Feb 26, 2021Physical Sciences0 likes Auto-indexed

Sparse Representation of HINODE/SOT/SP Spectra Using Convolutional Neural Networks

Keywords:spectroscopyphotosphereSunMachine Learningsolar spectropolarimetryConvolutional Neural Networksspectra compression
Poster thumbnail

Description

A fundamental problem in solar spectropolarimetry is relating observed spectra and their polarization to the physical parameters of the underlying atmosphere. One of the difficulties in this process is the fact that the spectra usually can be represented with a much smaller number of hyperparameters than what is suggested by the number of wavelength points used for sampling. Said differently, spectra can usually be compressed or described in a sparser basis. In this work, we use the neural networks to investigate the dimensionality of photospheric spectra, and to compare the compressed spectra with the maps of physical parameters used to generate the said spectra.

Identifiers

Authors

Serena Flint

University of Rochester

Ivan Milic

LASP

Additional Information

LanguageEnglish
Formatapplication/pdf
Version1
License
Creative Commons Attribution 4.0 International
Indexed fromZenodo