Machine-learning glare removal makes exoplanet imaging up to 15 times more sensitive

Host stars overwhelm direct imaging: even the brightest young giant planets can be about 100,000 times fainter. Published in Nature, the FLIP framework combines physical modelling with machine learning, generating simulated glare for a given star and subtracting the structured starlight from observations. The study reports sensitivity to planetary signals up to 15 times greater than traditional methods. It offers a stronger image-processing route for future searches for Earth-like exoplanets, but the team still needs to reduce computational cost and validate the method across larger datasets and wider observing conditions.