Galaxy Zoo in the Deep Learning Era

Deep learning is fundamental to Galaxy Zoo’s latest morphology catalogs. In this talk, I explore how we train accurate and reliable models on our volunteer labels, and consider the consequences and opportunities of such models.
One obvious consequence is scale. We will shortly release detailed morphological classifications for 1.4 million nearby galaxies in the DESI Legacy Surveys – a catalog which would have been impossible with volunteers alone. But the truly new opportunity is using the models to create your own catalogs. Having already learned to answer every GZ question at once, our models are easy to adapt to new surveys and new questions. We recently exploited this to create new (and order-of-magnitude larger) catalogs of mergers in HST and ringed galaxies in DECaLS. You can answer your own morphology questions with our public code and models.
Finally, we describe our very latest work on simultaneously learning from labelled and unlabelled galaxy images. Such approaches are ideally suited to Euclid and Rubin because they allow us to leverage both the millions of human labels collected over the last decade and the raw scale of unlabelled images these new surveys will produce.