Extensive studies of the neutron star equation of state from the deep learning inference with the observational data augmentation

A bstract We discuss deep learning inference for the neutron star equation of state (EoS) using real observational data mass and radius. make a quantitative comparison between conventional polynomial regression neural network approach EoS parametrization. For our method to incorporate uncertainties in observation, we augment training with noise fluctuations corresponding uncertainties. Deduced EoSs can accommodate weak first-order phase transition, histogram likely regions. also find that augmentation has byproduct tame overfitting behavior. To check performance improved by augmentation, set up toy model as simplest problem recover double-peaked function monitor validation loss. conclude could be useful technique evade without tuning architecture such inserting dropout.