High School Student's AI Model Uncovers 1.5 Million Hidden Cosmic Objects in NASA Archive

A California teenager has transformed a summer astronomy project into a peer-reviewed scientific breakthrough, utilizing artificial intelligence to uncover 1.5 million previously undocumented cosmic phenomena. Matteo Paz, a student at Pasadena High School, achieved the milestone by applying custom machine-learning algorithms to a massive, publicly available NASA dataset.[1][2]

The discovery centers on archival data from the Near-Earth Object Wide-field Infrared Survey Explorer (NEOWISE). Launched in 2009, the space telescope spent more than a decade scanning the sky in infrared light, cataloging everything from near-Earth asteroids to distant galaxies. Over its operational lifespan, NEOWISE accumulated a staggering dataset comprising nearly 200 billion rows of measurements.[1][4]

While the NEOWISE archive is a goldmine for astrophysicists, its sheer volume presents a formidable challenge. Traditional software and manual human review are sufficient for tracking bright, obvious objects, but they frequently miss faint, transient, or slowly varying light sources. These subtle variations often represent some of the most intriguing phenomena in the universe, such as brown dwarfs or distant active galactic nuclei.[3][4]

Paz initially joined the Planet Finder Academy—a program designed to immerse students in real-world astronomical research—in the summer of 2022. Working under the mentorship of Caltech scientist Davy Kirkpatrick at the Infrared Processing and Analysis Center (IPAC), the original plan was for Paz to manually study a small subset of the NEOWISE data.[1][2]

However, recognizing the limitations of manual analysis, Paz pivoted to a much more ambitious, computationally driven approach. Drawing on a strong background in theoretical mathematics and coding, he spent six weeks designing and training an automated machine-learning pipeline capable of processing the entire archive.[1][5]

The resulting AI model utilized advanced mathematical techniques, specifically Fourier transforms and wavelet analysis. These tools are highly effective at isolating time-based signals, allowing the algorithm to detect faint fluctuations in the infrared spectrum that standard sampling methods would overlook.[1][3]

The model began showing promise almost immediately, flagging objects whose brightness changed too subtly or unpredictably for conventional detection. By automating the search, the AI could sift through billions of data points in a fraction of the time it would take a human team, operating with a level of precision that eliminated human fatigue.[1][3]

The final output of the pipeline was staggering: a map of 1.5 million previously invisible or unnoticed cosmic objects. The findings were so robust that they formed the basis of a formal paper, recently published as a peer-reviewed breakthrough in The Astronomical Journal.[1][2]

Beyond the astronomical value of the newly discovered objects, the achievement highlights a profound shift in how modern science is conducted. For decades, processing datasets of this magnitude required access to massive institutional supercomputers and teams of specialized researchers.[4][5]

Today, the democratization of artificial intelligence and open-source machine learning libraries has fundamentally lowered the barrier to entry. Tools that were once the exclusive domain of elite laboratories can now be run on consumer hardware, empowering citizen scientists and students to make genuine contributions to frontier research.[5]

This paradigm shift is particularly relevant as the scientific community grapples with a growing data deluge. Next-generation observatories, such as the Vera C. Rubin Observatory, are expected to generate petabytes of data annually. AI-driven pipelines like the one developed by Paz will be essential for triaging this information and identifying targets for follow-up observation.[2][5]

NASA leadership formally recognized Paz's contributions earlier this year, underscoring the agency's commitment to open data initiatives. By making archives like NEOWISE publicly accessible, space agencies provide the raw material necessary for AI-enabled discoveries by the broader public.[1][4]

The 1.5 million newly identified objects now present a fresh challenge for the astronomical community: classifying them. Researchers anticipate that the catalog will yield a wealth of new variable stars, eclipsing binaries, and potentially entirely new classes of celestial bodies that have never been documented.[3][5]

Ultimately, this breakthrough serves as a powerful proof of concept for the future of scientific inquiry. It demonstrates that when open data is paired with accessible artificial intelligence, the next major discovery can come from anywhere—even a high school classroom.[1][5]