Registration is now open!
This workshop will provide brief introductions to probabilistic machine learning and modern uncertainty quantification techniques. Attendees will be given hands-on demonstrations with digiLab's twinLab probabilistic machine learning framework as well as the UM-bridge platform which simplifies deployment of simulations onto HPC resources for various UQ tasks. The workshop will be split into two sections:
digiLab: Applying Probabilistic ML to Enhance Fusion Simulations https://www.digilab.co.uk/
During this session, we will overview machine learning techniques, and why trust and explainability is so important in the field of fusion. We'll see how probabilistic techniques can help improve trust and explainability, and how these techniques are currently being applied in the field, from plasma science to systems engineering. Finally, we'll get hands on with twinLab, digiLab's platform for probabilistic machine learning, and see how this platform can enhance simulation workflows for plasma science & fusion.
Linus Seelinger (KIT): UQ and UM-bridge https://um-bridge-benchmarks.readthedocs.io/en/docs/
High-level analyses (uncertainty quantification (UQ), optimization, ML, ...) enable simulation models to deliver greater value and deeper scientific insight. UM-Bridge breaks down the technical complexity of such analyses, makes it easy to apply high-level analysis tools to complex numerical simulations, and allows even prototype applications to transparently scale to supercomputers. This tutorial gives an introduction to UQ and UM-Bridge, and includes hands-on UQ analyses of realistic simulation models.
