Discovering gravitational lenses with artificial intelligence
|PhD ceremony:||Mr C.E. (Enrico) Petrillo|
|When:||November 01, 2019|
|Supervisor:||prof. dr. L.V.E. (Léon) Koopmans|
|Co-supervisors:||dr. G.A. (Gijs) Verdoes Kleijn, dr. C. Tortora|
|Where:||Academy building RUG|
|Faculty:||Science and Engineering|
Gravitational lensing is a phenomenon that occurs when the light from a background galaxy is bent by a foreground galaxy before reaching the observer.Gravitational lenses are a unique tool to study the structure of galaxies and to measure fundamental cosmological parameters such as the expansion rate of the Universe.Thus, it is important to identify these occurences of aligned galaxies which are particularly rare, about one for each thousand galaxies.Traditionally, this search has heavily relied on the visual inspection of images by astronomers.However, it has becoming increasingly difficult to build complete samples of gravitational lenses in such a way because upcoming astronomical surveys are going to observe billions of galaxies. For this reason it is essential to develop automatic lens-finder algorithms.In this thesis, I have developed and trained machine learning algorithms based on artificial neural networks to identify gravitational lenses in the Kilo-Degree Survey (KiDS). KiDS is an optical survey that has observed part of the South sky with unprecedented image quality.The artificial neural networks have analyzed images of tens of thousands galaxies and have aided in discovering hundreds of new gravitational lens candidates for which the images are being collected at the website https://www.astro.rug.nl/lensesinkids.This way I have demonstrated that applying these kinds of automatic methods for finding gravitational lenses is feasible in large sets of imaging data. In fact, many millions of galaxies will be analyzed by a computer neural network instead of a human eye and its neural network. This way it will be possible to select thousands of lens candidates as input for astrophysical research with minimal human intervention.