Recent years have seen an explosion of interest in Machine Learning, driven in part by the use of ML technology in a wide variety of applications from natural language processing to computer vision, including in industry. Theoretical advances have had a significant positive impact on the past development of the field. Theoretically inspired techniques such as Boosting, SVM, Winnow, and others are widely used, and frameworks and analytical tools developed by the theory community are often used to guide the design of new algorithms. Yet many exciting new application settings and methods being currently developed in the field do not fit standard theoretical models. The goal of this workshop is to bring together theoretically-inclined researchers from the NIPS/ICML community with machine-learning-inclined researchers from the FOCS/STOC community to discuss these new directions and theoretical challenges. These include settings such as deep learning, structure learning, learning from new kinds of feedback and interaction, and more. The hope is to identify questions and directions where theoretical contributions can be especially impactful.