Abstract: We introduce new high-resolution galaxy simulations enhanced by a surrogate model that reduces the computation cost. Some stars explode at the end of their lives, known as supernovae (SNe), which play a critical role in galaxy formation. The energy released by SNe is essential for regulating star formation and driving feedback processes in the interstellar medium (ISM). However, due to insufficient mass resolution, conventional simulations have employed simple $\textit{sub-grid models}$, assuming a uniform environment, which fail to capture the inhomogeneity of the shell expansion of SNe within the turbulent ISM. Our new framework integrates numerical simulations and surrogate modeling, including machine learning and Gibbs sampling. The resulting distributions of the density and temperature of the ISM in the galaxy match those obtained from direct (resolved) numerical simulation. Our new approach achieves high-resolution fidelity while reducing computational costs by approximately 75 percent, effectively bridging the physical scale gap and enabling multi-scale simulations.