Graph-structured data is ubiquitous and occurs in several application domains. My talk will provide an overview of our group’s recent work on representation learning for relational data, including some concrete applications in the medical domain. One of the long-standing goals of machine learning is to infer and leverage relational structure, even if such structure is not available a-priori. Graph representation learning approaches have been limited to applications where relational structure is given or have used heuristics to construct affinity graphs before learning commences. We have recently proposed an end-to-end differentiable framework that jointly learns the discrete relational structure and parameters of graph convolutional networks for semi-supervised learning. This makes graph neural networks applicable to a much wider range of learning problems.