Let’s start with the figure that makes the rest of the story difficult to imagine: 200 billion. That’s about the number of rows of data a machine learning program created by a Pasadena high school student had to process. Each row is a moment when a retired NASA telescope captured a flash of infrared light somewhere in the sky. Stacked up over more than a decade of scanning, they add up to a table no one could read by hand. The person who wrote the code for review is Matteo Paz. When the flickers were sorted, their program flagged about 1.5 million potential new objects. In 2025, that work earned him first prize in a national scientific competition. We want to answer the obvious questions: what the code actually did, why “potential” is the word that matters, and how a teenager ended up doing it inside one of Caltech’s research centers. What is 200 billion raw infrared data entries? The data came from NEOWISE, an infrared space telescope that scanned the entire sky over and over again for more than ten years before being retired. Each pass added more detections. Paz’s mentor, IPAC senior scientist Davy Kirkpatrick, expressed the scale clearly: The count was inching toward 200 billion rows in the table for every detection the survey had made. Raw is the key word. This was not an ordered catalog of stars and galaxies. They were individual snapshots, the kind of firehose that most projects cut out before they even start, because the entire broadcast is so unwieldy. The value hidden in it changes over time. If you look at the same patch of sky over and over again, some points of light remain stable and others brighten and dim. The ones that vary are usually the interesting ones. Kirkpatrick’s original plan for the summer was modest: take a small patch of sky and look for variable stars. What was the code actually doing? The Paz program was built to capture small differences in infrared brightness through those repeated measurements. He described the method in a peer-reviewed paper in The Astronomical Journal, which he wrote alone. It combines signal processing mathematics with a neural network that learned to distinguish one type of blink from another. The result was not just a lot of “this one changes.” The program classified the sources into a small set of categories, separating fixed sources from the various ways light from an object can vary. Some things pulse on their own. Some dim on a schedule because a companion star passes in front of them. Some shine once and fade away. Sorting candidates into groups is what turns a raw list into something a scientist can actually search for, because a researcher looking for pairs of eclipsing stars doesn’t want to examine every distant, twinkling galaxy to find them. What catches our attention is the scope that Paz affirms for the focus beyond astronomy. He has said that the model can be used for other time-domain studies in astronomy and, potentially, for anything else that comes in a temporal format. That “potentially” is yours and it is worth keeping. Anything that arrives as a flow of measurements over time is a candidate in principle. Whether the model is actually useful in, say, financial data or sensor readings is an open question, not a proven result. 1.5 million “potentials” are not 1.5 million discoveries. The distinction that the headline number may obscure is this: the 1.5 million potential new objects are candidates, not confirmed finds. Some may be already cataloged sources. Some will be false alarms, data quirks rather than actual variable objects. The catalog is a list of things that deserve a closer look, and the search is a separate job. This is not a knock on the job. This is how surveys of this type work. A program designed to track 200 billion measurements is valuable precisely because it reduces an impossible search to a manageable set of clues. Among the candidates are objects whose brightness changes over time, and the Society for Science summary of Paz’s census of infrared variable objects lists supermassive black holes, newborn stars and supernovae among them. Confirming any of them requires follow-up observation and analysis. To treat all 1.5 million as finished discoveries would be to overstate what the catalog is and to underestimate why a leaked list of tracks is actually useful. The paper describing the method was published in November, and Paz and Kirkpatrick have said they plan to publish the full catalog so others can begin follow-up work. How did a high school student end up doing this? The short answer is a chain of outreach programs rather than a single lucky break. Peace’s route involved Caltech public lectures and a summer research program that paired him with Kirkpatrick in 2023. The project was to last six weeks. Paz had something bigger in mind from the beginning, and on the first day he told Kirkpatrick that he was considering working on an article, a much bigger goal than six weeks would normally allow. According to him, the mentor did not discourage him. What the tutoring achieved, according to Paz, was less technical instruction than space to think. He left room for the ambitious version of the project to survive, and the six weeks became the beginning of a much longer work. What happens now with a catalog of this size? The prize, first place and its $250,000 prize in the 2025 Regeneron Scientific Talent Search, seems more like a starting line than a finish line. Paz now works at IPAC as a Caltech employee and the catalog he created is still a set of tracks waiting to be reviewed. Which of the 1.5 million candidates are real and new, which are already known and which are just noise, are questions that the statement aims to convey to the community at large. A list of candidates is an invitation. What astronomers do with it, and what the method does when targeting the next avalanche of survey data, is work that has not yet been done. About this articleThis article is for general information and reflection. It is not professional advice. For your specific situation, consult a qualified professional. Standards Space Daily articles are edited and fact-checked before publication. We use artificial intelligence tools in the newsroom. See our editorial standards and masthead.
