Latest Galaxy Zoo paper submitted!

Hi Everyone!

This is to announce the submission of a new galaxy zoo paper entitled “Reproducing Galaxy Morphologies Via Machine Learning”.

First let me introduce myself as I am a newcomer to Galaxy Zoo. My name is Manda Banerji and I am a final year PhD student at University College London. I am interested in automated machine learning tools for morphological classification and became interested in comparing the classifications you all have worked on over the last couple of years to those produced by machine learning codes. So a couple of months back I started working with my PhD supervisor, Prof. Ofer Lahav, on using artificial neural networks to perform morphological classifications. The result has been this paper.

Basically what we have shown here is that using ~10% of the sample of objects that you have classified to train an artificial neural network, we are able to reproduce your classifications for the rest of the objects to an accuracy of greater than 90% provided we choose our neural network input parameters carefully.

There are of course caveats. A neural network is not as good as the human eye in recognizing unusual objects but by and large it does a decent job for the bulk of the galaxies. The most interesting result from this paper however was that if we limit our training sample to the brightest galaxies only and use these to classify fainter galaxies, this does not degrade our results. This means that using your visual classifications as a training set, we can use neural networks to accurately classify hundreds of millions of objects likely to become available over the next few years. These surveys will go considerably deeper and image fainter objects than in the Sloan Digital Sky Survey from which the Galaxy Zoo images were taken – therefore obtaining data for 100-1000 times more objects – and so your work has paved the way for accurate classifications with these future surveys too!

I hope to write another blog soon with a few more details on what we did and how the neural network works. Meanwhile, keep a look out for our paper. It has just been submitted to MNRAS and we eagerly await the referee’s reports!