Researchers have used convolutional neural network (CNN) architectures to develop a new artificial intelligence system that classifies tens of thousands of galaxies in a few seconds, a process that can take months to perform manually .
Astronomers classify galaxies by shape to understand how they form and evolve.
CNN architecture that outperforms previous models
Ultimately, the technique could deepen our understanding of how galaxies transform over time. As we can read in the study (preprint):
The key strengths of automated classification techniques, like our CNN approach, ultimately lie in their speed and ability to generalize. Although training a CNN can be a computationally expensive task, the speed with which it can classify galaxies once trained is orders of magnitude greater than could be possible with manual classification.
The team developed a CNN architecture that outperforms existing models in classifying galaxy morphologies into 3-class (elliptical, lenticular, spiral) and 4-class (+ irregular / miscellaneous) schemes. Their general classification accuracies were 83% and 81% respectively.
The researchers also claim that the system will be able to classify more than 100,000,000 galaxies at different distances from Earth and in different environments . According to the study’s lead author, Mitchell Cavanagh , a doctoral student at the International Center for Research in Radio Astronomy (ICRAR):
These neural networks are not necessarily going to be better than people because they are trained by people, but they are getting closer with more than 80% accuracy and up to 97% when classifying between elliptical and spiral. If you put a group of astronomers in a room and ask them to classify a bunch of images, there will almost certainly be disagreements. This inherent uncertainty is the limiting factor in any AI model trained with labeled data.