New Paper – Practical Galaxy Morphology Tools

Last year, we published the GZ DECaLS catalog: detailed morphology classifications for 314,000 galaxies. We classified so many galaxies by training AI models to learn from volunteers and work alongside them. This raises the question – what else can we do with those models?

It turns out that we can use them to make three new practical tools that will help both professional researchers and volunteers. You can read all about them in our new paper out today: https://arxiv.org/abs/2110.12735.

The first practical tool is a similarity search. You can type in the coordinates of a galaxy, and it will try to show you the most similar galaxies. Try it out on your favourite DECaLS galaxy. For now, it’s a simple demo website, but we hope to eventually integrate this into Galaxy Zoo.

The second is a new method for finding the galaxies most interesting to you personally. Imagine a website where you can rate galaxies by how interesting you find them. As you rate galaxies, the website shows you new ones for you based on your previous ratings – just like how Netflix suggests new series (I’m a big Bojack fan myself). The system is too complicated to create a simple demo to show you, but you can see some examples in the new paper. Thanks to funding from the Sloan Foundation, we’re making this even better and adding it as an official Zooniverse feature.

The third is about adapting the AI models to classify new kinds of galaxies. If a researcher wants a model that can find ringed galaxies, for example, they would usually have to start by gathering tens of thousands of examples of ringed galaxies with which to teach their new model. This takes a long time and a lot of effort, especially for rarer galaxies. However, a model already trained on Galaxy Zoo classifications needs just hundreds of example galaxies to learn to find rings as well. This will let researchers “fine-tune” models to help solve their own specific science problems. That includes me! I’m running a Galaxy Zoo Mobile project to make a new ring catalogue with this approach.

All these tools work because of your classifications. As well as using them directly in science catalogues, we need them to train better AI models. Thank you for your contribution.

If you have any spare time – maybe on the bus, or just sitting around scrolling – I would really appreciate your help finding ring galaxies by swiping left and right on Galaxy Zoo Mobile, part of our Zooniverse app (Apple, Android). I’m hoping to build the biggest catalogue of rings ever assembled so we can understand how they form. Please join in if you can.

Cheers,

Mike

P.S. You can find a few more technical details on my personal blog.