“I Mapped the Invisible”: An American High School Student Turns Astronomy on Its Head with His AI That Reveals 1.5 Million Forgotten Space Object

A high school student in California has made a remarkable contribution to astronomy by developing an artificial intelligence model that identified 1.5 million previously undetected space objects. Through his summer research at Caltech, Matteo Paz not only deepened the scientific potential of a retired NASA mission, but also authored a peer-reviewed paper based entirely on his own findings, published in The Astronomical Journal.

AI Meets Deep Space Data

The breakthrough began in the summer of 2022, when Paz joined Caltech’s Planet Finder Academy, a program led by Professor Andrew Howard designed to give students access to advanced astronomical research.

Under the mentorship of Davy Kirkpatrick, a senior scientist at IPAC (Caltech’s Infrared Processing and Analysis Center), Paz tackled an ambitious challenge—analyzing the massive dataset collected by the NASA telescope NEOWISE.

Originally designed to track near-Earth asteroids, NEOWISE captured thermal emissions across the entire sky for more than a decade. While the mission was successful in its primary objectives, it also collected enormous amounts of data on more distant cosmic objects.

These included variable sources that brightened, dimmed, or pulsed over time—signals often associated with phenomena like quasars, eclipsing binary stars, and exploding supernovae.

Building the model from scratch

Faced with what Kirkpatrick described as “creeping up towards 200 billion rows in the table of every single detection,” the Caltech team initially planned to analyze a small slice of the sky manually. Paz, however, had a different vision. With a background in computer science, theoretical math, and programming, he realized the task was ideal for an artificial intelligence solution.

In just six weeks, Paz began developing a Fourier and wavelet-based machine learning model to scan NEOWISE’s massive dataset and flag objects with variable brightness. His understanding of time-domain data analysis, supported by his advanced math education through Pasadena Unified School District’s Math Academy, gave him the tools to work at a level typically reserved for college researchers.

According to Phys.org, the model “began to show some promise” almost immediately. As Paz refined the pipeline, it began detecting subtle variations in infrared light—an indicator of potential new objects or activity in space.

Paz was supported not only by Kirkpatrick but also by other researchers at Caltech, including Shoubaneh Hemmati, Daniel Masters, Ashish Mahabal, and Matthew Graham. These scientists shared their expertise in machine learning techniques and the detection of temporal changes in astronomical objects.

As he progressed, Paz learned how NEOWISE’s data—based on specific observation rhythms—had limitations in identifying certain types of variables. Some objects changed too slowly or flashed too briefly to be captured by conventional analysis. His AI model helped overcome these challenges, flagging variable stars and other objects that showed measurable variation in brightness over time.

These findings were detailed in a paper Paz authored and published in The Astronomical Journal. A complete catalog of the detected variable objects is scheduled for release in 2025, offering new opportunities for astronomers worldwide to study the long-term evolution of distant stars and galaxies.

Expanding beyond astronomy

Although the work is rooted in space research, Paz sees much broader applications.

“The model I implemented can be used for other time domain studies in astronomy, and potentially anything else that comes in a temporal format,” he explained.

For example, the same approach could be applied to analyze stock market charts or track pollution, where periodic patterns play a key role.

Now a paid Caltech employee, Paz continues to work at IPAC while finishing high school. He’s mentoring other students in the Planet Finder Academy and refining his model’s capabilities across different datasets.

His early success highlights the power of machine learning, especially when combined with mentorship and cutting-edge tools. His mentor Kirkpatrick, who grew up in a farming community in Tennessee, sees in Paz the same spark that led him to science.

“If I see their potential, I want to make sure that they are reaching it,” he said. “I’ll do whatever I can to help them out.”