IvS Seminar: Pablo Huijse (28/11/2024)

Pablo Huijse - "Deep Learning in Observational Astronomy: Theory and Applications" (IvS seminar 28/11/2024)
Abstract: Artificial neural networks (ANNs), machine learning models inspired by the brain and proposed over four decades ago, have recently achieved widespread prominence by addressing complex pattern recognition challenges across various disciplines. This resurgence has been fueled not only by key methodological innovations but also, and perhaps more critically, by the availability of compute resources and large-scale datasets. The latter is particularly relevant to observational astronomy, where modern surveys generate data volumes far exceeding the capacity of manual analysis. In this context, deep neural networks-ANNs with multiple layers-have emerged as powerful tools for the automated analysis of unstructured astronomical data, especially images and time series. This seminar begins with an introduction to the basic theory of ANNs, the machine learning tasks they address, and the practical aspects of how to train them effectively. We then discuss the paradigm shift from traditional machine learning to deep learning, highlighting key neural architectures through examples using images and time series from astronomical surveys such as ZTF and Gaia. Finally, we present the case of the ALeRCE system, a machine learning-based community broker developed to process alerts from the Vera C. Rubin Observatory.