ELIS researcher Sander Dieleman wins Galaxy Challenge

Galaxy Zoo (http://www.galaxyzoo.org/) is an online crowdsourcing project that invites people to help classify galaxies from images. By classifying observed galaxies based on their shape, astronomers can come to new insights about their origin and their distribution in space.
There are more than one hundred billion galaxies in the observable universe. Better telescopes are continuously being developed, and as a result we are able to photograph more galaxies that are much farther away. Classification of such a large quantity of image data by humans is not feasible, so this process must be automated.
That is why Galaxy Zoo organised a competition: the Galaxy Challenge (http://www.kaggle.com/c/galaxy-zoo-the-galaxy-challenge). The goal of this competition was to build a system that can classify galaxies automatically. More than 300 participants competed against each other to develop the most reliable classification system.
Sander Dieleman of the Department of Electronics and Information Systems won this competition with a system based on convolutional neural networks.
These models are able to learn relations from large amounts of data autonomously. Their structure resembles that of the human visual system.
Since a few years, convolutional neural networks are being applied on a large scale for image recognition, by Google and Facebook among others. In the Reservoir Lab (http://reslab.elis.ugent.be/), the research group which Sander Dieleman is a part of, research is also being conducted about the use of convolutional neural networks for sign language recognition, music recommendation and automatic detection of epileptic seizures.
More information: