What is it about?
We use an unsupervised learning technique - a variational autoencoder (VAE) - to find the charactersitic features in images of galaxies. The galaxies images come from a computer simulation. The VAE tests the features by trying to rebuild the original image from the features (or image building blocks extracted).
Featured Image
Photo by Arnaud Mariat on Unsplash
Why is it important?
Modern galaxy surveys contain many tens or hundreds of millions of galaxies -- too many for even an army of citizen scientists to classify. Our approach is fast and avoids human biases.
Perspectives
This project came from Sam Howie's final year project as an undergraduate in the Department of Physics at Durham University.
Professor Carlton M. Baugh
Durham University
Read the Original
This page is a summary of: Deciphering galaxy images using machine vision – combining variational autoencoder and principal component analysis for feature extraction, Monthly Notices of the Royal Astronomical Society, November 2025, Oxford University Press (OUP),
DOI: 10.1093/mnras/staf1926.
You can read the full text:
Contributors
The following have contributed to this page
