Multi-Aspect Intelligence Fusion and Analytics: Models, Identifiability and Computation
Abstract
Abstract. Many modern military operations have been fully supported by powerful data acquisitions infrastructures, in which a large number of heterogeneous and autonomous data acquisition agents participate with various capacities. However, there has been an urgent need to develop theory and methods that can help effectively process and fuse such massive real-time multimodal data at scale, so that actionable intelligence can be gathered to improve situational awareness and to assist decision making. The overarching goal of the proposed work is to offer a new modeling, analysis and computational framework for challenging multimodal interaction data fusion problems. The specific thrusts proposed here are: Thrust 1: Tensor Modeling for Multimodal Analytics will design models and criteria that effectively represent the multimodal data interactions, so that different latent structures can be provably discovered and identified. Thrust 2: Partially Observed Modality Interaction will propose effective methods that can, under realistic conditions, provably identify the latent relationship between modalities whose interactions are not observed. Thrust 3: Computational Framework will address the computational issues associated with our proposed model identification criteria. The capability of our framework to tackle highly non-trivial and large-scale multimodal data analytics and fusion tasks, is appealing to a wide array of mission-critical applications demanding fast real-time decision support, such as early-warning systems, long-range surveillance, battlefield/battlespace information fusion and decision making, among others. The proposed framework represents an exciting paradigm shift for large-scale multimodal data analytics. If successful, we expect that our work will enable provable multimodal data analytics for many applications, especially cross-dimensional event/target detecting, pre-warning, and knowledge discoveries. It will also offer an array of effective and scalable numerical methods, with guaranteed performance and optimal tradeoff between solution quality and computational complexity.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 13, 2019
- Source ID
- W911NF1910247
Entities
People
- Mingyi Hong
Organizations
- Army Contracting Command
- United States Army
- University of Minnesota