Helsinki advanced computing hub
Helsinki Advanced Computing Hub provides fusion modelling with the highest level of artificial intelligence, big data handling, as well as expertise in fusion and computational physics, backed up by 500 specialists in the same location, at the Kumpula Campus.
Watch a success story video AI for fusion energy at Finnish Center for Artificial Intelligence site.
The Hub brings state-of-the-art computational methods to physicists
The Advanced Computing Hub is part of a larger EUROfusion project consisting of five hubs providing the computational support to 14 other projects, solving the still unanswered questions related to fusion energy. This network of projects will advance the understanding of fusion devices and improve the predictive capability of the simulation models. The Helsinki hub focuses on providing support in the utilisation of artificial intelligence, tools for validation, verification and uncertainty quantification, and help with GPU programming and data-management, among other things.
University of Helsinki
The Hub focuses on artificial-intelligence tools in fusion applications
CSC – IT Center for Science participates in the project by providing IT hardware platforms for the hub together with support from a senior scientist within the realm of high performance computing. The early focus is on code optimisation and the proper testing and porting of codes to CSC platforms, as well as support for the setup of data bases, while later the focus is planned to be on producing data and developing AI tools for fusion research.
Computer scientists and algorithm designers support international fusion experiments
Hands-on experience with new tools and platforms enables computer scientists and algorithm designers to improve physics modeling. Such close interaction with physicists is essential in supporting international experiments, especially the ITER project.
Bayesian neural networks that are based on a priori physical information and admit uncertainty quantification in a natural way, are already being used in predicting thermal-quench onset to enable real-time disruption avoidance. Representing molecular dynamics interaction potentials with Gaussian process regression can be exploited in speeding up simulations studying the behaviour of fusion-related materials such as EUROFER steels in high irradiance conditions. Finally, as an ultimate goal for the community, machine-learning emulators may one day enable the development of high-fidelity “flight simulators” for training and benefit of fusion reactor control operators.