Multimodal Subspace Learning and Modeling of Complex Systems

Abstract

This project will create a comprehensive framework for modeling multimodal complex and dynamic data. There are two unifying themes: (a) create the underlying mathematical foundation of subspace modeling; (b) exploit multimodal data by combining tools from random forest, information theory, and the subspace model:ing from the first theme. The final goal will be the development of a novel algorithm for non-negative matrix factorization. To model multimodal data and advance the state-of-the-art in the important applications, the incorporation of novel analysis and computation tools is critical. The PI will create and utilize tools from subspace modeling in the fonn of learning multimodal low-rank representations, modeling multimodal sparse networks, and solving for big data matrix decompositions. Witb the models developed, both multiple-instances and multirnodal can be studied with the same tools; and multiple-instances are often so diverse, e.g., due to noise and acquisition variability, that considering them as multimodal is beneficial. Close collaboration and interactions with DoD in general and ARL in particular are an integral part of this project.

Document Details

Document Type
DoD Grant Award
Publication Date
Jan 12, 2017
Source ID
W911NF1610088

Entities

People

  • Guillermo Sapiro

Organizations

  • Army Contracting Command
  • Duke University
  • United States Army

Tags

Fields of Study

  • Engineering

Readers

  • Distributed Systems and Data Platform Development
  • Linear Algebra
  • Neural Network Machine Learning.