Presentation #203.03 in the session “Machine Learning in Astronomy: Transient Discovery with Machine Learning (Meeting-in-a-Meeting)”.
Machine learning research has made major strides in the past decade and led to advancements in wide-ranging fields. We are using deep learning to increase the speed and sensitivity of exoplanet searches. Our deep convolutional neural network classifier, Astronet, is trained to determine whether any given signal detected in transiting exoplanet surveys like Kepler or TESS is due to a real planet or is some type of false positive. Our most recent version of Astronet is currently being used in the TESS Quick Look Pipeline, one of the methods used to produce the official TESS planet candidate catalog. Meanwhile, we are exploring whether deep learning can increase the sensitivity of radial velocity observations to smaller planets. Currently, radial velocity planet searches are limited by astrophysical noise from stellar magnetic activity. We have prototyped a machine learning solution by training neural networks to recognize and remove this noise signal. Our initial tests show promise for increasing the sensitivity of these observations.