Neural networks help Australian astrophysicist classify messy galaxies in minutes

An astrophysicist from the Australian National University has applied neural networks to automate the process of analysing and finding particular galaxies.

An artificial neural network is part of machine learning (under the umbrella of artificial intelligence) where a computer program mimics how the human brain transmits data through signals between neurons and their connections with other neurons. Raw data values are first input into nodes, which are like neurons, and are passed through the input layer to the output layer of the network.

PhD student Elise Hampton, from the ANU Research School of Astronomy and Astrophysics, was able to classify ‘messy galaxies’ from 1,188 galaxies with 300,000 data points in eight minutes using neural networks.

“For one person to do it would have taken years,” Hampton said.

‘Messy galaxies’ are turbulent and are powered by black holes that cause huge galactic winds. Brightly glowing centres is a feature of these type of galaxies. They are of interest to Hampton because they can be used to better understand how galaxies live and die.

“We believe these winds blow so much material out of the galaxies that they eventually starve themselves to death,” Hampton said.

The neural networks were first trained on about 4,000 galaxy spectra, which was analysed previously by astrophysicists, to be able to successfully carry out the classification task of identifying messy galaxies.

Robotic telescopes like the ANU 2.3-metre and Anglo-Australian Telescope measure “enormous numbers” of spectra, with neural networks speeding up and automating the task of pin-pointing galaxies that are of interest to astrophysicists.

Human interpretation of the spectre to distinguish light from stars forming, matter falling into black holes, and supersonic galactic winds is a slow and laborious process, ANU said.