Dr. Sotiria Fotopoulou, University of Bristol, a guest of Dr. Mara Salvato (MPE), is an expert in statistics and Machine Learning and leads the efforts in the Euclid Survey to identify and classify AGNs.
Source classification is one of the most fundamental tasks in astronomy. Classes range from broad-stroke categorisation such as separating stars from galaxies, to the most minute characterisation such as pin-pointing exact evolutionary stages of stars and morphological characteristics of galaxies. The large datasets of next generation extragalactic surveys, such as Euclid and LSST, will deepen our understanding of the Universe. However, at the same time, the scale of these data pose a challenge to standard processing methods.
Based on my involvement in the preparation of the Euclid survey, I will first showcase the application of supervised and unsupervised learning on source classification across a few astronomical applications. I will then discuss the identification of outliers and the impressive potential of human-in-the-loop methods, which utilise very efficiently expert domain knowledge. Finally, I will present some recommendations on good practices in the field, including estimating uncertainty on the classification model, the impact of missing data, and the use of benchmark datasets.
Meeting ID: 620 9129 6955
Mara Salvato (MPE) and ORIGINS Excellence Cluster