Cosmological simulations have become an essential tool for studying how galaxies form and evolve, and recent advances in machine learning have opened new ways of connecting these simulations to observations. A persistent challenge, however, is that simulations do not perfectly reproduce the real Universe, making it difficult to apply machine-learning models trained on simulations directly to observational data.
In this talk, I will present a new framework that addresses this challenge and enables us to extract information about the past assembly histories of galaxies from large observational surveys. Using a combination of cosmological simulations and machine-learning techniques, we infer the relative importance of stars formed within a galaxy and stars accreted through past mergers for nearly 10,000 nearby galaxies observed by the MaNGA survey.
The results provide some of the first large-scale, theoretically motivated constraints on the merger histories of nearby galaxies. More broadly, this work illustrates how modern machine-learning methods can help bridge the gap between simulations and observations, revealing aspects of galaxy evolution that are otherwise inaccessible from observational data alone.