Modeling Datasets with Field-Level and Halo-Centric Inference via Machine Learning

Neural networks can be utilized for improving the precision and accuracy of current and future multiwavelength cosmological and deep astronomical observations in two distinct methods. First, with cosmological field-level parameter inference from continuous datamaps. Now, novel astrophysical and observational parameters undergo halo-centric parameter inference to predict Circum-Galactic Medium (CGM) properties with only discrete points. In combining these, we aim to identify areas of interest for cosmological surveys and astronomical observations from featured neural network outputs. The CAMELS Multifield Dataset provides data-driven methods for constraining both cosmological and astrophysical (baryonic feedback) parameters. We can use parallel methods for new inferences on CGM properties, for example, using simulated IllustrisTNG data of HI and X-ray to probe cool cosmic neutral hydrogen gas and hot cosmic gas, respectively. With fewer data per set, multiple CGM parameters must be trained together for robust constraints – the network is flexible enough to handle this. This method will vastly optimize the decisions for observational surveys around galaxies, and provide a pipeline for future analyses using other simulations to create a more accurate picture of the Universe.