Femtosecond-level laser pulse shaping via machine learning modeling

Recent advancements in high-power, high-repetition-rate (>0.1 Hz) laser systems have demonstrated the ability to enhance secondary radiation source performance by controlling ultrashort (~10fs) laser pulse shapes. Pulse shaping at the femtosecond level is performed by regulating the spectral phase distribution away from its best-compressed level. We present here a set of machine learning models that predict the laser input parameters (spectral phase coefficients) needed to obtain a given pulse shape. This set of Artificial Intelligence (AI) models has been trained on experimental data, usually consisting of ~10,000 daily shots, acquired at the GALADRIEL facility. Performance results from several algorithms are shown and the impact of different datasets is discussed. We provide the results from real experiments where the AI models are quickly recalibrated on site to create custom pulse shapes.