Researchers have developed an AI surrogate modeling framework to predict tokamak plasma equilibrium in real-time, addressing the computational cost of traditional Grad-Shafranov solvers. The study benchmarks five neural network architectures—MLP, CNN, FNO, Transformer, and KAN—evaluating their accuracy, efficiency, and robustness on a large dataset. While the Transformer model achieved the highest accuracy on in-distribution data, the CNN demonstrated the best overall balance of speed and reliability, with CNN and FNO showing superior extrapolation capabilities on unseen data. AI
IMPACT Provides practical guidance for selecting AI surrogates for real-time plasma control in fusion applications.
RANK_REASON The cluster contains a research paper detailing a new AI methodology for a scientific application.
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
