Deep learning is making astrophysicists excited and afraid. We're excited because we can show a neural network every galaxy in the sky and ask questions, and afraid because we don't understand the answers. I'll talk about how you can investigate galaxy evolution with TensorFlow and crowdsourced labels, what we're discovering, and where it all falls apart.
Modern telescopes automatically detect far more galaxies than astronomers can ever actually look at. To turn a million galaxy images into something meaningful, science needs to learn from data science.
I'll give a crash course on galaxies and introduce you to The Zooniverse, a free platform for researchers to crowdsource labels. I'll show how I use TensorFlow and the PyData stack to build a galaxy image classification algorithm from noisy crowdsourced labels and then ensemble them together to outperform both humans and machines. I'll highlight where science and data science diverge, and what we can learn from each other.