Reliable Inference in Dynamic Data Fusion
Abstract
Proper inference in a decision or inference network requires that the commander (technically: the fusion center) have an understanding of the relative weight that he / she should place on the inputs each subordinate. Recent works have addressed the problem of estimating agents' behaviors in complex networks, of which social networks are a prominent example. These works are especially promising and would seem to be of considerable practical importance in a wide variety of command and control venues. However, these works are perhaps limited by their somewhat idealized assumptions: that the commander (fusion center) possess full information of all subordinates' histories, and that conditional statistical independence between these histories can be assumed. In the proposed project we intend to explore more general situations : of dependent sensors, of unknown structure of that (possible) dependence, of missing data and of subordinate identities that are either obscured / adulterated / entirely missing. For such dynamic fused inference problems we propose to extend results in a number of directions: exploring dependency amongst data sources (physical proximity or "group-think"), in term of useful communication strategies when the inference task and quantization are not necessarily matched, and even the unlabeled case in which the identity of each measurement's source is unknown - this is a form of the data association problem.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 15, 2021
- Accession Number
- AD1155230
Entities
People
- Peter Willett
Organizations
- University of Connecticut