À la Une

Soutenance de thèse Brandon Panos


M. Brandon Panos soutiendra en anglais, en vue de l'obtention du grade de docteur ès sciences, mention interdisciplinaire, sa thèse intitulée:

The Analysis of Solar Flares Using Machine Learning

Date: Lundi 10 mai 2021 à 16h00

Lieu: en ligne, sur Zoom. Pour participer, contactez Lucia.Kleint(at)unige.ch


Jury :

  • Prof. L. Kleint, University of Geneva, thesis director
  • Prof. S. Voloshynovskiy, University of Geneva, co-director
  • Prof. S. Krucker, FHNW, co-director
  • Prof. L. Harra, ETH Zürich, jury member
  • Dr. G. Cauzzi, National Sola Observatory, jury member
  • Prof. L. Fletcher, University of Glasgow, jury member
  • Prof. D. Schaerer, University of Geneva, jury member
My Ph.D. involved the application of machine learning techniques to answer unresolved questions in solar flare physics. Solar flares are massive eruptions on the Sun caused by the reconnection of its magnetic field. These flares can release the energy equivalent of 100,000,000 megatons of TNT within the span of a few minutes. Some of this energy can be used to hurtle large volumes of charged particles laced with magnetic fields towards earth, potentially damaging satellites, disrupting GPS systems, and causing the failure of major power grids. The aim of this Ph.D. was not only to predict solar flares but also to understand their underlying physics by studying large quantities of data.
The data for the research was principally collected by NASA's small explorer Interface Region Imaging Spectrograph (IRIS), which was launched into a low earth orbit in 2013 by a Pegasus-XL rocket. This satellite is equipped with both a slit-jaw imager, that takes high-resolution contextual pictures of the Sun’s atmosphere, as well as a spectrometer, that collects the light from a number of different atomic spectral lines. IRIS focuses most of its observational power on the region of the solar atmosphere that absorbs most of the flare's energy. 
In the first part of the defense, I will offer a high-level introduction to the topic of solar flares and machine learning. The remainder of the defense will then be dedicated to explaining the four major research outputs of the project. I will discuss 1) how unsupervised clustering techniques can be used to identify similar physics operating in all flares. 2) How we can measure the degree of communication between different atmospheric layers of the Sun using a mutual information neural estimator. 3) How one can use spectra from one line to predict the most probable response in another line, and finally 4) how IRIS spectral data can be used for flare forecasting. I will close the defense with a summary and a brief outlook at the future prospects of machine learning in solar and stellar physics.