Information Processing and Fusion via Sparse Factorization
Abstract
The objective of this project is to create a hybrid (discrete and continuous) sparse framework for information processing and fusion. A problem of merging importance is the extraction and fusion of information from user (or sensing active nodes) interactions. The key feature of this problem is to use observational data first to understand the relationship between the users and the content (i.e., factor analysis), and second to understand the relationships among the various content items (i.e., content analytics). A promising approach is to jointly perform factor analysis and content analytics while exploiting low-dimensional problem structure, which required the creation of a new statistical model that encodes the probability of discrete user responses and the continuous information content appropriate for sparse analysis. Research will focus on three thrusts: (1) creation of new statistical models for joint factor analysis and content analytics that leverage the sparse and low-rank structure of many emerging information extraction application; (2) design of novel algorithms to learn the new models to incomplete data; (3) validation of the framework and algorithms on a diverse range of data in cognitive sciences, recommendation systems, performance metering, psychology tests and dictionary learning.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 12, 2017
- Source ID
- W911NF1510316
Entities
People
- Richard G. Baraniuk
Organizations
- Army Contracting Command
- Rice University
- United States Army