Oak Ridge scientist Pei Zhang outlines Fusion FM AI project to train models for fusion research

Pei Zhang, a computational scientist in the Computational Sciences and Engineering Division at Oak Ridge National Laboratory and the AI lead on Fusion FM, described the new project as “a seed project as part of Genesis Mission.”

Fusion FM, Zhang said, is a joint effort among PBPL, Argonne and Oak Ridge to develop a fusion foundation model — a very large AI model trained on experimental and high-fidelity simulation data that researchers can use in daily fusion science work and in designing fusion power plants. “Fusion FM is a seed project as part of Genesis Mission,” Zhang said.

The project depends on exascale computing. When asked how Frontier helps, Zhang said the supercomputer both supplies the high-fidelity physics-based simulations that provide training data and can support the large-scale training runs needed for foundation models. In Zhang’s words, Frontier “can afford and allow the large scale training of AI foundation models,” and it also meets the computing needs of the simulations used to generate that training data.

Zhang said the Genesis Mission provides a platform that combines models, world-class supercomputers and scientific expertise so teams can develop and deploy foundation models in a single environment. She added that the resulting models are intended to help address broader national challenges, though she did not specify funding sources, a deployment timeline or precise computing budgets.

The interview noted PBPL as a partner but did not expand the acronym or provide further detail about that organization’s role. Zhang emphasized the dual role of Frontier: to generate the high-fidelity simulation data and to perform the training of the AI models. She did not provide specific metrics for data volumes, training runtimes or expected release dates for Fusion FM.