AI Verifies 118 Planets in TESS Data

University of Warwick researchers used the AI pipeline RAVEN to verify 118 planets from NASA TESS's first four years of observation data. Among these, 31 were introduced as newly confirmed worlds, and RAVEN was trained on hundreds of thousands of simulations to distinguish real planets from false signals together.

The total analysis covered 2.2 million+ stars, resulting in 2,000+ high-quality candidates and about 1,000 completely new candidates remaining.

Key results:

• Detection of an ultra-short-period planet orbiting in under 24 hours
• Confirmation of a rare Neptunian desert planet
• Discovery of a new multi-planet system orbiting the same star

A companion study estimated that 9-10% of Sun-like stars have close-in planets, with uncertainty reduced by up to 10x compared to Kepler. The occurrence rate of Neptunian desert planets was directly measured for the first time at 0.08%.

RAVEN combines detection, machine learning-based vetting, and statistical validation all at once, enabling direct connection to target selection for follow-up observations.