Siraj Raval explores one of the newest frontiers of deep learning: Geometric Deep Learning and why non-Euclidian geometry requires new kinds of deep learning approaches.