Yuna Cho
Pomfret School, Pomfret, Connecticut
Abstract
This study leverages advanced deep learning techniques, specifically Neural Score Matching (NSM) and Convolutional Neural Networks (CNNs), to accurately predict and recreate weak gravitational lensing data from the Cosmic Evolution Survey (COSMOS) field. Utilizing high-resolution imagery from the Hubble Space Telescope, rigorous preprocessing ensures data accuracy and reliability. Incorporating previous research results, the NSM and CNN methodologies enable accurate probabilistic reconstruction of weak gravitational lensing images. Results validated through simulations and application to actual COSMOS data demonstrate the model’s ability to capture uncertainties and reveal complex spatial patterns, particularly in regions with massive clusters. This interdisciplinary approach enhances the precision of weak gravitational lensing analysis and significantly advances our understanding of the universe’s structure, showcasing the potential of integrating deep learning with traditional astrophysical methods and contributing novel methodologies to astrophysics.
Please click to read the paper.
