Line intensity mapping (LIM) is an observational technique that measures collective line intensity emissions from distant galaxies and intergalactic media. Observed large-scale structures provide information on cosmology and galaxy formation and evolution. A limitation with LIM is line confusion, where the contributions of different lines from different redshifts are all confused. In this study, we propose to use machine learning to solve it. We devise a generative adversarial network that extracts designated signals from LIM data. By training it on mock observational maps that simulate future observations, our network properly reconstructs the peak positions and statistics of intensity fluctuations. In this talk, I will introduce our new machine-learning architecture, show the results of the reconstruction, and discuss the applicability of the network for future experiments.
