Autocorrelation-based convolutional neural network for reconstruction of noisy attosecond streaking traces

Attosecond light sources serve as crucial tools for investigating the ultrafast electronic dynamics in matter with remarkable temporal resolution. Traditional methods face difficulties in accurately measuring attosecond pulses, and the prevailing approach involves utilizing attosecond streak cameras coupled with inversion algorithms to reconstruct phase information. However, these algorithms often require multiple iterations and extensive computational time. This study investigates the utilization of autocorrelation graphs as inputs for a convolutional neural network (CNN) to invert streaking traces obtained by attosecond streak camera. We explore the noise resistance capability of autocorrelation within the CNN inversion and aim to provide a physical explanation for its effectiveness. The objective of this research is to enhance the accuracy and reliability of CNN inversion for attosecond streaking traces, enabling improved resilience against experimental noises.