Mining large time-domain astronomical surveys using machine learning

Time-domain astronomy is the field dedicated to studying celestial objects whose brightness varies over time. These variations may be periodic (as in pulsating variable stars), stochastic (such as active galactic nuclei), or transient in nature (e.g. supernovae). In recent years, we have seen the rise of synoptic time-domain surveys aiming to detect and characterise large quantities of these objects. The challenge of manually analysing the massive databases and data streams generated by these instruments has necessitated collaboration between computer scientists, engineers, and astronomers to develop computational methods capable of automatically identifying patterns and filtering astrophysically significant events. In this talk, we introduce the fundamental concepts of machine learning, a key field in this endeavour, and demonstrate, through examples, how these techniques support astronomical research. We then focus on recent developments in deep artificial neural networks and their application to images and time series from both ground-based and space-based astronomical surveys. Finally, we explore the emerging paradigm of astronomical data brokers, which provide machine learning-based classifications for data streams from large synoptic telescopes.

Short Bio:

Pablo Huijse holds a PhD in Electrical Engineering and has been actively engaged in the field of astroinformatics since completing his doctorate in 2014. His research focuses on the development of statistical and machine learning methods for processing large astronomical datasets. He began his career working with data from ground-based observatories in his home country, Chile. More recently, he relocated to Belgium to join the Institute of Astronomy at KU Leuven, where he now works with data from ESA space missions. More at: https://phuijse.github.io/