Page 19 - Demo

The model has been shown to be effective on both synthetic and experimental data
• In situ methodology has the benefit of being fast and easy
• It can be used to test diagnostics without needing to remove them from the machine or set up
test beds
• Machine learning significantly reduces the required dataset size compared to traditional deterministic methods
• Synthetic results are highly promising:
• Using a dataset with similar dimensionality as real experimental data, channels were calibrated
with error < 5 %
• Increasing the size of the dataset provides minimal benefits beyond ~10 shots, as the reconstruction algorithm has fully “learned” the source distribution at that point
• Experimental results are similarly good:
• Several known low-response channels are calibrated up to the correct signal response levels
Presenter: Gabriel Player
19