Laplacian Eigenspaces, Horocycles and Neuron Models on Hyperbolic Spaces

Ming-Xi Wang

Keywords: hyperbolic learning, hyperbolic neural network, Poincare embedding

Abstract: We use hyperbolic Poisson kernel to construct the horocycle neuron model on hyperbolic spaces, which is a spectral generalization of the classical neuron model. We prove a universal approximation theorem for horocycle neurons. As a corollary, this theorem leads to a state-of-the-art result on the expressivity of neurons of the hyperbolic MLR. Our experiments get state-of-the-art results on the Poincare-embedding tree classification task and the two-dimensional visualization of images.

One-sentence Summary: A novel spectral learning tool that can be applied to existing theoretical and experimental hyperbolic learning problems.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Reviewed Version (pdf): https://openreview.net/references/pdf?id=7lSMGo-ZgC