报告题目：Representation learning in hyperbolic space
报告人：Atsushi Suzuki 伦敦大学国王学院助理教授
报告时间：2023年6月19日10:00
报告地点：艺设西楼213会议室
报告对象：计算机科学与技术学院研究生及其他感兴趣师生
报告摘要：Hyperbolic space has attracted attention in machine learning this decade since even extremely low-dimensional hyperbolic space succeeded in representing entities with a large-scale hierarchical structure. This talk aims to help the audience to understand hyperbolic space’s properties and why it can contribute to some machine learning areas. At the end of this talk, the audience will be expected to be able to
- Explain hyperbolic space’s exponential space growth property,
- Judge whether hyperbolic space can contribute to applications that they are interested in, including natural language processing, and
- Have clues to design a machine-learning model using hyperbolic space (if we have additional time).
While hyperbolic space is usually defined as a Riemannian manifold in differential geometry, this lecture does not assume the preliminary knowledge of differential geometry.
报告人简介：Atsushi Suzuki has been working for King's College London, UK, as an assistant professor (UK Lecturer) since 2022 and for the University of Tokyo, Japan, as a visiting researcher since 2023. Atsushi was conferred the degree of Doctor of Information Science and Technology from the University of Tokyo, Japan, in 2020. Prior to this, Atsushi received the degrees of Master of Information Science and Technology and Bachelor of Engineering from the University of Tokyo, in 2017 and 2015, respectively. Atsushi was awarded Research Fellowships for Young Scientists by the Japan Society for the Promotion of Sciences from 2018 to 2020. Atsushi has published papers in top-leading academic journals and international conferences, including IEEE Trans. Inf. Theory, NeurIPS, ICML, ISIT, AAAI, IJCAI, and ICDM.
