Scholarlink

We present OssicoNN, a conditional invertible neural network for analysing high-resolution stellar spectra that has been trained and tested using data from the Gaia-ESO Survey. Our approach provides robust estimates of stellar parameters and chemical abundances, achieving uncertainties of 28 K in \(\mathrm {T}_{\mathrm {eff}}\) , 0.06 dex in log g, and 0.03 dex in [Fe/H]. The network derives comprehensive uncertainty quantification, combining both internal limitations and observational uncertainties. This combination of accurate parameter determination and reliable uncertainty estimation makes OssicoNN particularly valuable for the exploitation of large spectroscopic surveys.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Stellar Parameter and Uncertainty Estimation from High-Resolution Spectroscopy Using cINN

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • N. Candebat,
  • G. G. Sacco,
  • L. Magrini,
  • F. Belfiore,
  • M. Van der Swaelmen,
  • S. Zibetti

摘要

We present OssicoNN, a conditional invertible neural network for analysing high-resolution stellar spectra that has been trained and tested using data from the Gaia-ESO Survey. Our approach provides robust estimates of stellar parameters and chemical abundances, achieving uncertainties of 28 K in \(\mathrm {T}_{\mathrm {eff}}\) , 0.06 dex in log g, and 0.03 dex in [Fe/H]. The network derives comprehensive uncertainty quantification, combining both internal limitations and observational uncertainties. This combination of accurate parameter determination and reliable uncertainty estimation makes OssicoNN particularly valuable for the exploitation of large spectroscopic surveys.