Overview
- Developed by the UK Atomic Energy Authority, Johannes Kepler University Linz, and Emmi AI, GyroSwin is an AI surrogate trained on archives of high‑fidelity supercomputer simulations.
- The tool produces five‑dimensional gyrokinetic turbulence results in seconds, with claimed speed‑ups of up to 1,000 times over conventional methods.
- Researchers report that GyroSwin preserves key physical features such as fluctuation length scales and sheared flows, and it outperforms other AI approaches tested.
- UKAEA plans to apply the model to guide scenario design and optimisation on the MAST Upgrade device and the STEP prototype power plant.
- The team notes that supercomputers remain essential to generate training data and that the surrogate must be rigorously validated in regimes beyond current experiments; the project received part‑funding from the UK Government’s Fusion Futures Programme.