AI and ML in Astronomy

Which image above shows a spiral galaxy, and which an elliptical galaxy? This question may be simple enough to answer based on the images above, but what if you were given the following images and asked to differentiate between the two types of galaxies? The question now becomes much more challenging and interesting!

Credit: ESA

We would have to start thinking about the differences between the two types of galaxies, such as having a flattened disk and spiral arms (spiral galaxies) vs. being spherical with no spiral arms (elliptical galaxies).

If we had to manually examine and classify the billions of galaxies visible today through powerful telescopes, this would be a tedious and tremendously time-consuming task, even with expert help! How can this be made faster and easier for researchers?

Math to the rescue once again! Artificial intelligence and machine learning (AI/ML) models can be effectively trained to do this task; they can classify tens of thousands of galaxies in seconds or minutes! AI/ML models, in turn, are built upon mathematics:

• Galaxy images are represented as grids of pixels, and transformed into a mathematical input form usable by the AI/ML models.

• Math operations are used to extract and represent features such as spiral arms and disk shape, which, combined with training data, are used to "train" the models to generate correct outputs.

• The accuracy of the models' outputs is measured using statistical measures such as true positive rate and true negative rate.

Morpheus is an example of an AI model used to classify galaxies. AI/ML models are also being used in many other innovative and transformative ways such as to detect supernovae, classify stars, find exoplanets, image black holes, align telescopes, monitor space debris, and so on. Who knows, our AI (artificial intelligence) may help to find other AI (alien intelligence) one day!