Making Discoveries with Generative Models in the Era of the Nancy Grace Roman Space Telescope

With the advent of future big-data surveys, automated tools for unsupervised discovery are becoming ever more necessary. In this talk I will show how deep generative models can be used to learn a representation of expected data defined by the training set and then look for deviations from the learned representation by looking for the best reconstruction of a given object. We perform several experiments: 1. We first show that we are able to detect ~90% of interacting galaxies if we train our network only with a sample of isolated galaxies. 2. We then explore how the presented approach can be used to compare images of observed and simulated galaxies by identifying real objects not well rendered in the simulations. 3. We, finally, blindly look for interesting objects in the Hyper Suprime Cam Survey survey with millions of detected objects. We show that our approach successfully identifies outliers and is sensitive to subtle morphological deviations which are typically missed by more traditional moment-based approaches