Kennedi White
MIT Department: Aeronautics and Astronautics
Faculty Mentor: Prof. Kerri Cahoy
Research Supervisor: Christine Page
Undergraduate Institution: Howard University
Hometown: Savannah, Georgia
Website: LinkedIn
Biography
Kennedi J. White is a Savannah, Georgia native, majoring in Physics and English and minoring in Mathematics at Howard University. Growing up, she was infatuated with Greek, Roman, and Egyptian mythology, making her actively seek out fiction books with space and star imagery. These origins of scientific literature led Kennedi to aspire to earn an MS in science communication and a PhD to become a planetary scientist focusing on planetology and planetary atmospheres. Kennedi has previously interned with NASA Goddard and NASA JPL in the realm of lunar and planetary atmospheres. She is currently interning at MIT’s STAR lab, working on exoplanet detection. Kennedi aspires to have a career in research and scientific journaling to bridge the scientific knowledge gap between scientists and the general public. Outside of academics, Kennedi enjoys writing science fiction and fantasy novels, skateboarding, and making music.
Abstract
A.I. x Astro : Machine Learning Models for Exoplanet Detection
Kennedi White1, Christine Page2 and Kerri Cahoy2
1Department of Physics and Astronomy, Howard University
2Department of Aeronautics and Astronautics, Massachusetts Institute of Technology
Direct imaging techniques have contributed to an increase in exoplanet detection over the
past decade. Successful direct imaging instruments and techniques require the ability to suppress the observed object’s image, scattered light, diffraction pattern, and imperfections. Currently researchers are relegated to distinguishing exoplanets in a post analysis phase with no guarantee that the body being observed is an exoplanet or noise. Distinguishing which figures in an image is a planet or another celestial body with the current methods exhaust resources with little reward. Advancements in artificial intelligence (AI) have allowed for more effective methods to detect and analyze exoplanets as a means of characterization. Our research group is looking at ways that AI can be used to distinguish which parts of an image taken with a telescope and coronagraph are noise and which could be an exoplanet. A convolutional neural network (CNN) was trained on raw and labeled direct images to build its ability to identify exoplanets. The model proved to be significantly effective when trained and tested on synthetic images. On real images, the model over identifies exoplanets and other objects of interest. The implementation of this model can help with detecting regions thinterest in real time to maximize the effectiveness of future observations.
