Data-Driven Methods in Spectroscopy

Understanding the formation and evolution of the Milky Way and its constituents relies on making measurements of data that are faithful representations of the underlying physical system. Data-driven spectral models function in several capacities in this context; generating spectra, extracting encoded information, and connecting that information to theory. We review how data-driven methods have changed the spectroscopic landscape and assess their future: ▪ A core utility of data-driven models is that they deliver precise stellar measurements at scale and with modest computational cost, providing consistent parameters and abundances across surveys, including for low-resolution and low signal-to-noise spectra. This circumvents prior limitations and unlocks the statistical power of large datasets. ▪ Entirely new avenues of study have been opened up by inferred spectroscopic ages, distances, and evolutionary states, whereas residual analyses have uncovered nonstellar signals and allow a novel discovery space. ▪ Biases can be introduced or inherited by data-driven models; identifying and accounting for them is important for accurate interpretation. ▪ Data-driven models do not replace theoretical models; together, they enable the interpretation and parameterization of spectra. ▪ A key benefit of data-driven frameworks, alongside public releases of survey data, is democratization of spectroscopic analysis. ▪ Future facilities and computational tools will expand data-driven methods and their utility. Although this review focuses on the stellar temperature range of 3,000–7,000 K, the methods are general.