Page 4 - Demo

Machine learning is used at TAE to enable predictive modeling and analysis
• 𝑇# reconstructions:
• On both C-2 and C-2U, feedforward neural networks were trained to reconstruct 𝑇# time-series data from limited Thomson samples1,2
• Predictive Models:
• A convolutional neural network was trained on C-2U to
predict the staircase instability3 • Mode Activity Reconstructions:
• Collaboration with Google AI to use Bayesian Inference to mode structure from magnetic probe data4
1S. Aefsky, et al. Bulletin of the American Physical Society, Division of Plasma Physics. (2013) 2G. Player, et al. Bulletin of the American Physical Society, Division of Plasma Physics. (2017) 3C. Scott, et. al., Nucl. Fusion 60 (2020)
4P Norgaard, et al. Bulletin of the American Physical Society, Division of Plasma Physics. (2019)
Input Layer
Hidden Layers
Output Layer
Beam Power Scan C-2U Reconstruction
Presenter: Gabriel Player
4