Dimensionality Reduction of Streaming Big Data for Clustering, Classification and Visualization via Incremental
Abstract
One of the missions of the Naval Surface Warfare Center - Dahlgren is to “provide research, development, test and evaluation, analysis, systems engineering, integration and certification of complex naval warfare systems related to surface warfare, strategic systems, combat and weapons systems associated with surface warfare.” [1]. While there are likely a multitude of defense strategies being deployed to achieve such a mission, one active area of research to support such e↵orts is the dimensionality reduction, clustering, and classification of streaming Big Data. When examining the problem of clustering and classification resulting from streaming Big Data analysis, the key idea is to reduce the overall dimensionality of said data by exploiting statistical coherence between data sources while maximizing correlation between data streams. In the modern battle space, data is collected from a multitude of sources ranging from Electro-Optical, Radar, LIDAR, and Chemical Spectrometers, to name but a few. Subspace learning techniques (also known as Eigenspace methods) have shown great promise for dimensionality reduction while preserving meaningful information across disparate sensing modalities. Research Objective: Advance the Navy’s ability to analyze Big Data streams in real-time through a suite of tools grounded in incremental/adaptive Multi-linear subspace learning. Such a suite will be useful in many dimensionality reduction, clustering, classification, and visualization problems encountered within the U.S. Navy as well as the entire Department of Defense battle space. To meet this objective, we propose to develop efficient mechanisms for representing, characterizing, and analyzing multi-modal sensor data in real-time via incremental/adaptive Multi-linear Subspace Learning. The novelty in our approach is derived from a fundamentally di↵erent way to compute optimal subspace projections through tensor decompositions and extensions to incremental/adaptive eigenspace updating techniques. Meeting this objective will provide the U.S. Navy with advanced tools for dimensionality reduction, clustering, classification, and visualization of large streaming multi-modal data sets resulting in a better overall understanding of the battle space in real-time. Ultimately, such a suite of tools and algorithms will lead to better predictive capabilities for all Maritime Missions.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 30, 2020
- Source ID
- N001741910014
Entities
People
- Randy Hoover
Organizations
- South Dakota School of Mines and Technology
- United States Navy