Artificial Intelligence-Based Class Assessment and Rare Sources Screening in Fermi Blazars

Active Galactic Nuclei (AGN) are among the most captivating and dynamic entities in the cosmos. These astronomical marvels, marked by their intense electromagnetic emissions in broad energy range (from radio to $\gamma$-ray bands), are central to our comprehension of galactic evolution. Within the diverse subclasses of AGN, blazars emerge as an especially extreme and fascinating subset. Their distinctive attributes, such as rapid variability and pronounced polarization, render them indispensable for astrophysical inquiries. The primary aim of this dissertation is twofold: first, to probe the intricacies and challenges associated with binary classification or clustering in blazars, and second, to leverage these techniques to pinpoint blazars exhibiting unique properties. The opening chapter lays the groundwork, providing a comprehensive overview of AGN, delving into the classification and unification models of AGN, and spotlighting the identification process for blazars with rare physical attributes, marking them as the most extreme AGN subclass. Ch. \ref{chap:fermi} is allocated to the comprehensive study of the \textit{Fermi} Gamma-ray Space Telescope with its significant catalogs like 3FHL and 4FGL series. This chapter will elucidate the telescope's instrumentation, mission objectives, and the wealth of discoveries made possible by its observations. Ch. \ref{chap:methodology} delves into the methodologies adopted, with a special highlight on the transformative power of Artificial Intelligence (AI) in astrophysical research. AI's capability to process vast datasets, its unparalleled accuracy in pattern recognition, and its adaptability in evolving research landscapes underscore its significance in modern astrophysics. Ch. \ref{chap:dichotomy} through Ch. \ref{chap:neutrino candidates} offer an in-depth analysis of four groundbreaking studies. Each of these studies not only contributes to the broader narrative of binary classification or clustering but also advances our understanding of rare blazars. 1. Chapter on Radio Dichotomy in AGNs: This chapter investigates the optical B band and 6 cm radio wavelength data of AGNs, uncovering a distinct division in radio loudness, represented by $\log R$, at $\log R = \langle1.37 \pm 0.02\rangle$. This delineates radio-loud (RL) and radio-quiet (RQ) AGNs. A dual criterion for AGN classification is proposed, integrating radio luminosity with loudness. 2. Chapter on Blazar Jets Studied through VLBI Technique: This section presents a study of 407 VLBI-detected \textit{Fermi} blazars, revealing a correlation between $\gamma$-ray luminosity and apparent velocity, suggesting a link between jet knot movement and jet power. Gaussian mixture models are employed to identify 228 potential VFB candidates, using criteria like $\log L_{\gamma} > 45.40$, $\alpha_{\rm ph} > 2.24$, and $\log VI > 1.71$. 3. Chapter on Identifying TeV Blazar Candidates using Supervised Machine Learning: This chapter applies Logistic Regression (LR) to catalogs such as 4FGL-DR2 / 4LAC-DR2 to distinguish TeV blazars from non-TeV counterparts. The LR model assigns probabilities to each source, indicating the likelihood of being a TeV blazar. Utilizing an 80% threshold, 40 high-confidence TeV candidates are identified from the 4FGL-DR2 / 4LAC-DR2 blazars. 4. Chapter on Probing the Origins of High-Energy Cosmic Rays and Blazars as Potential Neutrino Emitters: Utilizing transfer learning (TL) and an artificial neural network (ANN), this chapter analyzes the \textit{Fermi}-LAT Gamma-ray Source Catalog (4FGL-DR3) to differentiate neutrino blazar (NB) from non-NB entities, identifying 273 NB candidates. In conclusion, Chapter 8 synthesizes the dissertation's findings, emphasizing the potential of AI methodologies in elucidating the complexities of blazar research and predicting the expanding role of AI in future astrophysical studies.