UK Atomic Energy Authority readies US artificial intelligence collaboration

The UK Atomic Energy Authority (UKAEA) has annouced plans to look at how it can work with the US Department of Energy’s Princeton Plasma Physics Laboratory (PPPL) to deliver federated supercomputing for fusion research.

During the Global Fusion Policy Summit held in London, UKAEA and PPPL signed a joint declaration of intent to explore linking UKAEA’s Sunrise AI supercomputer with PPPL’s Stellar-AI platform through a partnership known as the Sunrise–Stellar-AI Federation.

The aim is to enable researchers on both sides of the Atlantic to train artificial intelligence (AI) models on results from both national laboratories and develop AI tools for future fusion power plants, drawing on experiments completed using UKAEA’s Mast Upgrade facility in Oxfordshire and PPPL’s NSTX-U facility in New Jersey.

The federation will mean both advanced computers will be used to train the same AI models, which need large volumes of high-quality data to make accurate predictions, shortening the design cycle for future fusion power plants.

Joe Milnes, executive director for engineering and computing at UKAEA, said: “Fusion is one of the great scientific and engineering challenges of our time. To solve these challenges, fusion needs partnerships.”

The planned collaboration will also look at developing digital twins of both facilities, providing detailed virtual copies of a real fusion machine, supported by regular data from fusion experiments. This could help researchers test changes and predict how the machine will behave in software before applying experiments to hardware.

Shantenu Jha, head of computational science at PPPL, said: “Our goal is to let models and experiments move freely between the two systems. We will turn a collection of supercomputers into a single engine for fusion discovery.”

Researchers will also examine how to move computing jobs smoothly between platforms, which use different hardware, so scientists can pick the right supercomputing platform for each problem.

Jonathan Menard, chief scientist at PPPL, said: “By federating the Sunrise and Stellar-AI computer platforms, we can train AI on data from two leading fusion facilities, test ideas across two of the most capable computing systems in the world that are designed for fusion research, and accelerate the path to a compact fusion power plant.”

According to UKAEA and PPPL, when different facilities are combined, researchers can develop more powerful machine learning models that better capture the underlying physics and help accelerate future fusion projects, including the UK’s Spherical Tokamak for Energy Production and US concepts such as PPPL’s Spherical Tokamak Advanced Reactor.

The federation also wants to develop the next generation of fusion and AI experts through training, workshops, research opportunities and other international collaborations.

Rob Akers, director of computing programmes and senior fellow at UKAEA, said: “UKAEA and PPPL would draw on experimental data from two world-leading fusion machines, using simulation to fill the gaps and extend those datasets into regimes we have yet to explore. Together, we can develop digital twins of both machines to support the design of future fusion power plants, creating models of spherical tokamaks that are more predictive, more actionable, and ultimately more useful to fusion engineers.”