Manifold Learning for Subsequent inference: Structure Discovery: Exploitation in Networks

Abstract

Manifold learning is ubiquitous in machine learning and statistical pattern recognition – modern applications involve high-dimensional, multi-modal data, and successful methodologies require (often nonlinear) dimensionality reduction, whether explicit or implicit. But manifold learning for its own sake is rarely the goal; one desires a low-dimensional representation as just the first step in addressing the ultimate exploitation task. Thus the proper assessment of the manifold learning step is in the context of its utility for subsequent inference. Many modern applications involve data in the form of a network, and the first step in addressing the associated exploitation task is to identify structure in the network, with the endgame being exploitation of discovered structure for subsequent inference. We propose to formulate network structure discovery as a manifold learning problem in spectral decomposition space, and theoretical, methodological, and practical investigation into appropriate manifold learning mechanisms for the facilitation of various subsequent inference exploitation tasks, including testing, estimation, classification, regression, etc.

Document Details

Document Type
DoD Grant Award
Publication Date
Jul 29, 2020
Source ID
N001741910011

Entities

People

  • Carey E. Priebe

Organizations

  • Johns Hopkins University
  • United States Navy

Tags

Fields of Study

  • Computer science

Readers

  • Neural Network Machine Learning.
  • Systems Analysis and Design

Technology Areas

  • AI & ML
  • AI & ML - Machine Learning Algorithms
  • AI & ML - Neural Networks
  • Space