Graph Representation Learning Beyond Hyperbolic Geometry

Learning numerical representations of graphs is a fundamental problem in machine learning, with traditional Euclidean approaches often falling short in capturing important features such as hierarchies. Motivated by this limitation, there has been growing interest in developing geometric representation learning frameworks that better reflect the properties of target graphs, such as embedding hierarchies in hyperbolic spaces. In this talk, we discuss two such frameworks that go beyond hyperbolic geometry. This talk is based on work with Federico Lopez, Max Riestenberg, Michael Strube, Steve Trettel, and Wei Zhao.