A Neural Network for Identifying Exoplanets

Abstract: The Transiting Exoplanet Survey Satellite (TESS) has now been in operation for slightly more than two years, covering both the Northern and Southern hemispheres once. The TESS team uses the Science Processing Operations Center pipeline and the Quick Look pipeline to generate alerts for follow-up. Combined with other community efforts, over two thousand planet candidates have been discovered, with tens confirmed as planets. We present Udva, our pipeline that is complementary to these approaches. Udva employs a combination of transit detection, supervised machine learning, and detailed vetting to identify a few planet candidates that were missed by previous searches with high confidence. We find shallow transits with a high signal-to-noise ratio (SNR) that may represent more than one transit. Future work will include approaches to enhance stages that have been conservatively abandoned because they lacked one or two datums in order to boost the yield.

Keywords: planetary systems, planets and satellites: detection, techniques: photometric, methods: data analysis