Journal of the Korean Physical Society

pISSN 0374-4884 eISSN 1976-8524

Article

Cross-Disciplinary Physics and Related Areas of Science and Technology

Published online December 31, 2021     https://doi.org/10.1007/s40042-021-00346-1

Copyright © The Korean Physical Society.

Iterative peak-fitting of frequency-domain data via deep convolution neural networks

Seong-Heum Park, Hyeongseon Park, Hyunbok Lee, Heung-Sik Kim

J. Korean Phys. Soc. 79(12), 1199 - 1208 (2021)

Abstract

We combine deep convolution neural networks and conventional global optimization techniques to analyze x-ray photoemission data. By employing the most advanced form of convolution neural network architecture, we could iteratively subtract peak features from synthetic and actual photoemission spectra, after which conventional global optimization methods can be applied to calibrate the results. Our model could recognize defect-induced peaks in MoS2 and WS2 that could be easily overlooked, showing the power of deep neural network models in capturing subtle but existing features from experimental results.