Abstract: Fusion energy holds a promise of a virtually limitless, carbon-free energy source. However, it is a scientific and technological grand challenge to sustainably maintain high fusion performance, requiring fuel temperatures of ~100 million degrees, while simultaneously obtaining net energy gain and avoiding overheating of the components of the power plant. To achieve these goals, machine learning (ML) methods have been within the portfolio of approaches applied by the fusion energy research community for more than two decades. Within the past few years, the role of these activities has increased substantially, and various ML approaches are nowadays routinely applied or being developed for a versatile portfolio of tasks in fusion research. Examples of such tasks are several orders-of-magnitude speed-up of computationally demanding plasma turbulence simulations and data-driven predictions of rare, high-impact events that are challenging to predict based on traditional predictive methods. This talk provides a high-level overview of some of the ML applications in fusion energy research and aims to promote discussion within the FCAI community about the potential algorithmic overlaps with other fields of science or industry.

Speaker:  Dr. Aaro Järvinen is a senior scientist in the Fusion Energy and Decommissioning research group at the VTT Technical Research Centre of Finland. He is also a visiting scientist in the EUROfusion Advanced Computing Hub at the University of Helsinki and Academy Research Fellow of the Research Council of Finland. He has been actively involved in magnetic confinement fusion energy research since 2009 and has a broad experience in both numerical and experimental fusion research, including experience with several tokamak research reactors. Within the past few years, his research focus has centered around applications of machine learning (ML) methods to facilitate the development of fusion energy, primarily focused on both Bayesian methods to facilitate model validation and on representation learning algorithms for fusion reactor scenario predictions. Overall, he is passionate about the opportunities provided by the development of ML algorithms and high-performance computing systems in accelerating the progress towards abundant and clean fusion energy.

Affiliation: VTT

Place of Seminar:  Kumpula exactum CK111 (in person) & zoom ( Meeting ID: 640 5738 7231 ; Passcode: 825217)