Face identification AI bot can now recognize radio galaxies
Australian researchers are training an artificial intelligence program to identify radio galaxies. This program was previously used to recognize faces in Facebook photos.
The program is called ClaRAN and was created by big data specialist Chen Wu and astronomer Ivy Wong of the University of Western Australia. The AI is based on an open source version of Microsoft’s and Facebook’s object recognition software. In fact, it is a large neural network in which masses of data have been inserted. This concerns the quality of the data sets entered, after which the AI algorithms are trained to optimize the rest.
When recognizing the galaxies, ClaRAN mainly uses data from various radio telescopes, but also infrared data, which improves the reliability of the predictions. The program looks at more than 500 different images of the radio galaxies to arrive at a classification.
According to Wong, traditional computer algorithms are able to correctly identify 90 percent of the source material. But that still leaves ten percent, which amounts to about seven million galaxies that must be viewed by the human eye through their complex elongated structures. Previously, this required the help of volunteers. “If ClaRAN reduces this percentage of visual classifications to one percent, citizen astronomers can spend more time looking at new types of galaxies,” Wong said.
Radio galaxies radiate powerful radio jets from huge black holes located at the centers of the galaxies. These jets are fountains of bundled streams of gas. The supermassive black holes attract gas and dust from their surroundings, after which highly energetic jet streams spew charged particles at almost the speed of light back into space.
On the left is a radio galaxy with a jet, which ClaRAN has detected using only data from radio telescopes. Based on this, the program is still uncertain what it is about, but based on data from an infrared telescope (right), the AI is certain of its prediction and classification.