Title: Automated Knowledge Discovery from Scientific Articles Subtitle: Harnessing Link and Path Prediction Algorithms for Knowledge Discovery Abstract: 10,000 to 12,000 scientific articles on deep learning were generated last year; PubMed contains about 27 million biomedical research articles and NYTimes publishes 230 stories every single day. It is impossible to manually peruse these articles looking to connect the dots and generate new insights. However, automating this process is no easy task. In this talk, we explore the use of knowledge graphs for performing knowledge discovery. More specifically we review embedding based link prediction algorithms in knowledge graphs. We also present results on the importance of negative sampling for generating embeddings, and the usefulness of metadata, such as categories, for improving knowledge discovery. Finally, we propose a path prediction problem that would allow automated discovery of chains of facts.