Noncommutative Probability and Information Theories for Inference from Networked Multi-sensor Data
Abstract
The objective of this project is to establish new probability and information theories that will capture various forms of noncommutativity in the multimodal interdependent data of large heterogeneous sensor networks. The research will also investigate collaborating humans as integral parts of such systems. The new models will capture noncommutativity phenomena that are due to causal relations between the variables in space and time as well as due to recently validated phenomena of interference between human observations and multiple observers in sensor and social networks, and in human team work and collaborations. The research is organized in four interrelated and complementary thrusts. Thrust 1: Directed Information and Network Information Theory. Effort will establish new extended versions of Directed Information and conditional Directed Information and use them to develop new rate distortion, sensor fusion and performance limitations. These new concepts will be established for networked spatio-temporal data towards a Network Information Theory. Thrust 2: Noncommutative (Quantum-like) Probability Models for Directed Networked Data and Inference. Effort will create new probability models not based on Kolmogorov axioms and the associated Information Theory, inference and likelihood algorithms. The new models will allow incompatible events and will be evaluated against the classical ones in the major application classes of interest, including distributed human multi-agent inference. Thrust 3: Noncommutative Probability Models Based on Various Logics. Research will investigate the various causes of noncommutativity that we will discover and validate, and associate them with different event logics. We will develop a taxonomy of these event logics and classes of probabilistic models that account for both causal directivities as well as constraints between the agent operations. This is an effort to develop a unified theory for these new models. Thrust 4: Application and Model Validation-Verification. The new theories, models and algorithms created in Thrusts 1-3 will be further extended and applied to progressively more complex versions of the application class driving the research: sensor fusion in heterogeneous sensor networks, with emphasis on imaging sensors. Methodologies and algorithms will be designed to objectively test and evaluate which class of probabilistic models provides more accurate results.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 23, 2017
- Source ID
- W911NF1510646
Entities
People
- John Baras
Organizations
- Army Contracting Command
- United States Army
- University of Maryland