Geometricity and Embedding

摘要: In this chapter, we compare and contrast two approaches to the problem of embedding non-Euclidean data, namely geometric structure preserving embedding. Under first heading, explore how spherical can be used embed data onto surface sphere optimal radius. Here both elliptic hyperbolic geometries, i.e., positive negative curvatures. Our results on synthetic real show that performs well under noisy conditions deliver low-distortion embeddings for a wide variety datasets. Hyperbolic seems much less common (at least in our datasets) is more difficult accurately embed. second Ihara zeta function hypergraphs manner which reflects their underlying relational structure. Specifically, polynomial characterization derived from leads an captures prime cycle hypergraphs.