Adaptive Sparse and Low-Rank Models for Real-World Visual Recognition

Abstract

When data is partially-labeled, heterogeneous and imbalanced, the performance of existing methods for visual data analysis can be much worse than expected. The objective of this project is to overcome the fundamental limitations of the current sparse and low-rank modeling techniques to create a new class of methods and algorithms for visual data analysis that will significantly improve the performance of data analysis. Proposed research will focus on the following specific tasks: (1) create sparse and low-rank modeling techniques for imbalanced data; (2) establish theory and algorithms with performance guarantees for learning sparse and low-rank models for multi-modal heterogeneous data; (3) establish theory and algorithms for learning sparse and low-rank models with partially-labeled data; and (4) create efficient scalable sparse and low-rank modeling techniques for large-scale problems.

Document Details

Document Type
DoD Grant Award
Publication Date
Jan 12, 2017
Source ID
W911NF1610126

Entities

People

  • Vishal Patel

Organizations

  • Army Contracting Command
  • Rutgers University
  • United States Army

Tags

Fields of Study

  • Computer science
  • Engineering

Readers

  • Neural Network Machine Learning.