Constraining Cosmology and Baryonic Physics via Deep Learning From Weak Lensing
Video
Team Information
Team Members
Tianhuan Lu, PhD Candidate, Department of Astronomy, Graduate School of Arts and Sciences
Jose Zorrilla Matilla, PhD Candidate, Department of Astrophysical Sciences, Princeton University
Daniel Hsu, Associate Professor of Computer Science, Columbia Engineering
Faculty Advisor: Zoltan Haiman, Professor of Astronomy, Faculty of Arts and Sciences
Abstract
Ongoing and planned weak lensing surveys are becoming deep enough to contain information on angular scales down to a few arcmin. To fully extract information from these small scales, we develop a convolutional neural network architecture to learn and constrain cosmological and baryonic parameters from the simulated weak lensing convergence maps. We find that in a HSC-like survey, our network achieves a 1.7× tighter constraint in Ωm−σ8 space (1σ area) than the power spectrum and 2.1× tighter than the peak counts, showing that the network can efficiently extract non-Gaussian cosmological information even while marginalizing over baryonic effects.
Team Lead Contact
Tianhuan Lu: tl2854@columbia.edu