Large stellar surveys are revealing the chemodynamical structure of the Galaxy across a vast spatial extent. However, the many millions of low-resolution spectra observed to date have not yet been fully leveraged. In chapters 2 and 3, we employ data-driven spectroscopic models to the low-resolution LAMOST survey (𝑅 = 1800). In chapter 2, we employ The Cannon, a data-driven approach for estimating chemical abundances, to obtain detailed abundances, using the GALAH survey as our reference. We deliver five (for dwarfs) or six (for giants) estimated abundances representing five different nucleosynthetic channels, for 3.9 million stars, to a precision of 0.05 - 0.23 dex. Using wide binary pairs, we demonstrate that our abundance estimates provide chemical discriminating power beyond metallicity alone. We show the coverage of our catalogue with radial, azimuthal and dynamical abundance maps, and examine the neutron capture abundances across the disk and halo, which indicate different origins for the in-situ and accreted halo populations. LAM- OST has near-complete Gaia coverage and provides an unprecedented perspective on chemistry across the Milky Way. Stars with unusual levels of enrichment in a particular element are of great interest, but often pose a problem for data-driven methods. In chapter 2, we present a simple method, for the de- tection of 𝑋-enriched stars, for arbitrary elements ?, even from blended lines. Our method does not require stellar labels, but instead directly estimates the counterfactual unrenriched spectrum from other unlabelled spectra. We apply this method to the 6708 Å Li doublet in LAMOST DR5, identifying 8,428 Li-enriched stars seamlessly across evolutionary state. We comment on the ex- planation for Li-enrichement for different subpopulations, including planet accretion, nonstandard mixing, and youth. The Galactic disk exhibits complex chemical and dynamical substructure thought to be induced by the bar, spiral arms, and satellites. In chapter 4, rather than calculating spectroscopic quantities, we use them to understand the Milky Way. We explore the chemical signatures of bar resonances in action and velocity space and characterize the differences between the signatures of corotation and higher-order resonances using test particle simulations. Thanks to recent surveys, we now have large datasets containing metallicities and kinematics of stars outside the solar neighborhood. We compare the simulations to the observational data from Gaia EDR3 and LAMOST DR5 and find weak evidence for a slow bar with the “hat” moving group (250 km s⁻¹ ≲ 𝑣_𝜑i ≲ 270 km s⁻¹) associated with its outer Lindblad resonance and “Hercules” (170 km s⁻¹ ≲ 𝑣_𝜑 ≲ 195 km s⁻¹) with corotation. While constraints from current data are limited by their spatial footprint, stars closer in azimuth than the Sun to the bar’s minor axis show much stronger signatures of the bar’s outer Lindblad and corotation resonances in test particle simulations. Future datasets with greater azimuthal coverage, including the final Gaia data release, will allow reliable chemodynamical identification of bar resonances. Finally, in chapter 5, we present KORG, a new package for 1D LTE (local thermal equilib- rium) spectral synthesis, which computes theoretical spectra from the near-ultraviolet to the near- infrared, and implements both plane-parallel and spherical radiative transfer. It is compatible with automatic differentiation libraries, and easily extensible, making it ideal for statistical inference and parameter estimation applied to large data sets. We outline the inputs and internals of KORG, and compare its output spectra to those produced by other codes. We use five example wavelength regions across 3660 Å – 15050 Å to show that the residuals between KORG and the other codes are no larger than that between existing codes themselves. We show that KORG is 1–100 times faster than other codes in typical use.

In the last few decades, the emergence of large-scale infrared surveys has led to a revolution in the study of star formation. In particular, NASA’s Spitzer Space Telescope has recently carried out mid- and far-infrared observations of numerous star formation regions with unprecedented resolution and sensitivity, and has uncovered thousands of forming stars. In combination with present and future large-scale near-infrared and sub-mm surveys, spectral energy distributions from near-infrared to mm wavelengths will be available for these thousands of young stars. Never before has there been such a wealth of multi-wavelength data for so many young stars. Traditional techniques for studying the physical properties of young stars through their spectral energy distributions have usually focused either on the analysis of many sources using simple observational diagnostics such as colours or spectral indices, or on the analysis of a few sources through the detailed modelling of their full spectral energy distributions. The work presented in the first part of this thesis aims to bridge these two techniques through the efficient modelling of the spectral energy distributions of many young stars. In particular, the technique developed for this work makes it straightforward to find out how well different physical parameters are constrained, whether any parameters are degenerate, and whether additional data would resolve the degeneracies. In the second part of this thesis, a census of intrinsically red sources observed by Spitzer in the Galactic plane is presented, including a catalogue of over 11,000 likely young stellar objects. This sample of sources is the largest uniformly selected sample of young stars to date, and effectively provides a map of the sites of star formation in the mid-plane of the Milky-Way. In parallel, this census has uncovered over 7,000 candidate asymptotic giant branch stars, of which over 1,000 are variable at 4.5 or 8.0 microns.