Page 9 - Demo

Machine learning-driven calibrations require much less data than traditional methods
• ML-powered calibration requires only a handful of shots (𝑁.-/,. < 𝑁0-*((#1.)
• Calibration process:
1. Over multiple similar plasma shots, adjust
NPA energy range
2. Reconstruct the NPA source distribution using ML
3. Determine systematic deviations from the source distribution
4. Use ML to calibrate channels to minimize these systematic deviations
• To verify this method, we tested on synthetic data, with the method shown in the chart
Experimental Inputs: Estimated noise levels, NPA channel energies, number of shots
Synthetic Inputs: Source distribution, randomized channel response levels
Test Methodology Refine Methodology
Apply to experimental dataset & calibrate
Presenter: Gabriel Player
9