Towards a radially-resolved semi-analytic model for the evolution of disc galaxies tuned with machine learning

We present a flexible, detailed model for the evolution of galactic discs in a cosmological context since $z\approx 4$, including a physically-motivated model for radial transport of gas and stars within galactic discs. This expansion beyond traditional semi-analytic models that do not include radial structure, or include only a prescribed radial structure, enables us to study the internal structure of disc galaxies and the processes that drive it. In order to efficiently explore the large parameter space allowed by this model, we construct a neural network-based emulator that can quickly return a reasonable approximation for many observables we can extract from the model, e.g. the star formation rate or the half mass stellar radius, at different redshifts. We employ the emulator to constrain the model parameters with Bayesian inference by comparing its predictions to 11 observed galaxy scaling relations at a variety of redshifts. The constrained models agree well with observations, both those used to fit the data and those not included in the fitting procedure. These models will be useful theoretical tools for understanding the increasingly detailed observational datasets from IFUs.