ACCELERATING KINEMATIC LENSING INFERENCE WITH NEURAL NETWORKS

Abstract

Cosmological surveys in the next decade will provide us with an unprecedented amount of data for Kinematic Lensing (KL) studies. KL infers the cosmic shear signal by jointly forward modeling the observed photometric image and velocity field of a disk galaxy, allowing for shear measurements with greatly reduced statistical noise. We show that there are good prospects for a future KL survey using the Dark Energy Spectroscopic Instrument (DESI), and that a pilot measurement can already be made using data from the DESI Peculiar Velocity (DESI-PV) survey. However, the process of KL inference for cosmic shear using MCMC will be time-consuming, and will become unfeasible with a large number of galaxies. In this work, we explore ways to accelerate KL inference using neural networks, in preparation for the DESI-KL pilot measurement and future KL endeavors using large survey data. Specifically, we created an algorithm to mass-generate galaxy images and spectra, which can then be used for neural network training and validation. We also develop a neural network schematic for emulating image and spectra generation, which we plan to implement and test in a future work.