Hyperbolic Graph Neural Networks

Abstract

Learning from graph-structured data is an important task in machine learning and artificial intelligence, for which Graph Neural Networks (GNNs) have shown great promise. Motivated by recent advances in geometric representation learning, we propose a novel GNN architecture for learning representations on Riemannian manifolds with differentiable exponential and logarithmic maps. We develop a scalable algorithm for modeling the structural properties of graphs, comparing Euclidean and hyperbolic geometry. In our experiments, we show that hyperbolic GNNs can lead to substantial improvements on various benchmark datasets.

Published at NeurIPS 2019

Cited by in corpus (20)

• Hyperbolic Graph Convolutional Neural Networks
• Geoopt: Riemannian Optimization in PyTorch
• From Trees to Continuous Embeddings and Back: Hyperbolic Hierarchical Clustering
• Dorylus: Affordable, Scalable, and Accurate GNN Training with Distributed CPU Servers and Serverless Threads
• Hyperbolic Graph Attention Network
• Hyperbolic Deep Neural Networks: A Survey
• Knowledge Association with Hyperbolic Knowledge Graph Embeddings
• Coordinate Independent Convolutional Networks -- Isometry and Gauge Equivariant Convolutions on Riemannian Manifolds
• Privacy-Preserving Representation Learning on Graphs: A Mutual Information Perspective
• HCGR: Hyperbolic Contrastive Graph Representation Learning for Session-based Recommendation
• Hyperbolic Variational Graph Neural Network for Modeling Dynamic Graphs
• A Hyperbolic-to-Hyperbolic Graph Convolutional Network
• Mix Dimension in Poincaré Geometry for 3D Skeleton-based Action Recognition
• Learning Hyperbolic Representations of Topological Features
• Switch Spaces: Learning Product Spaces with Sparse Gating
• Are Hyperbolic Representations in Graphs Created Equal?
• Lorentzian Graph Convolutional Networks
• HyperExpan: Taxonomy Expansion with Hyperbolic Representation Learning
• Graphs for deep learning representations
• Isometric Graph Neural Networks