Research

Following up on revolutionary 2018 research in which Google's neural network had first identified planets in Kepler telescope data, I'm advancing exoplanet discovery boundaries with machine learning. While the original Google system had previously been restricted to analysis of already-identified planetary signals from NASA's supercomputer processing, my method is entirely different. I am developing and tuning a deep neural network powered by GPUs that can efficiently detect ultra-short and short-period planets from raw data using available computational power rather than requiring supercomputer facilities (Wang et al. 2024, MNRAS). My current work involves scaling this technology to find potentially habitable small planets that previous methods would have missed. The search for Earth-like planets remains challenging, only some 50 such small planet candidates with radii larger than 1.4 Earth radii have been identified in Kepler's large dataset. My deep learning system aims to break this ceiling, finding smaller habitable planet candidates with radii smaller than 1.4 Earth radii that are hard for traditional detection methods to identify. Why does this matter? If we can identify more of these small, potentially habitable worlds, we'll start to be able to do robust statistical studies of planets that might harbor life. Right now, our sample size is just too small to draw solid conclusions about how common Earth-like planets really are in our galaxy. This research could help us answer one of the biggest questions in astronomy: are worlds like ours rare or common?