Manifold Learning for Network Inference

• Activity Number: 533 - SLDS CPapers NEW 2
• Type: Contributed
• Date/Time: Wednesday, August 1, 2018 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Mingyue Gao* and Carey E Priebe and Minh Tang
• Companies: The Johns Hopkins University and Johns Hopkins University and Johns Hopkins University
• Keywords: Manifold Learning; Network ; Testing; Random Dot Product Graph; Adjacency Spectral Embedding; Structure Discovery

Abstract

Manifold learning is widely used in applications involving high-dimensional data, and many methodologies have been developed for achieving the low-dimensional representation which is desirable for subsequent inference. Here, we consider a framework for network structure discovery via manifold learning applied to the Adjacency Spectral Embedding (ASE) representation space or Laplacian Spectral Embedding (LSE) representation space for Random Dot Product Graph (RDPG). By investigating hypothesis testing powers of the ASE (or LSE) representation and of the low-dimensional representation after manifold learning, we show that the RDPG network inference procedure developed here yields higher power than inference in ambient ASE representation space.