Hyperbolic Space
Hyperbolic space, a non-Euclidean geometry with unique properties, is increasingly used in machine learning to represent hierarchical and complex data structures that are poorly captured by traditional Euclidean methods. Current research focuses on adapting existing models, such as transformers and graph neural networks, to hyperbolic space, developing novel algorithms for clustering and classification within this geometry, and exploring its application in diverse fields like image recognition, natural language processing, and recommendation systems. This work is significant because it offers improved performance and efficiency in handling hierarchical data, leading to advancements in various machine learning tasks and potentially impacting diverse applications.
Papers
A Geometry-Aware Algorithm to Learn Hierarchical Embeddings in Hyperbolic Space
Zhangyu Wang, Lantian Xu, Zhifeng Kong, Weilong Wang, Xuyu Peng, Enyang Zheng
Conformally Natural Families of Probability Distributions on Hyperbolic Disc with a View on Geometric Deep Learning
Vladimir Jacimovic, Marijan Markovic