NASA’s Miner++ AI Brings Machine Digs Into TESS Archive to the Hunt for Nearby Earth-Like Worlds

NASA launched its TESS mission in 2018 to survey nearly the entire sky to find thousands of exoplanets. Scientists have discovered so far over 6,000 such planetary discoveries, much due to the data gathered by NASA's Kepler and its own TESS mission, with the latter discovering over 700 so far. For such an information flood, researchers are turning to the help of machine learning (AI) to identify real transits from the noise.

Deep Learning Aids Exoplanet Discovery

According to NASA, scientists developed ExoMiner, based on deep learning, confirming 370 planets from the data collected from the Kepler Space Telescope. A version of ExoMiner designated ExoMiner++, designed to evaluate data from the TESS space telescope and the original Kepler space telescope data, found ~7,000 planet candidates in the data from the TESS space telescope.

This artificial intelligence works through the shallow dip in each light curve generated from each light signal of the candidate planets to verify whether the signal was an actual planet or an "impostor" such as an eclipsing binary star. A team of researchers created the neural net algorithm that studied ~10,800 light curves from TESS data to discover three planet candidates previously overlooked.

TESS Mission and Future Prospects

TESS surveys nearly the entire sky of bright, nearby stars. Unlike Kepler's narrow field, TESS covers many targets, and its data are compatible. NASA makes these data and AI tools public – for example, ExoMiner++ is open-source on GitHub – so any researcher can join the hunt.

Looking ahead, NASA's upcoming Nancy Grace Roman Telescope (launch mid-2020s) should record tens of thousands of exoplanet transits, all to be released publicly. With open data and powerful AI, the pace of discovery is set to quicken.