Nicolaou, Constantina; (2023) Machine Learning for Multi-Messenger Astronomy. Doctoral thesis (Ph.D), UCL (University College London).

Text Nicolaou_Thesis.pdf - Other Access restricted to UCL open access staff until 1 January 2025. Download (13MB)

Abstract

The direct observation of gravitational waves (GWs) in 2015 marked the beginning of a new era of GW astronomy, unlocking an independent probe for studying the Universe. GW170817, was the first event detected in both GWs and electromagnetic (EM) observations. The implications of this multi-messenger event in the field of physics are far-reaching. Multi-messenger events can independently estimate the Hubble constant. In Chapter 2, I demonstrate the presence of a potential systematic error associated with the peculiar velocity of the host galaxy of nearby GWs, biasing the H0 estimate. I study the GW170817 event and formulate a Bayesian model that accounts for this error. Under the proposed model an unbiased estimate of the Hubble constant from nearby GW sources is obtained, H0 = 68.6+14.0−8.5 kms−1 Mpc−1, which is crucial when considering the H0 tension. In Chapter 3, I present the study of detecting and classifying GWs from core collapse supernovae (CCSNe), which are promising multi-messenger events yet to be observed. Simulated CCSNe signals were injected into real detector noise data. I implement a two-step approach comprised of wavelet-based transient detection and machine learning for classification. I compared the performance of 1D, 2D CNNs (convolutional neural networks) and LSTM (long short-term memory) and showed that 2D CNNs perform the best overall. Large galaxy surveys, play an instrumental role in EM observations of multi messenger events and studies of their properties. DESI is expected to observe 35 million galaxies. In Chapter 4, I apply Variational Autoencoders to detect anomalous spectra in DESI data. The dataset used in this analysis is composed of ∼ 208,000 spectra. The outliers identified fall into two broad categories: spectra with unique physical features and spectra with artefacts. The latter can be used to improve the DESI spectroscopic pipeline while the former can lead to the identification of transients, unusual objects and potential scientific discoveries

Download activity - last month

Export as JSONXMLCSV

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item