When: Thursday 20 January at 14.00
Where: Microsoft Teams - Online
Speaker: Mike Warmsley – University of Manchester
Hosted by: Hugh Dickinson
Galaxy Zoo recruits members of the public to classify galaxies, allowing morphology measurements of hundreds of thousands of galaxies. At the same time, the advent of deep learning has transformed our ability to make automated measurements. In this talk, I will describe how we combine crowdsourcing and deep learning to do better science than with either alone. I will discuss how, when training new deep learning models, we ask our volunteers to label the galaxies which would be most informative for our models. I will share how we build models which learn from galaxies on which the volunteers are uncertain and predict full posteriors for every galaxy. I will also share some early science results from the latest Galaxy Zoo project and some advice on using the morphology catalog for your own science projects. Finally, I will present new results exploring how our trained deep learning models can solve several practical science tasks for which they were never trained. Doing so requires only minimal new labels, challenging the longstanding view that deep supervised methods require new large, labelled datasets for practical use in astronomy.