Article illustration 3

In an extraordinary feat blending artificial intelligence and astronomy, 17-year-old Matteo Paz from Pasadena has identified 1.5 million previously overlooked cosmic objects within NASA’s NEOWISE infrared data archive. What began as a high school summer project at Caltech’s Planet Finder Academy has evolved into peer-reviewed research published in The Astronomical Journal, challenging conventional approaches to astronomical data analysis.

The Data Deluge Challenge

NASA’s NEOWISE telescope, originally tasked with asteroid detection, amassed over 200 billion infrared observations across a decade—a dataset so vast that manual analysis was impractical. "We were creeping up toward 200 billion rows of observations," noted Davy Kirkpatrick, Caltech/IPAC senior scientist and Paz’s mentor. Traditional methods could only sample fragments, leaving transient phenomena like quasars, binary stars, and supernovae hidden in the noise.

Building the Anomaly Hunter

Paz, leveraging skills honed in Pasadena Unified’s Math Academy, engineered an automated detection pipeline in just six weeks. His model combined:
- Fourier transforms to identify periodic brightness variations
- Wavelet analysis to capture non-repeating transient signals
- Machine learning classifiers to filter meaningful patterns from noise

The anomaly extraction pipeline developed by Paz (Credit: The Astronomical Journal)

The system specifically targeted temporal anomalies—objects exhibiting subtle brightness changes too slow, brief, or irregular for human analysts to consistently identify. "The model began to show promise almost immediately," Kirkpatrick told Phys.org. "As Paz refined it, the results kept getting more interesting."

Scientific Impact and Interdisciplinary Potential

Paz’s pipeline detected over 1.5 million variable sources, creating the largest catalog of its kind from NEOWISE data. This treasure trove will enable:
1. Targeted follow-ups by observatories like Vera Rubin and JWST
2. New studies on stellar evolution and galactic dynamics
3. Validation of theoretical models for cataclysmic cosmic events

Crucially, the methodology transcends astronomy. Paz notes the pipeline’s architecture—designed for temporal pattern detection—could revolutionize fields from finance (market fluctuation prediction) to neuroscience (identifying neural signal patterns).

Democratizing Discovery

Now a paid Caltech researcher while finishing high school, Paz embodies a shift in scientific accessibility. His work proves that open data + algorithmic ingenuity + accessible compute resources can democratize discovery. As AI tools lower barriers to extracting insights from big data, we may see more breakthroughs emerging from unexpected quarters—transforming not just what we see in the cosmos, but who gets to map it.

Source: Paz, M. et al. (2025). "A Machine Learning Pipeline for Anomaly Detection in NEOWISE Temporal Data." The Astronomical Journal. Original reporting via Daily Galaxy