Using variational autoencoders to explore the latent space of a large-scale galaxy image dataset.