view article
Figure 11
(a) The XAS spectra within our benchmarking dataset have a range of noise levels, with only a few of them being very noisy. (b) By grouping the compounds into noise bins, we can better understand how denoising methods perform across various signal-to-noise levels. At the lowest noise levels, all methods perform similarly but, once the noise level increases, the discrepancies also increase. Overall, the Gaussian processes and autoencoder denoising methods on average perform better across most noise level bins and especially when dealing with the noisiest signals. The noisiest compound within the dataset is the V K-edge BiVO4 XAS spectrum shown in (c)–(e). We also highlight regions that were `stretched out' during warping, resulting in overfitting of the noise, which is avoided by the GP denoiser.

Journal logoJOURNAL OF
SYNCHROTRON
RADIATION
ISSN: 1600-5775
Volume 33| Part 3| May 2026| Pages 658-682
https://doi.org/10.1107/S1600577526001712
Follow J. Synchrotron Rad.
Sign up for e-alerts
E-alerts
Follow J. Synchrotron Rad. on Twitter
Twitter
Follow us on facebook
Facebook
Sign up for RSS feeds
RSS