Bring fusion energy closer: challenges for probabilistic inference in the multiphysics, multiscale environment of fusion devices

Track: Open Problems

Keywords: fusion, probabilistic inference, simulation-based inference

TL;DR: Fusion energy research can be greatly accelerated by probabilistic inference that reliably combines multimodal diagnostics with physics simulators to estimate plasma and device state.

Abstract: Fusion energy research can be greatly accelerated by probabilistic inference that reliably combines multimodal diagnostics with physics simulators to estimate plasma and device state. Probabilistic inference is becoming more routine across fusion devices worldwide, but can be greatly expanded in both the data and techniques used. We outline some of the open challenges in this path, inherent to the multiscale, multiphysics nature of fusion devices, namely high-dimensional, time-dependent posteriors; fidelity hierarchy of physics simulators for forward models; and wide range of multimodal diagnostics at varying spatiotemporal resolutions. We highlight some potential techniques, machine learning and related, to begin to tackle these problems.

Submission Number: 50