journal menu
![[Figure 4]](yi5107fig4mag.jpg)
|
|
Figure 4
Illustration of how the quality of reconstruction of PAX spectra from deconvolved spectra can be used to estimate the optimal regularization strength. The panels on the left show example simulated data with 105 simulated detected electrons and a regularization strength of 7.7 meV. (A) Example ground truth spectrum and an estimate of it obtained by deconvolving simulated data. (B) Corresponding root mean squared error (RMSE) of the deconvolved spectrum as a function of the regularization strength and the number of simulated detected electrons. (C) PAX spectrum obtained by averaging a training set of data (data that were used in deconvolution) and its reconstruction from the deconvolved result. The incident photon energy of 778 eV (Co L3) combined with the Ag 3d binding energies near 370 eV (Panaccione et al., 2005  ) give electron kinetic energies near 405 eV in the PAX spectrum. (D) Corresponding RMSE of the reconstruction of the training data for the same parameters as (A). (E) PAX spectrum obtained by averaging a validation set of data (data that were not used in deconvolution) and its reconstruction from the deconvolved result. (F) Corresponding RMSE of the validation data for the same parameters as (A). The data of panel (F) are shown with the minimum of each curve subtracted and a small offset added in order to highlight the locations of minima. Since the validation reconstruction error is minimized at similar regularization strengths as the deconvolved error, we can estimate the optimal regularization strength from the validation reconstruction error. |
![[Figure 4]](yi5107fig4.jpg)
 | JOURNAL OF SYNCHROTRON RADIATION |
ISSN: 1600-5775
Open 
access
access
