Spinelli, Claudia
(2021)

*A deep learning approach to the weak lensing analysis of galaxy clusters.*
[Laurea magistrale], Università di Bologna, Corso di Studio in

Astrofisica e cosmologia [LM-DM270], Documento full-text non disponibile

Il full-text non è disponibile per scelta dell'autore.
(

Contatta l'autore)

## Abstract

In the present work we propose an innovative method which can potentially allow the unbiased estimate of galaxy cluster structural parameters from weak gravitational lensing maps using deep learning techniques. This method represents a viable alternative
to more classical and time consuming approaches which are not ideal when dealing with large datasets.
We readapted the Inception-v4 architecture to our purpose and implemented it in Pytorch. This model is trained with labeled data and then applied to unlabeled data in order to predict from the input maps, for each cluster, the virial mass, the concentration, the number of substructures and the mass fraction in substructures. The determination of these quantities is particular important because of their possible applications in several cosmological tests.
The training and test sets consist of maps produced with the MOKA software, which generates semi-analytical mass distributions of galaxy clusters and computes convergence and reduced shear maps. The simulated halos are placed at different redshifts, i.e. z = 0.25, 0.5, 0.75, 1. The complexity and the level of realism of the simulations has been increased while performing a sequence of experiments. We train the model on noiseless convergence maps, which are then replaced
with noiseless reduced shear maps. Finally, the same model is trained using reduced shear maps which include shape noise for a given number density of lensed galaxies.
We find that our model produces more accurate and precise
measurements of the virial masses and concentrations compared to the standard approach of fitting the convergence profile. It is able to learn information about the triaxial shape of the clusters during the training phase. Consequently, well known biases due to projection effects and substructures are strongly mitigated.
Even when a typical observational noise is added to the maps, the network is capable to measure the cluster structural parameters well.

Abstract

In the present work we propose an innovative method which can potentially allow the unbiased estimate of galaxy cluster structural parameters from weak gravitational lensing maps using deep learning techniques. This method represents a viable alternative
to more classical and time consuming approaches which are not ideal when dealing with large datasets.
We readapted the Inception-v4 architecture to our purpose and implemented it in Pytorch. This model is trained with labeled data and then applied to unlabeled data in order to predict from the input maps, for each cluster, the virial mass, the concentration, the number of substructures and the mass fraction in substructures. The determination of these quantities is particular important because of their possible applications in several cosmological tests.
The training and test sets consist of maps produced with the MOKA software, which generates semi-analytical mass distributions of galaxy clusters and computes convergence and reduced shear maps. The simulated halos are placed at different redshifts, i.e. z = 0.25, 0.5, 0.75, 1. The complexity and the level of realism of the simulations has been increased while performing a sequence of experiments. We train the model on noiseless convergence maps, which are then replaced
with noiseless reduced shear maps. Finally, the same model is trained using reduced shear maps which include shape noise for a given number density of lensed galaxies.
We find that our model produces more accurate and precise
measurements of the virial masses and concentrations compared to the standard approach of fitting the convergence profile. It is able to learn information about the triaxial shape of the clusters during the training phase. Consequently, well known biases due to projection effects and substructures are strongly mitigated.
Even when a typical observational noise is added to the maps, the network is capable to measure the cluster structural parameters well.

Tipologia del documento

Tesi di laurea
(Laurea magistrale)

Autore della tesi

Spinelli, Claudia

Relatore della tesi

Correlatore della tesi

Scuola

Corso di studio

Ordinamento Cds

DM270

Parole chiave

deep learning,gravitational lensing,weak lensing,cosmology,galaxy clusters

Data di discussione della Tesi

12 Marzo 2021

URI

## Altri metadati

Tipologia del documento

Tesi di laurea
(NON SPECIFICATO)

Autore della tesi

Spinelli, Claudia

Relatore della tesi

Correlatore della tesi

Scuola

Corso di studio

Ordinamento Cds

DM270

Parole chiave

deep learning,gravitational lensing,weak lensing,cosmology,galaxy clusters

Data di discussione della Tesi

12 Marzo 2021

URI

Gestione del documento: