Workshop
Machine Learning in Multibody Systems
by Grzegorz Orzechowski (LUT University) and Peter Manzl (University of Innsbruck / University of Augsburg).
Date: June 15.
Time: To be announced.
Description: This hands-on workshop provides a comprehensive introduction to machine learning with a focus on deep learning applications for multibody systems. We give an overview and explore the available toolset for practitioners and researchers in the field of multibody dynamics, and practical applications are shown. This workshop is designed for interested researchers with little to no experience in machine learning and neural networks. Sessions will feature both theoretical information and individual coding for small tasks.
- Session 1: Get an introduction to machine learning with an overview of common problems and their associated standard solutions.
- Session 2: Explore supervised learning and how neural networks can be used to learn input-output mapping using labeled data from measurements or simulations.
- Session 3: Apply Reinforcement Learning (RL) to multibody systems and see how the need for labeled data is bypassed in RL by learning from the Interaction with the environment.
- Session 4: See the latest state-of-the-art Large Language Models (LLM), generative text-based models, and their application to the generation of multibody models. In context learning and prompt tuning are applied. The LLMs can be run either locally (if you have a strong laptop) or online using, e.g., ChatGPT or Claude.
Requirements:
- Participants must bring their own laptop with Anaconda and Python (link) installed. We recommend using Python ≥ 3.10. Please note that some files will be distributed in Jupyter Notebook format.
- The code during the workshop will be distributed through a GitHub repository.
- Ideally, participants should have some basic experience with Python and the additional libraries like exudyn (link), matplotlib (link), jupyter (link), stable-baselines3 (link), and pytorch (link) already installed to get going quickly.
