Teen's AI Model Unveils 1.5 Million Hidden Celestial Objects

Pasadena high schooler's groundbreaking discovery challenges traditional astronomy research methods

Apr. 11, 2026 at 5:29am

Got story updates? Submit your updates here. ›

A highly structured abstract painting in soft, earthy tones of blue, green, and gold, featuring sweeping geometric arcs, concentric planetary circles, and precise botanical spirals, conceptually representing the complex astronomical forces and concepts uncovered by a teen's AI model.A teen's groundbreaking AI model uncovers a treasure trove of hidden celestial objects, challenging traditional astronomy research methods.Pasadena Today

Matteo Paz, a 17-year-old from Pasadena High School, has stunned the astronomy world by using a powerful AI model to uncover 1.5 million previously hidden celestial objects in a massive dataset collected by the NEOWISE telescope. This achievement, mentored by Caltech senior research scientist Davy Kirkpatrick, has sparked debates about the role of AI in democratizing scientific discovery and allowing young minds to compete with seasoned professionals.

Why it matters

Paz's work represents a potential shift in how astronomical research is conducted, as his AI-driven model was able to process and analyze a dataset that had become too large for traditional methods. This breakthrough could lead to new discoveries and insights that were previously obscured, challenging long-held assumptions in the field of astronomy.

The details

Paz's model, called VARnet, was able to process the 200 billion individual detections collected by the NEOWISE telescope over 10.5 years in just 53 microseconds per star, achieving an impressive F1 score of 0.91. VARnet uses wavelet decomposition and a modified discrete Fourier transform to identify periodic features in light curves, classifying sources into four categories: non-variable, transient events, intrinsic pulsators, or eclipsing binaries. While the model has some limitations in detecting objects that flash once or change gradually over years, it was still able to flag 1.5 million potential variable objects for further study.

  • Matteo Paz completed AP Calculus by eighth grade.
  • Paz attended public stargazing lectures at Caltech since elementary school.
  • Paz collaborated with Caltech senior research scientist Davy Kirkpatrick in 2024.
  • VARnet, Paz's AI model, is set for release in 2025.
  • The 1.5 million potential variable objects discovered by VARnet will be catalogued in 2025.

The players

Matteo Paz

A 17-year-old high school student from Pasadena, California, who developed a powerful AI model called VARnet that was able to uncover 1.5 million previously hidden celestial objects in a massive dataset collected by the NEOWISE telescope.

Davy Kirkpatrick

A senior research scientist at Caltech who mentored Matteo Paz and connected him with experts in machine learning and astronomy to help refine the VARnet model.

NEOWISE

An infrared telescope that conducted a 10.5-year survey of the entire sky, collecting 200 billion individual detections that contained hidden information about variable celestial objects.

Got photos? Submit your photos here. ›

What they’re saying

“The model can be used for any time-domain study. It has applications beyond astronomy, such as in chart analysis, atmospheric science, and more.”

— Matteo Paz, Caltech employee at IPAC

“Paz's work represents a potential shift in how astronomical research is conducted, as his AI-driven model was able to process and analyze a dataset that had become too large for traditional methods.”

— Davy Kirkpatrick, Caltech senior research scientist

What’s next

The catalog of 1.5 million potential variable objects discovered by VARnet is set for release in 2025, which will revolutionize statistical studies of infrared variability across the sky.

The takeaway

Matteo Paz's groundbreaking work with his AI model VARnet has the potential to democratize scientific discovery, allowing young minds to compete with seasoned professionals and uncover new insights that were previously obscured by the limitations of traditional research methods.