ExoMiner, NASA's new deep neural network that leverages the Pleiades supercomputer, recently helped scientists to add a whopping 301 new exoplanets to the total exoplanet tally of 4,569.

Deep neural networks, such as ExoMiner, are machine learning methods that learn a task when provided enough data. In ExoMiner's case, it can detect real exoplanets from imposters or false positives by learning past confirmed exoplanets. It was designed based on various tests and properties used in confirming new exoplanets.

 ExoMiner Deep Neural Network Finds 301 New Exoplanets Added to the Total Count of NASA's Kepler Mission
(Photo: Wikimedia Commons)
Artist Concept Planetary System (NASA/JPL-Caltech)

ExoMiner Deciphers What is and What is Not a Planet

According to NASA, ExoMiner helps scientists to comb through data and decipher which cosmic bodies are a planet and which are not. The data NASA gathered using the Kepler spacecraft and K2 helped in training the new deep neural network to find which of the thousands of stars in its field of view could potentially host multiple exoplanets.

Exoplanet scientist Jon Jenkins of NASA's Ames Research Center said that ExoMiner is not a black box, unlike other exoplanet-detecting machine learning programs. It can easily explain which features of the data have led it to reject or confirm a planet.

ExoMiner project lead and Universities Space Research Association machine learning manager Hamed Valizadegan added that when ExoMiner says that something is a planet, it is sure that it is indeed a planet. More so, the new deep neural network is highly accurate and more reliable than both existing machine learning programs and human experts that it is meant to mimic because of the biases that humans have when labeling.

"These 301 discoveries help us better understand planets and solar systems beyond our own, and what makes ours so unique," Jenkins explains as NASA quoted.

ALSO READ: Earth's Atmosphere Seems to be 'Unique', No Other Exoplanets Could Compare

Confirmed vs Validated Exoplanet

In the press release by the Universities Space Research Association scientists, they noted that a confirmed exoplanet has features that can only be explained by a planet after undergoing different observation techniques. On the other hand, scientists use statistics to tell how likely or unlikely it is to be a planet based on the data before it becomes a validated exoplanet.

In their paper, titled "ExoMiner: A Highly Accurate and Explainable Deep Learning Classifier that Validates 301 New Exoplanets"  published in the Astrophysical Journal, researchers explained that ExoMiner validated 301 exoplanets using existing data from possible or candidate planets in the Kepler Archive.

Miguel Saragoca Martinho, the main engineer behind ExoMiner, said that the modular design of the deep neural network enabled them to explain why a candidate planet is an exoplanet or false positive.

Furthermore, researchers demonstrate in their paper how ExoMiner is more precise at ruling out imposters or false positives and revealing the genuine characteristics of orbiting planets in distant stars.

Researchers believe that ExoMiner will have more opportunities to prove its capacity with the data coming from NASA's Transiting Exoplanet Survey Satellite (TESS) and ESA's PLAnetary Transits and Oscillations of stars (PLATO) with just a little fine-tuning.

 RELATED ARTICLE: NASA Discovered "Hot Jupiter" Exoplanet, Bigger But Less Massive Than Solar System's Largest Planet

Check out more news and information on Exoplanets in Science Times.