Machine learning-based analysis of an electron spectrometer for high-repetition-rate laser-driven particle acceleration experiments

Accurately and rapidly diagnosing laser-plasma interactions is often difficult due to the time-intensive nature of the analysis and will only become more so with the rise of high-repetition-rate lasers. Whereas image analysis often takes several seconds even with a well-constructed algorithm, a laser-driven experiment operating at 10 Hz would need parameters of interest extracted in less than 100 ms to allow for real-time feedback and control. Machine learning-based diagnostic analysis can address this problem while maintaining a high degree of accuracy. We report on the application of machine learning to the analysis of a scintillator-based electron spectrometer for high-intensity, laser-plasma experiments at the CSU ALEPH facility. Our approach utilizes a neural network trained on synthetic data and tested on experiments to extract important electron distribution parameters. Leveraging transfer learning, we improved the accuracy of the neural network for analyzing experimental data at the speeds required in high repetition rate experiments.