Reducing acquisition time and radiation damage: data-driven subsampling for spectro-microscopy

Meier, M.; Lazzarino, L.; Shustin, B.; Al Daas, H.; Quinn, P.

doi:10.1107/S1600577526006491

Figure 4
The amount of information retained with different subsampling strategies. The information measure is (1 − ||D − D_completed||_F) / (1 − ||D − D_optimal||_F) × 100%, where $[{\bf{D}}_{{\rm completed}}]$ is the completed dataset (with filled in entries) and $[{\bf{D}}_{{\rm optimal}}]$ is the best rank-10 approximation to $[{\bf{D}}]$ according to the Eckart–Young–Mirsky theorem (Eckart & Young, 1936

), i.e. truncated rank-10 singular value decomposition (SVD). Solid lines show the mean across trials, while shaded bands represent ±1 standard deviation. The dataset is DS1 from Townsend et al. (2022

). For the raster sampling approaches, completion is done with LoopedASD, and the data shown are the average of ten iterations. For the CUR sampling approaches, the CUR decomposition is the completion and the data shown are the average of 100 iterations.

JOURNAL OF
SYNCHROTRON
RADIATION

ISSN: 1600-5775

Volume 33| Part 5| September 2026| Pages 1440-1454

https://doi.org/10.1107/S1600577526006491

Open

access