I am interested in understanding when and why deep learning works well, and what we need to do to make it better. In particular, my PhD research has been focused on understanding generalization and representations by using a mix of theory and empirics.