Abstract: In recent years, graph representation learning has emerged as a powerful tool in the data mining and machine learning community with numerous applications in knowledge graphs, natural language processing, recommender systems and multimodal data analytics. The algorithms for doing graph representation learning have also undergone rapid advances. Early algorithms tended to rely on matrix factorization and statistical dimensionality reduction techniques, but because of deep learning, the quality of the representations has significantly improved. Early algorithms relied on 'shallow' features such as random walks, but more recently, higher-order structures have started to be incorporated in the representation learning algorithms, including through Graph Convolutional Networks. In this talk, we provide an overview of graph representation and the evolution of the algorithmic landscape in this area.
Bio: Dr. Mayank Kejriwal is a research assistant professor and research team lead in the University of Southern California. He holds joint appointments at USC's Information Sciences Institute and in the Department of Industrial & Systems Engineering. His research is in emerging technologies, including AI, complex systems and knowledge graphs, and their applications to social good. He regularly collaborates with government and industry, with project areas spanning e-commerce, machine common sense, network science and crisis response. His latest book, published by MIT Press in March, 2021 is titled 'Knowledge Graphs: Fundamentals, Techniques and Applications.'