I will discuss some of the recent advances in using non-Euclidean geometry in machine learning, mostly focusing on hyperbolic geometry. After introducing some important concepts such as curvature, I will discuss how hyperbolic embeddings can be used to model entailment or hypernymy relations. Next, I will present extensions of basic deep learning architectures to hyperbolic spaces, showcasing how this geometry can be leveraged in standard ML pipelines.
Bio: I am a postdoctoral researcher in the group of prof. T. Jaakkola and prof. R. Barzilay at CSAIL-MIT. Previously I obtained my PhD from the Data Analytics Lab at ETH Zurich under the supervision of prof. Thomas Hofmann. I am broadly interested in representation learning for text, graphs or images through statistical or geometric models that could be devised and understood in a mathematically principled manner. In particular, I have recently explored finding and learning latent hierarchical structures in data via hyperbolic geometry.