The Computing Division Technical Meetings are a platform for

  • presenting Pinboard papers under review for journal or conference publication,
  • inviting speakers who are current or prospective UKAEA collaborators at external organisations,
  • presenting work done in PhD projects funded by or co-supervised by UKAEA,
  • presenting work done during summer placements or other secondments to UKAEA.

 

If you would like to invite a speaker on a topic that would be of interest to one or more Units within the Computing Division, but is not currently collaborating on a UKAEA project, please consider nominating them for a Computing Division Cross-Disciplinary Seminar.

These meetings are normally recorded. Recordings of past meetings can be found here:

CD Technical Meetings Archive on UKAEA Sharepoint

CD Technical meeting (HD2): From Diagnostics to Deployment: Towards Open, FAIR Data and Software Infrastructure for AI-Driven Fusion Science at UKAEA

→ Europe/London
Samuel (Computing) Jackson (UKAEA)
Description

Abstract

Realizing the promise of AI/ML for fusion energy requires infrastructure spanning the full data lifecycle, from raw diagnostic signals, through curation and benchmarking, to real-time deployment in plasma control. This talk presents a progress on data, infrastructure, tools, and models being developed by UKAEA and collaborators to address this pipeline.

We start at the data layer, where FAIR-MAST [7, 6] and progress on its successor, the Fusion Data Service (FDS), provide open, scalable access to fusion diagnostic data and metadata data, aligned with community standards such as IMAS and DCAT, and underpinned by a shared vocabulary for datasets, devices and workflows through the FUEL ontology [12]. For further interoperability, we developed libtokamap [13] to map experimental data into the IMAS schema.

Turning raw signals into curated, labelled, AI-ready datasets is the job of TokTagger [4], a open source, web-based platform for interactive labeling of diagnostic data of different modalities either by hand or from annotation models, and NiceShot [5], an interactive dashboard for exploring, visualizing, searching, clustering, and filtering relevant shots by metadata across whole experimental campaigns.

Curated data enables downstream collaborations such as TokaMark[11], an open benchmark of over 11,000 real MAST shots across 14 different tasks, including forecasting and reconstruction tasks across heterogeneous diagnostics covering magnetic, kinetic, radiative, actuators, and equilibrium data. From this benchmark, we built TokaMind [3], a multi-modal, multi-fidelity transformer model for fusion diagnostics that supports lightweight adaptation to new tasks.

Finally, we will turn to operational deployment through the Real-Time Virtual Circuit (RTVC) programme [2, 1, 8, 10, 9], which replaces pre-computed lookup tables for tokamak shape control with neural-network emulators of the plasma response trained on synthetic FreeGSNKE equilibria. RTVC preserves the interpretability and modularity of existing control architectures while adapting in real time to the plasma state, and has reached experimental deployment on MAST Upgrade utilizing common emulators across distinct scenarios without scenario-specific retraining.

Together, these projects trace a path from raw data to real-time AI in fusion science. We will close by discussing the governance, reproducibility and interoperability practices needed to sustain this infrastructure as it scales across facilities and matures toward broader AI integration in fusion workflows.

References

[1] A. Agnello et al. Emulation techniques for scenario and classical control design of tokamak plasmas. Physics of Plasmas, 31:043901, 2024.

[2] N. C. Amorisco et al. FreeGSNKE: A Python-based dynamic free-boundary toroidal plasma equilibrium solver. Physics of Plasmas, 31:042517, 2024.

[3] Tobia Boschi, Rodrigo H. Ordonez-Hurtado, Eszter Sz´ekely, Adriano Agnello, Robert Akers, Andrea Loreti, C´ecile Rousseau, Alexander Whittle, Stanislas Pamela, Juan Bernabe-Moreno, Mykhaylo Zayats, Nicola C. Amorisco, George K. Holt, Samuel Jackson, Alessandra Pascale, and Vassil Alexandrov. TokaMind: A multi-modal transformer foundation model for tokamak plasma dynamics, 2026.

[4] S. Jackson, M. Field, J. Blake, A. Saleem, N. Bhatia, R. Costa, N. Cummings, S. Khan, P. Sharma, S. Pamela, and A. Gonzalez-Beltran. TokTagger: An open source tool for labelling and tagging tokamak data, 2026.

[5] Samuel Jackson. NiceShot!: An interactive dashboard for exploring tokamak plasma shot data, May 2026. Python package; provides scatter plots, correlations, time traces, filtering, clustering, and outlier detection over shot statistics.

[6] Samuel Jackson, Saiful Khan, Nathan Cummings, James Hodson, Shaun de Witt, Stanislas Pamela, Rob Akers, and Jeyan Thiyagalingam. An open data service for supporting research in machine learning on tokamak data. IEEE Transactions on Plasma Science, 2025.

[7] Samuel Jackson, Saiful Khan, Nathan Cummings, James Hodson, Shaun de Witt, Stanislas Pamela, Rob Akers, Jeyan Thiyagalingam, and The MAST Team. FAIR-MAST: A fusion device data management system. SoftwareX, 27:101869, 2024.

[8] K. Pentland et al. The FreeGSNKE pulse design tool (FPDT): a computational framework for evolutive plasma scenario and control design. Plasma Physics and Controlled Fusion, 68:075035, 2026.

[9] K. Pentland et al. Real-time virtual circuits for plasma shape control via neural network emulators: dynamic validation in closed-loop simulations, 2026. Submitted.

[10] A. Ross et al. Real-time virtual circuits for plasma shape control via neural network emulators, 2026. Submitted.

[11] Cecile Rousseau, Samuel Jackson, Rodrigo H. Ordonez-Hurtado, Nicola C. Amorisco, Tobia Boschi, George K. Holt, Andrea Loreti, Eszter Sz´ekely, Alexander Whittle, Adriano Agnello, Stanislas Pamela, Alessandra Pascale, Robert Akers, Juan Bernabe Moreno, Sue Thorne, and Mykhaylo Zayats. TokaMark: A comprehensive benchmark for MAST tokamak plasma models, 2026.

[12] UKAEA. FUEL: FUsion Energy Lexicon, an ontology for fusion energy data and processes, 2026. Provides structured semantic definitions for fusion energy-related information and workflows.

[13] UKAEA. libtokamap: A C++20 library for flexible data mapping and transformation with support for multiple data sources and mapping types, 2026. Plugin-based architecture supporting JSON-configured value, data source, expression, dimension, and custom mappings, with built-in caching.

Participants
  • Jony Castagna
    • 1
      Talk
      Speaker: Samuel (Computing) Jackson (UKAEA)
    • 2
      Q&A
      Speaker: Samuel (Computing) Jackson (UKAEA)