This lecture is about the role of representation in deep learning. It begins with how the representation learnt depends on the task being solved. A basic method, helpful to understand to modern representation learning, called "Greedy layer-wise unsupervised pre-training", is described.
