Hyperbolic embeddings for graph compression
arXiv:2607.11379
The paper presents a fast lossless graph compression method that leverages modern hyperbolic embedding techniques and demonstrates up to 42% improvement over existing methods on real-world networks.
Abstract
Network theoreticians hypothesize that the structure of real-world networks has a geometric origin. Especially, hyperbolic geometry was proven insightful in representing and modeling of scale-free networks. Embedders are algorithms used to find a geometric representation of a network. In this study, we introduce a fast lossless graph compression algorithm based on modern hyperbolic embedders. Experimental validation on real-world and generated networks shows that our algorithm beats state-of-the-art by up to 42% on real-world graphs.