Machine-learning classification of the NEOWISE infrared sky survey
Matteo Paz · Pasadena, California
First Place ($250,000) · Regeneron Science Talent Search · 2025
Machine-learning algorithms applied to the full raw NEOWISE infrared sky survey, sorting objects by small variations in their infrared emission into ten classes.
Processed roughly 200 billion raw entries and identified about 1.5 million new candidate variable objects.
The dataset was public and had been public for years. What was missing was a method that could get through it — which is a reminder that the bottleneck in observational science is often analysis, not access.
- Large-scale infrared survey processing
- Supervised and unsupervised classification
- Variability detection in time-series photometry
- Candidate catalogue construction
- Public archives contain unanalysed results; the contribution is the method that reaches them
- At 200 billion rows, the engineering of the pipeline is part of the science
- A classification into named classes is more useful than a list of anomalies
Archive entries are summaries compiled for study. Read them for how a project was structured, argued, and defended — not as a template to reproduce. Copying someone else's project is the one thing that will end yours.
Compiled by ResearchForge from the Society for Science public award announcement linked on this record. Names, hometowns, project titles and reported results are as published. The "why it worked" and "what to take from this" notes are ResearchForge's editorial reading of that public record — they are not statements by the students and not the judges' rationale. All rights to the original projects remain with their authors; no project materials are rehosted here.