10:00 - 11:00
Injecting Prior Information and Multiple Modalities into KB Embeddings
Knowledge graphs have become an increasingly crucial component in machine intelligence systems, powering ubiquitous digital assistants and inspiring several large-scale academic projects across the globe. However, most existing knowledge graphs are noisy and incomplete. In this talk, I will describe approaches that perform relational reasoning to complete and correct such knowledge graphs. First, I will provide background material and introduce recent advances in knowledge graph embeddings that use latent continuous representations of entities and relations to predict missing facts in the knowledge base. Then, I will present some of my past research on hybrid models that combine prior information, specified as symbolic logical statements, with dense, neural representations of the KB. Finally, I will describe some of my recent research on incorporating multiple modalities, like text, numbers, and images, in the KB embeddings in order to extract using all types of available evidence. With research in these directions, we hope to facilitate easy-to-use, accurate machine learned systems for completing knowledge graphs.
