AI Discovers Rare Celestial Objects in Hubble Images

For more than three decades, the Hubble Space Telescope has captured millions of images and gathered an immense trove of astronomical data, unlocking the mysteries of the universe. Analysing these images in detail, however, has traditionally been a painstakingly slow process. Now, researchers at the European Space Agency (ESA) have developed an artificial intelligence (AI) system capable of scanning nearly 100 million Hubble images in just two and a half days, identifying over 1,300 unusual and rare celestial objects.

The Hubble archive is one of the largest repositories of astronomical imagery in the world. It serves as a critical resource for understanding galaxy evolution, gravitational phenomena, and cosmic structures. Yet, the sheer volume of data has long posed a significant challenge. Even teams of scientists or volunteers cannot quickly sift through millions of images to detect subtle anomalies. Current astronomical surveys generate information at a rate far exceeding human analytical capacity.

To address this, ESA researchers David O’Ryan and Pablo Gomez created a neural network-based AI system named ‘Anomalimatch’. Modelled on the human brain, Anomalimatch was trained to recognise rare cosmic phenomena, including jellyfish galaxies and gravitational arcs. Following extensive training, the AI scanned the Hubble Legacy Archive’s 100 million images in just 60 hours, producing a preliminary catalogue of unusual objects.

Upon verification by astronomers, the AI-generated list confirmed over 1,300 genuinely unusual cosmic objects. Remarkably, more than 800 of these had never been documented in any scientific record. Among the notable discoveries are colliding galaxies, whose shapes are distorted by mutual gravitational forces, sometimes forming elongated trails of stars. Several jellyfish-like galaxies were identified, emitting gaseous tails reminiscent of their aquatic namesake. Dozens of objects were found that defy existing astronomical classification altogether.

According to the researchers, future data from ESA’s Euclid mission and NASA’s Nancy Grace Roman Space Telescope will be immense, making AI-assisted analysis essential. Importantly, AI is not replacing human expertise. As Gomez explains, “The AI acts as an untiring assistant. It rapidly scans and flags suspicious objects, but final verification is made by astronomers.” This synergy of machine speed and human judgment is expected to accelerate the uncovering of the universe’s deepest mysteries.

Summary of Key Discoveries by Anomalimatch

The deployment of AI in astronomical research demonstrates a powerful new method to navigate the data deluge from modern space telescopes, combining computational efficiency with human analytical expertise.