Abstract

As a suite of large-sky spectroscopic surveys comes online, automated spectral classification techniques are needed more than ever. For white dwarfs—the evolutionary endpoint of the vast majority of stars—spectral classification is vital for understanding their properties, yet still almost exclusively done by eye. Upcoming surveys will return of order 10^5 white dwarf spectra, highlighting the need for automated tools that are fast, but do not miss rare or unique objects, as supervised machine learning models often do. We present the use of dimensionality reduction, an unsupervised method, on white dwarf spectra from the DESI EDR . I will outline the theory behind dimensionality reduction, as well as results showing its effectiveness in classifying white dwarf spectra. I will also discuss two extensions of the technique: the highlighting of spectral regions, and its use in a pseudo-supervised manner.