High-sensitivity extreme-ultraviolet transient absorption spectroscopy enabled by machine learning.

Area of Science:

• Spectroscopy
• Machine Learning
• Materials Science

Background:

• Spectroscopy experiments often suffer from noise, limiting the sensitivity of measurements.
• Conventional noise reduction methods, like pump-on/pump-off referencing, have limitations, especially for broadband light sources.

Purpose of the Study:

• To introduce a novel denoising scheme for spectroscopy using machine learning.
• To demonstrate improved noise suppression in XUV transient absorption spectroscopy.

Main Methods:

• Measuring probe spectra before and after sample interaction.
• Utilizing machine learning to capture correlations between spectral components.
• Applying artificial neural networks for pixel-wise noise reduction.

Main Results:

• Achieved up to a tenfold improvement in noise suppression compared to conventional methods.
• Demonstrated effective noise reduction without requiring wavelength calibration of reference spectra.
• Successfully applied the scheme to XUV transient absorption measurements.

Conclusions:

• The novel denoising scheme offers significant noise reduction for spectroscopy.
• The method is adaptable to various experiments, particularly beneficial for low repetition-rate sources.
• Enhanced sensitivity enables detailed studies of electron and lattice dynamics in materials.