Planned intervention: On Wednesday April 3rd 05:30 UTC Zenodo will be unavailable for up to 2-10 minutes to perform a storage cluster upgrade.
Published May 19, 2022 | Version v1
Presentation Open

Previously Undiscovered Exoplanets Detected with Deep Learning

Creators

  • 1. Henry M. Gunn High School (PAUSD)

Description

Using deep learning with the Adam optimization algorithm in this research, I detected more than ten previously undiscovered exoplanets, as well as many confirmed and candidate exoplanets, in the Kepler data. Although some of the exoplanet transit signals were evident, others were not as strong and need further evaluation. By using my own code, open source libraries, and deep learning packages TensorFlow, I developed a Python program for exoplanet search. The program first normalizes the transit light curves, trains the deep learning model, folds the light curves to intensify the transit signals, then uses the model to search for exoplanet transits in the Kepler light curves. Among the newly detected exoplanets, most of them are ultra-short period (USP) exoplanets with orbital periods shorter than a day, and the two others are short period exoplanets with periods between 1 to 10 days. Because the Kepler mission lasted for 9.6 years and observed each star for a selected period of time, there are much more Kepler Objects of Interest (KOI) with shorter periods than those with long periods in the NASA database. This may be a reason as to why the orbital periods of the detected exoplanets in this study are shorter than 10 days. Meanwhile, the detection of these new exoplanets, especially the USP exoplanets, can shed light on their kind and expand our views on their planetary systems. Finally, these findings show that artificial intelligence such as deep learning can be an effective technological tool to used detect objects of interest in astronomy big data.

Files

NEW - AYu - ESO - Using Deep Learning to Detect Exoplanets.pdf

Files (2.0 MB)