automated representation learning is behind many recent success stories in
machine learning. It is often used to transfer knowledge learned from a large
dataset (e.g., raw text) to tasks for which only a small number of training
examples are available. In this paper, we review recent a