B1438
Title: Population-level inference for networks via graph embeddings
Authors: Jesus Arroyo - Texas A&M University (United States) [presenting]
Avanti Athreya - Johns Hopkins University (United States)
Vince Lyzinski - University of Maryland, College Park (United States)
Abstract: The problem of inferring population properties from observed network samples is considered. We approach this problem via dimensionality reduction by projecting the data onto a low-dimensional space. It is shown that this procedure can yield accurate inferences; however, in the presence of shared structure across the networks, classical approaches such as PCA are sub-optimal and can have reduced power. We show that a graph embedding that exploits a common low-rank structure can yield improvements, and we present a central limit theorem for the resulting network projections. Applications of this methodology are introduced in simulations and real data, including two-sample testing, anomaly detection, and network classification.