How does machine learning help us understand galaxies?

The scale of the universe is difficult to fathom. Our world is inside a star system, which is in turn inside a galaxy, which is in turn inside a galaxy cluster, which is in turn inside a supercluster. Understanding our own corner of the universe requires comparing it to others. To this end, some researchers specialize in galaxy morphology and evolution. Machine learning is proving to be a useful tool for comparing galaxies. To learn more about galaxy morphology and evolution, and about how machine learning helps understand it, we spoke to Clár-Brid Tohill. In her PhD studies at The University of Nottingham, she specializes in using machine learning to study galaxies.

What is galaxy morphology and evolution?

Morphology is the name given to the visual shape or structure of a galaxy. By looking at the morphology of a galaxy we can infer many different things about its past and ongoing formation. We can infer, for instance, how many stars are being formed or if a galaxy has merged with another galaxy in the past. Galaxies that formed most of their stars early on in their formation appear very different to galaxies where star formation is still ongoing. Astronomers therefore classify galaxies based on their morphology. Such classification is fundamental to understanding galaxy evolution.

Galaxy evolution studies try to explain how the first galaxies formed in the early universe and how they evolved throughout time to resemble the familiar spiral and elliptical type galaxies we observe in our local universe. One way this is investigated is to record how galaxy morphology changes the further back in time we look. In space, the more distant something you observe is, the further back in time you are seeing as the light from that object takes a long time to reach Earth. Thus, by looking at distant galaxies and comparing their morphology to nearby galaxies, we can try to link the two and better understand the processes involved in their evolution.

How does machine learning help us understand galaxy morphology and evolution?

The future of astronomy consists of many “big data” surveys that plan to image billions of galaxies and generate orders of magnitude more data than we have currently. Using humans to classify these galaxies therefore becomes infeasible, and even the computational methods we use currently will be expensive and inefficient at analyzing the amount of data we will be collecting. This is where machine learning comes in. Machine learning techniques can be trained to analyze and compute certain parameters thousands of times faster than current methods. Machine learning is thus a suitable and practical replacement to human processing.

There has already been a lot of work carried out applying machine learning techniques to various problems in astronomy, from galaxy classification to computing galaxies’ morphological parameters. Deep learning in particular is very promising for this problem as you can extract information directly from the image itself. This means that fewer preprocessing steps have to be carried out – sometimes none at all. This significantly reduces time requirements and can lead to a more efficient and robust method of analyzing large datasets.

In my work, for example, I use deep learning to compute different morphological parameters of galaxies. I have found that, as well as being much more efficient, deep learning can also be more accurate and reliable than traditional methods. It should thus be suitable for analyzing future datasets.

How did you come to focus your PhD studies on this topic?

I have always been interested in astronomy and especially in a career in research. I decided to focus my PhD studies on galaxy morphology and evolution because I believe it is an exciting and challenging field with a lot more to be discovered!

When I started my PhD, machine learning had started to become more popular and had successfully been applied to various problems in astronomy – galaxy classification, identification of merging galaxies, and many others. While it has proven to be a useful tool for astronomers, I believe it is only in its infancy. In upcoming years, we will see just how powerful a tool it can become. It is a fast-paced industry. I am excited to see where it can take astronomy and my research in the future.