5.2 Information Processing and Fusion: Fusion of Statistically Dependent Heterogeneous Information Sources
Abstract
The goal of this project was to investigate models, measures and methodologies for the fusion of statistically dependent heterogeneous information for a variety of inference tasks. We intended to understand the various ways in which statistical dependence among sensor measurements affects performance for inference networks and to develop collaboration and fusion methods that exploit statistical dependence to enhance inference performance in benign as well as in adversarial environments. In most research on distributed inference problems, sensor observations are assumed to be statistically independent or at least conditionally independent for analytical tractability. In real world, however, observations are statistically dependent. One barrier in this area is the inability to accurately model statistical dependence in a form that can be used for information fusion to enhance inference performance. The problem is exacerbated when the information sources are diverse and from different modalities. The second barrier is to define statistical dependence when the marginal distributions are not Gaussian, e.g., they are heavy tailed for example. Finally, dependence models must be scalable in a computationally efficient manner so that they can be employed for a large number of information sources. The significance of the work performed in this project was that it will enable the development of efficient fusion methodologies for dependent heterogeneous data from diverse and multimodality sources even in adversarial environments.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 16, 2023
- Accession Number
- AD1224624
Entities
People
- Pramod Varshney
Organizations
- Syracuse University