Position: Opportunities Exist for Machine Learning in Magnetic Fusion EnergyICML Oral
Spangher, L.†, Wang, A.M.†, Maris, A.D.†, Stapelberg, M., Mehta, V., Saperstein, A., et al. (†equal contribution)
Machine learning has arrived in fusion research, but the traffic has been mostly one-way: plasma physicists picking up ML tools. This paper argues the other direction is underexploited, and is written for an ML audience.
Fusion produces large, well-instrumented, physically structured datasets, and its open problems (disruption prediction, real-time control under hard latency constraints, surrogate modeling of expensive simulations, learning from small and heterogeneous experimental sets) line up closely with active areas of ML research. The field is also unusually open to outside collaborators, with public datasets and a culture of multi-institution work.
Presented as an oral at ICML 2024, an acceptance category covering under 2% of submissions. Co-first-authored with Lucas Spangher and Allen Wang.
CitationSpangher, Lucas, Andrew D. Maris, Allen Wang, et al. "Position: Opportunities Exist for Machine Learning in Magnetic Fusion Energy." Forty-first International Conference on Machine Learning (2024).