Experimental validation of equation of state (EOS) models for high-energy-density applications typically involves measuring the shock Hugoniot, which helps constrain the high-pressure, high-density EOS. However, EOS measurements at below ambient density in the expanded regime are less common. Recent experimental advances allow the isentrope to be inferred from measurements of the density profile of an adiabatically expanding proton-heated plasma, offering the potential to assess various theoretical EOS models over a wide range of below ambient density conditions. On the theoretical side, determining the expanded regime EOS poses a considerable challenge for ab initio methods like density functional theory based molecular dynamics (MD) due to the big simulation cells and large number of orbitals required to simulate these low density, high temperature conditions. We thus take a more computationally friendly approach by leveraging the power of MD driven by machine learned interatomic potentials (ML-IAPs) trained on ab initio simulations. We directly simulate the adiabatic expansion of heated diamond at conditions relevant to experiments using large cells with up to 1.3 million atoms. These simulations, using several different variants of carbon ML-IAPs, are used to extract release isentropes and to study possible microscopic chemistry and phase transitions at play. These insights gained from our MD simulations can serve to guide the design and interpret observations of experiments.