Title: SDSS-IV MaStar: Data-driven Parameter Derivation for the MaStar Stellar Library

Abstract: The MaNGA Stellar Library (MaStar) is a large collection of high-quality empirical stellar spectra designed to cover all spectral types and ideal for use in the stellar population analysis of galaxies observed in the Mapping Nearby Galaxies at Apache Point Observatory (MaNGA) survey. The library contains 59,266 spectra of 24,130 unique stars with spectral resolution $R\sim1800$ and covering a wavelength range of $3,622-10,354$ \r{A}. In this work, we derive five physical parameters for each spectrum in the library: effective temperature ($T_{eff}$), surface gravity ($\log g$), metalicity ($[Fe/H]$), micro-turbulent velocity ($\log(v_{micro})$), and alpha-element abundance ($[\alpha/Fe]$). These parameters are derived with a flexible data-driven algorithm that uses a neural network model. We train a neural network using the subset of 1,675 MaStar targets that have also been observed in the Apache Point Observatory Galactic Evolution Experiment (APOGEE), adopting the independently-derived APOGEE Stellar Parameter and Chemical Abundance Pipeline (ASPCAP) parameters for this reference set. For the regions of parameter space not well represented by the APOGEE training set ($7,000 \leq T \leq 30,000$ K), we supplement with theoretical model spectra. We present our derived parameters along with an analysis of the uncertainties and comparisons to other analyses from the literature.