Physics-aware Machine Learning in Earth Observation and Modelling Workshop ML4Earth

Wednesday 20 Sept 2023, 09:00 → 17:00 Europe/Berlin

Kolpinghaus, Munich

The national center of excellence ML4Earth invites you to participate in the Physics-aware Machine Learning in Earth Observation and Modelling workshop to be hosted at the Conference Center Kolpinghaus Munich on September 20, 2023. This event is organized in collaboration with the Helmholtz AI and the International Future Lab AI4EO.

During the morning, you will learn about numerical methods for physics simulations with deep learning methods in Prof. Nils Thuerey’s talk, and about data-efficient deep learning using physics-informed neural networks in Prof. Maziar Raissi’s keynote. Later, we will discuss on a round table with Prof. Thuerey and Dr. Shijie Jiang the present and future of Physics-aware Machine Learning.

In the afternoon, there will be a hands-on introduction to Physics-aware Machine Learning, and to end the day, the ML4Earth Hackathon will be introduced.

We are looking forward to meeting you in person. However, if you can’t be there, we will also be happy to get in touch with you online (https://tum-conf.zoom-x.de/j/66875415974?pwd=NlJtdllnbWI5Q095RUpPdmlzbTEzZz09).

 

Kind regards,
ML4Earth Team

 

Prepare for the Physics-aware Machine Learning Tutorial

We're looking forward to the physics-aware machine learning tutorial tomorrow. In this tutorial, you'll be training neural networks in PyTorch to carry out time integration for the shallow water equations, and adding several physical constraints to improve performance. 

In preparation for the tutorial, you may wish to review the following materials

Python: https://www.learnpython.org

Pytorch: https://pytorch.org/tutorials/beginner/deep_learning_60min_blitz.html

Please bring your own laptop and charger. We will use colab for online interactive jupyter notebooks, so you don't need to install anything. However, if you want to run everything locally or on another remote system, please install the required libraries (see below).

Best regards,
David Greenberg & Ali Bekar

Required python libraries: 

via conda:
h5py
torch
numpy
pdb
matplotlib
pandas

via pip:

tensorboard
setuptools==59.5.0

Make sure you use the above version of setuptools, not the latest version!