AI solved Hubble mystery and discovered thousands of “strange” objects

Using Artificial Intelligence, researchers analyzed more than a million images captured by Hubble in just three days.

And if I had to analyze more than 1.7 million images already recorded by the Hubble telescope since its launch in the 1990s? The task is not easy and, knowing this, researchers at the European Space Agency (ESA) created AnomalyMatch, an artificial intelligence model that not only explored almost 100 million sections of the observatory’s data but also identified more than 1300 anomalies.

In fact, the Hubble data represents the greater volume of observational information available for study in the history of astronomy, but the problem is that the amount of data is so vast that it is practically impossible for researchers to analyze it. That’s where AnomalyMatch comes in, which has allowed scientists to explore almost 100 million images in less than three days.

The authors trained the model to teach it to detect foreign objects through pattern recognition — nothing more appropriate, since the model was created to analyze images in a way similar to how our brain processes visual information.

AI in astronomy

The images captured by Hubble represent the largest volume of observational data ever recorded in the history of astronomy. In fact, the amount of information obtained by the famous telescope is so immense that there is not enough time for researchers to investigate what is there.

Registered anomalies are simply objects that look different than expectedand hundreds of them had gone unnoticed by researchers. Many, in fact, defy commonly used classifications — most showed distant galaxies interacting with each other, a process that results in the formation of galaxies with an appearance reminiscent of jellyfish and gas “tentacles.”

In any case, both NASA and the study scientists considered this to be a significant advance in studies of this type. “The discovery of so many undocumented anomalies in Hubble data reinforces the potential of this tool for future studies,” commented Pablo Gómez, a researcher who worked on building the model.