Presentation #225.01 in the session Fundamental Properties II.
The Transiting Exoplanet Survey Satellite (TESS) has produced millions of photometric light curves containing information about stellar variability across a wide range of timescales. We apply unsupervised machine learning on Lomb-Scargle (LS) periodograms computed from 2-minute cadence light curves observed by TESS to conduct a census of different types of periodic and quasiperiodic stellar variability. By utilizing a one-dimensional deep autoencoder and feature engineering, we produce compressed, low-dimensional representations of the LS periodograms. We use the low-dimensional representations to perform large-scale clustering and novelty detection on TESS data from the first two years. We validate our pipeline using the SIMBAD Astronomical Database, the ASAS-SN Variable Stars Database, as well as catalogs of recent TESS discoveries, such as M-type stars exhibiting complex rotational modulation. The indiscriminate survey produced by TESS offers a unique opportunity to investigate the relationships between variability properties and physical properties such as type, age, metallicity, surface gravity, and mass. Our homogeneous census of periodic and quasiperiodic stellar variability will lead to a better understanding of the underlying mechanisms.