Type: Talk
Session: Machine Learning and Artificial Intelligence
Author: Siddharth Chaini

Abstract: The advent of synoptic sky surveys has ushered in an era of big data in time-domain astrophysics, necessitating data science and machine learning as essential tools for the study of celestial objects. The Vera C. Rubin Observatory will further amplify this need by observing over 37 billion objects in its Legacy Survey of Space and Time (LSST). As we design and adapt machine learning algorithms for the LSST data, it is critical to consider interpretability, computational cost, and the adaptability of the method for specific scientific cases. Distance-based methods offer promising solutions on all these fronts, yet these approaches have not been extensively explored within the context of astrophysical classification. We recently developed a new classifier, DistClassiPy, and demonstrated its utility in the classification of variable stars. By exploring 18 distance metrics, we demonstrate that this classifier meets state-of-the-art performance while requiring lower computational resources and offering improved interpretability. To facilitate the application of DistClassiPy to transient science, we have also developed a custom feature extraction method utilizing autoencoders. Furthermore, we are in the process of extending DistClassiPy's capabilities to include anomaly detection.