UTILIZING MACHINE LEARNING TECHNIQUES FOR DISPERSION MEASURE ESTIMATION IN FAST RADIO BURSTS STUDIES
Abstract
Fast Radio Bursts (FRBs) are highly energetic, millisecond-duration bursts of energy of- ten detected from extragalactic sources. Almost two decades after their initial discovery, the sources and underlying physical mechanisms of FRBs remain unknown. As FRB signals propagate through space, they interact with matter, resulting in a temporal dispersion with frequency. The amount of dispersion is quantified through the Dispersion Measure (DM), which specifies the time delay before the arrival of a pulse as a function of frequency. The DM plays a crucial role in estimating the distance to the FRB source and can reveal information about the physical conditions at the source. Furthermore, accurate DM estimations have far-reaching implications beyond the direct study of FRBs. They can significantly impact other areas of astrophysics, such as the study of the intergalactic medium, the mapping of cosmic structures, and providing insights into the fundamental laws of physics. These aspects highlight the importance of precise DM measurements in advancing our understanding of the universe. However, the accurate estimate of the DM is often complicated and subject to human error. Automating and improving DM estimations has become an even more important task with the increasing number of detected FRBs and the commissioning of new FRB experiments. In this thesis, we explore various deep-learning models for estimating FRB DMs. To enhance the results of train- ing models, we create a set of simulated FRB data, as real FRB data are often noisy, to develop extensive training sets. We discuss the outcomes of different deep learning models and identify the most promising ones. This research is crucial for understanding and analyzing future FRB data, which are expected to grow exponentially by the commissioning of new experiments and facilities such as the Square Kilometer Array (SKA).