AI and the World's Hard Problems · Physics · arXiv · 2026-10-22
Fast2Peak: High-Fidelity Restoration Algorithm by Single-Pixel Masking towards Real-Time Spectral Measurements
Rapid spectroscopy pushes measurements into photon-starved conditions where weak spectral features are obscured by stochastic noise. Fast2Peak (F2P), a self-supervised Transformer, turns spectral restoration into a context-prediction problem, using single-pixel masking to exploit the local continuity of physical signals and suppress random noise without clean reference spectra. It restores spectra within milliseconds, enabling real-time acquire-and-process operation demonstrated in time-resolved electrochemical SERS and hyperspectral Raman imaging, and generalizes across Raman, NMR, XRD and FTIR.