School of Physics Colloquium

Speaker: Sultan Hassan (NYU)

Title: Optimal methods for extracting information from upcoming surveys.

Abstract: Extracting the maximum amount of astrophysical and cosmological information remains a challenge in the current and future surveys. These include, for instance, the James Webb Space Telescope (JWST), the Euclid, the Spectro-Photometer for the History of the Universe, Epoch of Reionization, and Ices Explorer (SPHEREx), the Nancy Grace Roman Space Telescope (Roman), and many more. Due to the unprecedented sensitivity and large field of view, future surveys will enable detecting the majority of high redshift sources (quasars and galaxies) on extremely large scales.  Hence, a new generation of theoretical models and statistical tools is required to maximize the scientific return of future surveys. In this talk, I will present several machine learning (ML) models and techniques that are capable of simultaneously generating new synthetic diverse examples of large-scale maps, and enabling high-dimensional inference at the field level. In addition, I will address one of the biggest challenges of using ML to extract information from surveys, namely the out-of-distribution (OOD) generalization problem. Interpretability approaches are a natural way to gain insights into this problem. I will show how similarity of learning representation may be used as an inductive bias to improve the performance of pre-trained Convolutional Neural Networks (CNNs) on OOD samples. This analysis shows that exploring representation similarity against performance might offer meaningful insights into complex deep learning (black box) models to generalize them to OOD samples.

Bio: Sultan Hassan is from Sudan. He obtained his PhD from the University of the Western Cape, South Africa in 2018. He held several prize fellowships starting with the SKA fellowship in South Africa, then moved to New Mexico State University where he was a Tombaugh fellow for two years till 2020, followed by a Flatiron research Fellow at the Center of Computational Astrophysics in New York till 2022, and recently a NASA Hubble Fellow at New York University as of January 2023. Sultan works at the intersection of Astrophysics, Cosmology and Machine learning. His expertise ranges from cosmic dawn, reionization, galaxy formation to large scale intensity mapping techniques.