Physics Attention Transformer accelerates fusion instability prediction

Researchers have developed a Physics Attention Transformer (PAT) model to rapidly predict the growth rate of vertical instabilities in fusion reactors like Alcator C-Mod and SPARC. This transformer-based approach significantly outperforms other machine learning models, such as FNO2D and DeepONet, in predicting both the scalar growth rate and the spatial distribution of perturbed toroidal current density. The PAT model achieves low mean absolute errors on held-out data and demonstrates potential for real-time control applications in future fusion devices. AI

IMPACT This model could enable faster, real-time control of plasma instabilities in fusion reactors, accelerating research and development in fusion energy.

RANK_REASON Publication of a research paper detailing a new ML model for a scientific application. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• Alcator C-Mod
• arXiv
• DeepONet
• FNO2D
• MEQ-FGE-L
• Physics Attention Transformer
• SPARC

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →