Classifying TESS Star Data using Neural Networks with Attention Mechanisms

NASA's Transiting Exoplanet Survey Satellite (TESS) is one of many satellites that provides astronomers with a plethora of valuable data concerning exoplanets and stars they orbit. TESS collects data by taking repeat observations of the sky every 30 minutes over the course of 30 days, resulting in time series data which allows scientists to observe how stellar brightness changes over time. This data also allows for the analysis of various trends among different star types, resulting in some stars remaining stable while others vary over time due different factors such as star spot rotations, eclipses, or pulsations of the stars themselves. While this data is extremely valuable, individually classifying stars is extremely time-consuming and non-trivial. The goal of this project is to implement a Neural Network with Self-Attention Mechanisms that will learn to accurately classify star data, as well as correctly calculate the periodicity of a specific class of stars. Our approach involves implementing two different neural networks; a Convolutional Neural Network (CNN) as a baseline, and a Perceiver model that contains the self-attention mechanisms.