Loading…
Loading contentLoading…
Loading contentA modern survey sees more of the sky in a night than an astronomer could study in a lifetime — millions of alerts, billions of objects, petabytes of images. This is how astronomy keeps up: the machine-learning methods that classify and discover at scale, the brokers that triage the alert stream in real time, and the data engineering that keeps it all honest.
The techniques astronomy borrows and adapts — classification, regression, clustering, representation and self-supervised learning, foundation models, and anomaly detection.
7 entriesWhere ML meets the sky — galaxy morphology, supernova classification, photometric redshifts, transit and lens finding, source extraction, and real-time alerts.
7 entriesThe community systems that classify the survey alert stream in real time — ALeRCE, ANTARES, Fink, and Lasair.
4 entriesWhat makes ML work and stay trustworthy — training and benchmark datasets, feature extraction, and honest model evaluation.
4 entriesSorting galaxies by their shape — spiral, elliptical, irregular, merging — from survey images. One of the earliest large-scale meetings of astronomy and machine learning, building on the labels gathered by citizen-science projects like Galaxy Zoo to train automatic classifiers for surveys too large to inspect by eye.
Estimating how far away a galaxy is from its brightness in a few broad colour bands, without taking a full spectrum. It is far less precise than a spectroscopic redshift but can be done for the hundreds of millions of galaxies in an imaging survey, underpinning weak-lensing and large-scale-structure cosmology.
Classifying the flood of alerts that a survey like Rubin issues — millions each night when something on the sky changes — quickly enough to catch the fleeting events worth following up. It is the problem the community alert brokers exist to solve.
Finding and measuring the individual stars and galaxies in an astronomical image, and separating real sources from noise and artefacts — the first step of nearly every imaging pipeline. Machine-learning methods increasingly complement the classical algorithms, especially in crowded or blended fields.
Searching survey images for the rare, distinctive arcs and rings of strong gravitational lensing — where a foreground mass bends the light of a background galaxy. The lenses are rare enough, and the images numerous enough, that automated finders are the only way to build large samples.
Deciding what kind of exploding star a transient is — often from its light curve alone, before or without a spectrum. Fast, automatic classification is essential when a survey finds thousands of supernovae a night and only a few can be followed up in detail.
Picking the tiny, periodic dips of an exoplanet transit out of a noisy stellar light curve — and telling a real planet from the many kinds of false positive. Machine learning now helps sift the enormous light-curve archives of transit surveys for the faintest candidates.
Each ML method, application, workflow, and alert broker is a first-class knowledge-graph entity resolved through the Scientific Data Engine, reusing the Rubin Observatory and alert stream, the alert systems, the photometry and lensing methods, the galaxy morphologies, the transit method, the Type Ia supernova class, the redshift concept, and the reproducibility and data-pipeline practices already in the graph. Curated from NASA, NOIRLab, and the Rubin/LSST community. Benchmark datasets and brokers are named only where real. See source quality.