In recent decades, astronomy has undergone a major transformation thanks to advances in instrumentation, sensors and computing power. Ground and space telescopes have undertaken extensive multispectral surveys, generating enormous amounts of data. Observations at different wavelengths, extended time series and catalogs listing billions of sources are now commonplace.
The flip side is that this exponential growth of data has made manual analysis impractical and limited the use of traditional methods. Even automated techniques struggle to handle high dimensionality, complex noise, and unusual phenomena. This is why artificial intelligence techniques, particularly machine learning, have become essential for extracting patternsclassify objects, detect transient events and explore vast astronomical databases.
A recent example is the development of an AI model capable of detecting anomalies in observational data. This model can identify signals that deviate from expected behavior. Astronomers applied it to data from the Hubble telescope, which allowed them to discover new phenomena that had gone unnoticed during traditional manual analyses. This type of approach demonstrates how AI contributes to data analysis and scientific discovery.
Revolution in astronomy
Table of Contents
At the beginning of the last century, the main problem in astronomy was the scarcity of data, which prevented exhaustive analysis. However, in recent decades the situation seems to have reversed, astronomy now has such a quantity of data that manual analysis has become impossible. This phenomenon is due to technological advances made in the fields of detectors, telescopes and computer infrastructures.
Surveying the sky at multiple wavelengths, continuous observations, and long-duration space missions have generated massive volumes of observational data.
Faced with the increase in the volume and complexity of data, traditional methods of information extraction prove insufficient. The need to identify trends, rare events and non-linear correlations has created a high demand for artificial intelligence and machine learning techniques. These approaches enable the analysis of large volumes of data with greater precision, while being automated and scalable.
The model AnomalyMatch
A new artificial intelligence model has been developed to search for anomalies across the entire Hubble Space Telescope archive. This model, presented in a study published in the journal Astronomy & Astrophysicsis baptized AnomalyMatch. It was designed for identify rare or unexpected patterns without resorting to previous classifications. The Hubble archive was an ideal target because it contains continuous observations spanning several decades.
The model was trained using observational data spanning approximately 35 years of Hubble observations. Instead of analyzing entire images, the neural network was trained on nearly 100 million small image fragments, each containing just a few pixels, allowing it to capture subtle local structures. This strategy reduced the computational cost and increased the sensitivity of the system to statistical deviations from the dominant population of astronomical objects.
Discoveries in AI
By applying AnomalyMatch at the Hubble telescope archives, researchers have identified more than a thousand objects classified as abnormal in less than three days of processing. Some of these detections have never been published before. This demonstrates that vast volumes of archived data still conceal many unexplored phenomena. Most anomalies are associated with energy eventssuch as galaxy mergers.
Besides galactic mergers, the model revealed classes of rare objects, like “jellyfish” galaxies. These galaxies are so named because of their long tails of star-forming gas, protoplanetary disks, and gravitational lensing systems. Some detections do not fit any known category, suggesting new physical regimes or new stages of evolution.
Benefits
Using AI in astronomy offers advantages over traditional methods, particularly in terms of speed and scalability. Machine learning models can analyze massive volumes of data in timeframes not possible with traditional approaches. These models also identify complex patterns and nonlinear correlations without requiring explicit rules.
In addition to its speed, the flexibility of AI facilitates its application to various astrophysical problems, from identifying potentially habitable exoplanets to improving images of black holes. These methods have already demonstrated their usefulness as complementary tools to traditional analyses. It is important to emphasize, however, that AI does not replace scientists; rather, it constitutes a tool for understanding the complexity and scale of modern astronomy.
News references
O’Ryan and Gómez 2026, Identifying Astrophysical Anomalies in 99.6 Million Images from the Hubble Telescope Archive Using AnomalyMatch, Astronomy and Astrophysics
