Femtosecond pulse parameter estimation from photoelectron momenta using machine learning

Abstract Deep learning models have provided huge interpretation power for image-like data. Specifically, convolutional neural networks (CNNs) demonstrated incredible acuity tasks such as feature extraction or parameter estimation. Here we test CNNs on strong-field ionization photoelectron spectra, training theoretical data sets to ‘invert’ experimental Pulse characterization is used a ‘testing ground’, specifically retrieve the laser intensity, where ‘traditional’ measurements typically lead 20% uncertainty. We report crucial augmentation techniques required successfully train and return consistent results from experiments, including accounting detector saturation. The same procedure can be repeated apply in range of scenarios ionization. Using predictive uncertainty estimation, reliable intensity uncertainties few percent extracted, which are consistently lower than those given by traditional techniques. interpretability methods reveal parts distribution that most sensitive directly associated with holographic interferences. employed provide an accurate convenient ways extract parameters, represent novel interpretational tool spectra.