Deep learning in attosecond metrology.

Area of Science:

• Physical Chemistry
• Quantum Dynamics
• Spectroscopy

Background:

• Time-resolved photoelectron spectroscopy is crucial for studying ultrafast electron dynamics.
• Analyzing attosecond streak camera data is computationally intensive and challenging.
• Existing algorithms are often specialized and require significant computational resources.

Purpose of the Study:

• To develop and benchmark deep neural networks for analyzing attosecond streak camera spectrograms.
• To extract information about near-infrared pulses and electron wavepackets.
• To assess the model's performance on simulated and real-world data, including uncertainty quantification.

Main Methods:

• Application of deep neural networks (DNNs) for data analysis.
• Benchmarking DNNs against simulated data.
• Domain-shift testing to evaluate performance on real-world experimental data.
• Quantification of model predictive uncertainty.

Main Results:

• DNNs achieve competitive retrieval quality for electron wavepackets and laser pulses.
• The models demonstrate superior tolerance to noisy data conditions.
• A significant reduction in computational time (orders of magnitude) was observed.
• Successful domain-shift from simulated to real-world data was illustrated.

Conclusions:

• Deep neural networks provide an efficient and robust alternative for analyzing attosecond streak camera data.
• This AI-driven approach accelerates the investigation of ultrafast electron dynamics.
• The method enhances data analysis capabilities in time-resolved spectroscopy.