High School Student's AI Discovers 1.5 Million New Celestial Objects in NASA Data

Matteo Paz, an 18-year-old high school senior from Pasadena, California, has upended the astronomical community by uncovering 1.5 million previously unknown celestial objects.[1][2]

Using a custom-built artificial intelligence algorithm, Paz mined archived data from a retired NASA telescope, identifying subtle cosmic phenomena that had gone unnoticed by human researchers and standard software.[5][6]

His breakthrough earned him the $250,000 first-place prize at the Regeneron Science Talent Search, the nation's most prestigious pre-collegiate science competition, and drew formal praise from NASA leadership.[1][4]

The project relied on data from NASA's Near-Earth Object Wide-field Infrared Survey Explorer (NEOWISE). Launched in 2009, NEOWISE spent over a decade scanning the sky in infrared, primarily hunting for near-Earth asteroids and comets.[2][6]

Over its operational lifespan, the telescope amassed nearly 200 billion rows of measurements. While the primary mission was successful, the dataset also captured the shifting heat signatures of "variable objects," which fluctuate in brightness over time.[1][3]

Paz's journey began in the summer of 2022 at the Caltech Planet Finder Academy. Under the mentorship of Davy Kirkpatrick, a senior scientist at Caltech's Infrared Processing and Analysis Center (IPAC), Paz was tasked with manually sifting through a small subset of the NEOWISE data to find variable stars.[1][3]

Rather than analyzing the data by hand, Paz leveraged his background in theoretical math and coding—honed in the Pasadena Unified School District's advanced Math Academy—to automate the process.[2][3]

He developed a machine-learning model dubbed VARnet. The system utilizes Fourier transforms and wavelet analysis to filter out noise and isolate faint, time-based signals in the infrared spectrum.[2][7]

The algorithm achieved sub-millisecond processing speeds per light curve on standard graphics processing units, allowing it to chew through the massive NEOWISE archive with unprecedented efficiency.[6][7]

Out of 1.9 million flagged sources, approximately 1.5 million had no prior record in existing astronomical catalogs. The newly discovered objects include volatile newborn stars, slowly pulsating stars, quasars, and potential supernovae.[5][6]

In a rare feat for a high school student, Paz published his methodology and findings as the sole author of a peer-reviewed paper in The Astronomical Journal.[3][7]

The success of VARnet highlights a massive shift in modern astrophysics: the democratization of big data. Tools that once required institutional supercomputers can now be deployed by independent researchers to decode the universe.[2][5]

Because the algorithm is designed to analyze any time-series data, researchers note it could eventually be adapted for terrestrial applications, such as environmental monitoring, neuroscience, or financial market analysis.[2][6]

Now employed at Caltech's IPAC, Paz plans to publish the complete catalog of the 1.5 million objects with Kirkpatrick, providing a massive new target list for the global astronomical community to study.[3][6]