A Spectral Framework for Graph Sampling

Abstract

This proposal seeks to investigate methodologicaJly and theoretically sound techniques for sampling interdependent data that make analysis feasible and lead to significant improvements in inference. This project will employ techniques from Markov chains, spectral graph theory, and statistical for interdependent data to describe a methodological and theoretical framework for scale-adaptive sampling on graphs. It will examine globally-adaptive Markov chain sampling to obtain a representative sample of the nodes in the graph. It will explore an áanti-transitive" sampling mechanism and attempt to demonstrate that the globally-adaptive property derives from the property that representative samples can be obtained more efficiently than standard link-tracing approaches. It will generate appropriate inferences for a sampled graph by characterizing the spectral sensitivity to sampling-induced edge dependence for two broad classes of sampling mechanisms (link-tracing and motif). ll will develop a class of locally-adaptive sampling mechanisms to test for local structure in massive graphs.

Document Details

Document Type
DoD Grant Award
Publication Date
Jan 12, 2017
Source ID
W911NF1510423

Entities

People

  • Karl Rohe

Organizations

  • Army Contracting Command
  • United States Army
  • University of Wisconsin–Madison

Tags

Readers

  • Neural Network Machine Learning.
  • Statistical inference.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Machine Learning Algorithms
  • AI & ML - Neural Networks