AI and the World's Hard Problems · Physics · arXiv · 2026-10-21
Probabilistic denoising for reliable signal extraction in spectroscopy
The authors introduce a probabilistic denoising framework that simultaneously extracts denoised signals and element-wise predictive uncertainties from noisy data. Applied to three-dimensional angle-resolved photoemission spectroscopy data, the model reliably recovers the spectral features of a cuprate superconductor from Poisson-distributed noise averaging only 0.02 electrons per voxel. The predicted uncertainties can be propagated into superconducting gap analyses for quantitative parameter extraction with scientifically meaningful error bars, and the approach also extends successfully to two-dimensional X-ray diffraction data.