A statistical framework for the quantitative spectroscopy of luminous blue stars

Context . Quantitative spectroscopy of luminous blue stars relies on detailed non-local thermodynamic equilibrium (non-LTE) model atmospheres whose increasing physical realism makes direct iterative analyses computationally demanding. Aims . We introduce Machine-learning assisted Uncertainty inference (MAUI), a statistical framework designed for an efficient Bayesian inference of stellar parameters using emulator-based spectral models. Methods . MAUI employs Gaussian-process-based emulators trained on a limited set of non-LTE simulations, combined with Markov chain Monte Carlo sampling to explore posterior distributions. We validate the approach with recovery experiments and demonstrate it on Galactic late-type O dwarf and early-type B dwarf and subgiant stars. Results . The emulator reproduces the predictions of full atmosphere models within the quoted uncertainties while reducing computational cost by several orders of magnitude. Posterior distributions are well calibrated with a conservative coverage across all stellar parameters. Conclusions . The emulator-driven Bayesian inference retains the accuracy of classical analyses at a fraction of the computational expense, which enables posterior sampling that would be prohibitive with direct model evaluations. This positions emulators as a practical tool for high-fidelity spectroscopy of massive stars as atmosphere models grow more demanding.