Recovery of TESS Stellar Rotation Periods Using Deep Learning

We used a convolutional neural network to infer stellar rotation periods from set of synthetic light curves simulated with realistic spot evolution patterns. convolved these real TESS containing minimal intrinsic astrophysical variability allow the learn systematics and estimate despite them. In addition periods, we predict uncertainties via heteroskedastic regression credibility period predictions. most credible half test data, recover 10%-accurate for 46% targets, 20%-accurate 69% targets. Using our trained network, successfully stars literature measurements, even past 13.7-day limit generally encountered by searches using conventional period-finding techniques. Our method also demonstrates resistance half-period aliases. present training introduce software butterpy synthesize star evolution.