Astronomers at the University of Warwick have deployed artificial intelligence to discover 118 previously undetected exoplanets hiding in NASA's Transiting Exoplanet Survey Satellite (TESS) data, demonstrating how machine learning can extract discoveries from archives that traditional analysis methods missed.
The breakthrough came through RAVEN (RAnking and Validation of ExoplaNets), an automated pipeline that analyzed observations of 2.2 million stars from TESS's first four years of operations. Beyond the 118 validated planets, RAVEN identified over 2,000 high-quality planet candidates—nearly 1,000 entirely new—that warrant follow-up investigation.
Dr. Marina Lafarga Magro, the study's lead author and postdoctoral researcher at Warwick, explained that the AI system distinguishes genuine planetary transits from false positives by recognizing patterns across hundreds of thousands of simulated scenarios—a scale of analysis impossible for human researchers to conduct efficiently.
The types of planets RAVEN uncovered reveal systematic gaps in traditional detection methods. The AI found ultra-short-period planets completing orbits in under 24 hours, worlds that create subtle signals easily lost in data noise. It also identified planets in the so-called "Neptunian desert," a theoretically predicted scarcity zone where such worlds shouldn't exist—yet apparently do.
Close-orbiting multi-planet systems presented another category where human analysis struggled. Dr. Andreas Hadjigeorghiou, who led pipeline development, noted these complex systems produce overlapping transit signatures that confuse conventional algorithms but which RAVEN's pattern recognition disentangled.
The methodology represents a paradigm shift in exoplanet research. Rather than waiting for new telescope data, RAVEN demonstrates that treasure troves of undiscovered worlds already exist in archived observations, overlooked by earlier analysis passes. NASA's TESS mission alone has generated petabytes of stellar brightness measurements—far more than astronomers can manually scrutinize.
Dr. David Armstrong, Associate Professor at Warwick and senior co-author, emphasized the efficiency gains:
