Data Science: NAMS Scientists Utilize Deep Neural Networks to find 301 More Planets
November 22, 2021
NAMS data scientists, Dr. Hamed Valizadegan, Senior Scientist, Machine Learning (USRA) and Miguel Martinho, Associate Scientist, Data Science (USRA) have made some exciting contributions in recent discoveries for the Kepler mission, using a new deep neural network called ExoMiner. Scientists from USRA, NASA, and other institutions recently added 301 newly validated exoplanets to the total exoplanet tally. The planets are the latest to join the 4,569 already validated planets orbiting a multitude of distant stars. The results are published in the Astrophysical Journal. This extraordinary discovery and methodology has also been recognized by USRA, NASA Ames, NASA JPL, and featured on the NASA Exoplanet website in addition to being covered by many other news media outlets.
Deep neural networks are machine learning methods that automatically learn a task when provided with enough data. ExoMiner is a new deep neural network that leverages NASA’s Pleiades supercomputer and can distinguish real exoplanets from different types of imposters, or "false positives." The design is inspired by various tests and properties human experts use to confirm new exoplanets and it learns by using past confirmed exoplanets and "false positive" cases. ExoMiner supplements the work of scientists who comb through data to decipher what is considered a planet, which is a time-consuming task with massive datasets from missions like Kepler.
"When ExoMiner says something is a planet, you can be sure it's a planet," said Hamed Valizadegan, Ph.D., ExoMiner project lead, Manager and Senior Scientist, Machine Learning (USRA). He is also the lead author of the paper published in the Astrophysical Journal. “ExoMiner is highly accurate and in some ways more reliable than both existing machine classifiers and the human experts that it is meant to emulate because of the biases that come with human labeling."
The Astrophysical Journal paper explains that ExoMiner validated the 301 planets using data from the remaining set of possible planets – or candidates – in the Kepler Archive. All 301 machine-validated planets were originally detected by the Kepler Science Operations Center pipeline and promoted to planet candidate status by the Kepler Science Office, but until ExoMiner, no one was able to validate them as planets. None of the newly confirmed planets are believed to be Earth-like or in the habitable zone of their parent stars. They do share similar characteristics to the overall population of confirmed exoplanets in our galactic neighborhood. Miguel Saragoca Martinho, Associate Scientist, Data Science and the main engineer behind implementing ExoMiner (USRA), said, “The modular design of ExoMiner allows us to explain why it says something is planet or false positive. That is peace of mind for domain experts when using a black-box machine classifier such as ExoMiner.”
The research also demonstrates how ExoMiner is more precise and consistent in ruling out false positives and better able to reveal the genuine signatures of planets orbiting their parent stars – all while giving scientists the ability to see in detail what led ExoMiner to its conclusion. As the search for more exoplanets continues – with missions using transit photometry such as NASA’s Transiting Exoplanet Survey Satellite (TESS) and the European Space Agency's upcoming PLAnetary Transits and Oscillations of stars (PLATO) mission – ExoMiner will have more opportunities to prove it is up to the task.
"Now that we've trained ExoMiner using Kepler data, with a little fine-tuning, we can transfer that learning to other missions, including TESS, which we're currently working on," said Valizadegan. "There's room to grow."
