To address the computational bottleneck of analyzing type II supernova samples from surveys like Legacy Survey of Space and Time, we present two stella-based neural network surrogates: the interaction model for low-energy explosions with potential circumstellar material (CSM) interaction and the photospheric model for standard interaction-free SNe IIP. Each surrogate follows a two-stage design in which an autoencoder first compresses stella spectral energy distributions (SEDs) into a latent representation and a separate neural-network emulator then maps physical parameters to that latent space. Both models incorporate latent mixup regularization to improve latent-space continuity, while using different network backbones tailored to their respective regimes: ResNet blocks for the interaction model and 2D CNNs for the photospheric model. On test sets, the interaction model and photospheric model achieve normalized SED reconstruction MSEs of $\approx 9.1\times {10}^{-5}$ and $1.0\times {10}^{-4}$, respectively. For the low-luminosity SN 2005cs, the interaction model favors a low-mass progenitor ($M_{\mathrm{ZAMS}}=10.40_{-0.05}^{+0.04}{M}_{\odot }$) with evidence for confined dense CSM, highlighting the likely presence of CSM interaction and offering a physical scenario consistent with direct imaging that helps resolve the historical mass discrepancy. For SN 2012aw, serving as a validation benchmark, the interaction model demonstrates consistency with previous studies by recovering a progenitor mass ($M_{\mathrm{ZAMS}}=11.05_{-0.06}^{+0.06}{M}_{\odot }$). For the archetypal SN 1999em, the photospheric model derives a progenitor mass ($M_{\mathrm{ZAMS}}=10.05_{-0.04}^{+0.07}{M}_{\odot }$) broadly consistent with direct preexplosion imaging limits without explicit CSM modeling, demonstrating that the photospheric model captures the essential physics of standard type IIP explosions. This model-level implementation can reduce full Bayesian parameter-inference runtime from days to minutes, providing a practical foundation for near-real-time physical characterization in large survey streams.

Close

PrevNext

View figure in article

• Novae & supernovae
• Transient & explosive astronomical phenomena
• Astrophysical & cosmological simulations