C2D)

The official GitHub repository for the Argoverse 2 dataset.

A PyTorch implementation for PyramidNets (Deep Pyramidal Residual Networks)

Official implementation of CATs: Cost Aggregation Transformers for Visual Correspondence NeurIPS'21

You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling

Pytorch implementation for "Distribution-Balanced Loss for Multi-Label Classification in Long-Tailed Datasets" (ECCV 2020 Spotlight)

Bridging Composite and Real: Towards End-to-end Deep Image Matting

[WWW 2021] Source code for "Graph Contrastive Learning with Adaptive Augmentation"