Speaker: Julian Kates-Harbeck The prediction and avoidance of disruptions in tokamak fusion plasmas represents a key challenge on the way to stable energy production from nuclear fusion. A fusion plasma is a complex dynamical system with some unknown internal state which emits a time series of possibly high dimensional observable data that is captured by sensory diagnostics. Using such diagnostic data from past plasma shots with both disruptive and non-disruptive outcomes, we train a deep recurrent neural network to predict the onset of disruptions in an online setting. To deal with very large amounts of data and the need for iterative hyperparameter tuning, we also introduce a distributed training algorithm that runs on MPI clusters of GPU nodes and provides strong linear runtime scaling. Our approach demonstrates competitive predictive performance on experimental data from the JET tokamak, and we highlight promising avenues for extending our method to cross-tokamak prediction as well as to high-dimensional diagnostic data such as temperature and density profiles. Julian grew up in Munich, Germany. He got his bachelorâ€™s degree in physics at Stanford and later earned a masterâ€™s in computer science (with a focus on AI and machine learning) at the same university. In the period in between, he co-founded a tech startup, where he was responsible for hiring, product and strategy. He is currently pursuing a PhD in biophysics (specifically evolutionary dynamics on networks) at Harvard. For his studies there, he was awarded the National Science Foundation GRFP, Department of Defense NDSEG, and Department of Energy CSGF fellowships. Julian has written about biophysics, machine learning, astrophysics and plasma physics.