Open and FAIR Fusion for Machine Learning Applications

This DOE-funded multi-institutional collaboration aims to develop a Fusion Data Platform for Machine Learning (ML) applications using Magnetic Fusion Energy data. Now entering its second year, this Open Science effort has accomplished several important milestones. One main goal of the project is to develop the MDSplusML framework, which will build on top of existing MDSplus installations but will expand its APIs to enable efficient AI/ML workflows. Multiple cashing technologies have been investigated to accelerate data retrieval and their relative cost-to-benefit ratio has been studied in order to pick the top performing candidate for inclusion in MDSplusML. Additionally, mapping of Alcator C-Mod data onto the common ITER data ontology has begun, with the intention of releasing an open source API for metadata queries. Then, the Physics-based, interoperable DisruptionPy workflow developed at the PSFC has been revamped and expanded to study disruption physics across different devices and is scheduled to be released as open-source at the next major release. Furthermore, Open Research Products include the open-source publication of several curated databases for disruption prediction, tearing mode instability, and confinement and transport, which are all underway. Finally, the first undergraduate summer school on Data Science and Fusion applications has been organized at W&M and leveraged as a platform to train a diversified workforce in Fusion sciences and AI/ML tools.