Recent advancements in artificial intelligence (AI) and machine learning (ML) offer numerous opportunities for learning models of complex physical phenomena directly from data. However,
these models often come without error guarantees, act as black boxes, can require large amounts of training data, and cannot always be trusted in the same way as traditional high-fidelity, physics-based models that are derived from first principles. However, AI/ML models have proven useful in multi-fidelity computations, where AI/ML models are used for speeding up computations while high-fidelity, physics-based models remain in the loop to establish trust in predictions and decisions. This workshop will explore mathematical and computational foundations of learning models from data as well as their integration into multi-fidelity methods, bridging the gap between data-driven and physics-based approaches in fusion plasma physics.
Key challenges to be addressed include active data acquisition for generating training data and as well as training methods for learning AI/ML models of high-dimensional kinetic equations, developing methods for learning generative models of stochastic, turbulent, and chaotic systems, and integrating AI/ML models with physics-based models in multi-fidelity methods. Additionally, the workshop will discuss effective use of AI/ML models in multi-fidelity methods for robust and reliable uncertainty quantification, design, and control in the context of fusion plasma physics. By tackling these challenges, the workshop aims to provide a deeper understanding of how AI and physics-based models can work together to advance plasma simulations and facilitate tasks that are infeasible in terms of high-fidelity, physics-based simulations such as the design and control of optimized fusion devices.
Anima Anandkumar
(California Institute of Technology)
Ionut-Gabriel Farcas
(University of Texas at Austin)
Greg Hammett
(Princeton Plasma Physics Lab)
Benjamin Peherstorfer
(Courant Institute of Mathematical Sciences)
Yunan Yang
(Cornell University)
Copyright. All Rights Reserved.