Perspectives for AI/ML Utilization in Fusion Research
Explore the perspectives and success stories related to the utilization of AI/ML in fusion research, including insights on fast surrogate models, efficient sampling of parameter space, and language/spell checking applications. Discover the importance of validating models, the need for error estimates, and considerations for incorporating AI/ML in the fusion community.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
You are allowed to download the files provided on this website for personal or commercial use, subject to the condition that they are used lawfully. All files are the property of their respective owners.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.
E N D
Presentation Transcript
Perspectives for utilization of AI/ML Fredric Granberg And ACH-VTT Team Faculty of Science / Fredric Granberg / E-TASC GM I www.helsinki.fi/yliopisto 11-15.11.2024 1
AI/ML in fusion research The aim is to raise the question about how to treat AI/ML in the fusion community I do not advocate for or against, neither am I an expert, and probably I will contradict myself What do we need, or do we even need any discussion about this? Wild west? Guidelines? At the moment all proposals needs to include aspects of AI/ML Can be trivial Faculty of Science / Fredric Granberg / E-TASC GM I www.helsinki.fi/yliopisto 11-15.11.2024 2
Some examples on success stories AI/ML/NN fast surrogate models to heavy simulations/or combinatorial problems Speep-up simulations Utilize accuracy of more precice models at larger scale AI based algorithms for efficient sampling of paramater space Can reduce human time for sampling, and be better than a grid LLMs for language/spell checking / document searching Good starting point for non-native speakers Faculty of Science / Fredric Granberg / E-TASC GM I www.helsinki.fi/yliopisto 11-15.11.2024 3
Common things All of these can be validated or at least checked with a accurate model If a strange result arises, it can be checked with a existing model/human Many current AI/ML papers are good models with small validation errors, but no new physics nor insights Faculty of Science / Fredric Granberg / E-TASC GM I www.helsinki.fi/yliopisto 11-15.11.2024 4
Problems Error estimate is not existing in many of these models It can be included in to the models, to have an estimate of the certainty of a point Not bullet-proof, but a good start, which can be improved Example: A NN estimate of X was generated by training of experimental X by choosing some parameters they thought will affect X. A nice evolution was extrapolated, but no possibility to get an error estimate, neither can it be validated (within many years/decades) Faculty of Science / Fredric Granberg / E-TASC GM I www.helsinki.fi/yliopisto 11-15.11.2024 5
What should we do? Anything goes, as long as it is AI/ML? Guidelines? Standard EUROfusion AI methods/software ? Faculty of Science / Fredric Granberg / E-TASC GM I www.helsinki.fi/yliopisto 11-15.11.2024 6
ACH-VTT We have had very little AI/ML related tasks I think it can be counted on one hand in the whole E-TASC period from beginning Maybe still in its infancy, so no need for support? Examples this far Anomaly detection for simulations Bayesian optimization Fast surrogate models for heavy codes Faculty of Science / Fredric Granberg / E-TASC GM I www.helsinki.fi/yliopisto 11-15.11.2024 7
THANK YOU FOR YOUR ATTENTION Happy to discuss this further during brea Faculty of Science / Fredric Granberg / E-TASC GM I www.helsinki.fi/yliopisto 11-15.11.2024 8
