My main research interest is trying to figure out how to learn good unsupervised world models and how to use them for vastly speeding up reinforcement learning. I have a fondness for probabilistic and bayesian modelling, models that learn by trying to predict the future and models that can be trained end-to-end.