Optimal methods for retrieving information from upcoming surveys
NHFP Hubble Fellow, The Center For Cosmology And Particle Physics, Department Of Physics, New York University
Due to the unprecedented sensitivity and large field of views, extracting the maximum amount of information remains a key challenge in future surveys. In this talk, I will present different promising methods to constrain the physics of the first light epoch and intensity mapping. In particular, I will discuss the implications of the UV radiation background inferred by recent JWST observations for radio experiments aimed at detecting the redshifted 21-cm hyperfine transition of diffuse neutral hydrogen. I will show how JWST observations can be used to place constraints on the presence of a 21 cm signal as well as the excess of radio background at Cosmic Dawn. This first part of the talk will highlight the importance of connecting different experiments to perform joint analysis to maximize the scientific return of future surveys. The second part will focus on the use of Machine Learning (ML) in cosmology. While very successful, ML models are usually regarded as black-box. Interpretability remains a key challenge due to the complexity of state-of-the-art deep learning models. I will show how a similarity measure metric of neural network representation, namely the centered kernel alignment (CKA), maye be used to examine the relationship between similarity and performance of pre-trained Convolutional Neural Networks (CNNs) on the CAMELS Multifield Dataset (CMD). By comparing representations between layers of two randomly-initialized CNN architectures, a correlation between similarity and accuracy in recovering cosmological parameters is observed. This analysis shows that exploring representation similarity against performance might offer meaningful insights into complex deep learning models to generalize them to out-of-distribution samples (OODs).
Date: Jeudi, le 7 septembre 2023 Heure: 11:30 Lieu: Université de Montréal Université de Montréal, Pavillon MIL, Local A-3561