Learning with Knowledge Graphs

Knowledge Graphs (KGs) provide ways to organize, manage and retrieve structure information, playing an important role in a number of AI applications, such as recommender systems, question answering or natural language generation. However, a problem that is common to all KGs is their incompleteness. In this talk we briefly discuss different techniques to do knowledge graph completion, with a special emphasis on link prediction methods. Link prediction aims to complete queries wherein the answer is always an element of a closed vocabulary. Then we will learn how to perform link prediction in temporal knowledge graphs, wherein facts may be framed in time. This will be followed by an introduction to the numerical attribute prediction problem. Different to link prediction, the answer to the queries addressed in this problem is a numerical value. Finally, we will see that knowledge graphs and recommender systems share a number of similarities.

Alberto García-Durán is a senior researcher at NEC Labs Europe, Germany. He did his PhD at CNRS, France. During his PhD he interned at MILA lab, Canada. Prior to that, he studied Sound&Image Engineering, and Telecommunications Engineering in Spain. At a high level, his research interests lie under the umbrella term Graph AI.