Dynamics of Information Fusion and Polling in Social Networks
Abstract
This project deals with the statistical signal processing and controlled sensing of interacting social sensors. Social sensors provide information about their environment (state) to a social network after interaction with other social sensors Ð this information fusion is modeled by a social learning protocol. The aim of this project is to develop mathematical models, structural results and algorithms for controlled sensing with social learning. The objectives fall under three inter-related themes: Task 1. Sequential Bayesian social learning where risk averse, rationally inattentive and anticipatory local decision makers interact with a global decision maker to achieve a global objective such as quickest change detection. Task 2. Interactive sensing in large scale networks where efficient polling strategies based on the friendship paradox are developed to estimate the labels (opinions) of nodes in both directed and undirected graphs. Also mean field dynamics for information diffusion in reactive networks are developed. Task 3. Analyzing massive social network data sets to justify the models and algorithms. The research involves the interplay of Bayesian social learning, partially observed Markov decision process (stochastic control), statistically efficient polling, and mean field dynamics. The problems considered in this project transcend classical statistical signal processing (which deals with extracting signals from noisy measurements) to address the deeper issue of how multi-agent decision systems and signal processing algorithms interact over a network.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 24, 2019
- Source ID
- W911NF1910365
Entities
People
- Vikram Krishnamurthy
Organizations
- Army Contracting Command
- Cornell University
- Defense Advanced Research Projects Agency