AI for Fusion Energy Challenge

Closing the gaps on fusion science with the help of AI

Within the IAEA Coordinated Research Project on AI for Fusion, this AI for Good Challenge aims to provide a platform for collaboratively explore the potential of ML in enabling predictive modelling for fusion energy systems. Fusion energy is produced when two light elements, i.e., hydrogen isotopes, combine to form a single heavier one, with the consequent release of energy. The scientific community is actively engaged in making fusion a commercially viable alternative energy source – scientists and engineers worldwide are collaborating to make this a reality. Through this Challenge, participants will use data from three distinct fusion devices called “tokamaks” (Alcator C-Mod, J-TEXT, and HL-2A) to develop a cross-machine disruption prediction model using ML, with strong generalization capabilities. Participants will acquire hands-on experience in AI/ML in areas relevant to fusion energy science and compete for prizes, recognition, and certificates. 

Within the IAEA Coordinated Research Project on AI for Fusion, this AI for Good Challenge aims to provide a platform for collaboratively explore the potential of ML in enabling predictive modelling for fusion energy systems. Fusion energy is produced when two light elements, i.e., hydrogen isotopes, combine to form a single heavier one, with the consequent release of energy. The scientific community is actively engaged in making fusion a commercially viable alternative energy source – scientists and engineers worldwide are collaborating to make this a reality. Through this Challenge, participants will use data from three distinct fusion devices called “tokamaks” (Alcator C-Mod, J-TEXT, and HL-2A) to develop a cross-machine disruption prediction model using ML, with strong generalization capabilities. Participants will acquire hands-on experience in AI/ML in areas relevant to fusion energy science and compete for prizes, recognition, and certificates. 

Compute platform

ITU provides a state-of-the-art, free-of-charge compute platform to participants of the Challenge who do not have adequate access to compute in their respective institutions. The compute platform will provide participants with access to: 

 

  • Free GPUs and CPUs
  • Hosted Jupyter notebook server
  • Python kernel
  • Pre-installed machine learning packages, e.g. PyTorch and Tensorflow

In some of the problem statements, a baseline or reference solution may be offered which may include implementations using Jupyter notebooks.

The AI for Fusion Energy Challenge features one problem statement

Multi-Machine Disruption Prediction

Develop a disruption prediction model with cross-machine generalization capabilities, using J-TEXT and HL-2A as the current fusion machines and Alcator C-Mod as the future fusion machine.

Contributors will be asked to sign a data sharing agreement upon joining the challenge. 

AI for Fusion Energy Challenge Timeline

September – October 2023

November 2023

December 2023

Competition Phase

Evaluation Phase

Awards

AI for Fusion Energy Challenge Timeline

September – October 2023

Competition Phase

November 2023

Evaluation Phase

December 2023

Awards

Contributors

The realization of this data challenge was made possible thanks to the work of Diakhere Gueye (IAEA), Dr Chengshuo Shen (HUST), Dr Jinxiang Zhu (PSFC), Dr Zheng Wei (HUST), Dr Zhongyu Yang (SWIP), Dr Cristina Rea (PSFC), Matteo Barbarino (IAEA), and Thomas Basikolo (ITU).

Intern
International Atomic Energy Agency (IAEA)
Research Scientist
Massachusetts Institute of Technology (MIT)
Fusion Scientist
International Atomic Energy Agency (IAEA)
Young Expert
Programme Officer
International Telecommunication Union (ITU)
AI for Fusion Energy Challenge partners

Hosts of problem statements

Sponsorship Inquiries​


    Are you sure you want to remove this speaker?