Sufficient Statistics for Decentralized Estimation,
Abstract
Decentralized estimation problems involve several agents receiving separate noisy observations of a common stochastic process, and each seeks to generate a local estimate of the state of that process. In the general case, these estimates are desired to be consistent in some way, and thus may be jointly penalized with the state via a cost functional to be minimized. In many cases, each agent need only keep track of its local conditional state probability distribution in order to general the optimal estimates. This paper examines the boundary between problems where this statistic is sufficient and those where it is not; when it is not, the additional information which must be kept appears to have additional structure as illustrated by an example. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 1982
- Accession Number
- ADA124941
Entities
People
- Robert R. Tenney
Organizations
- Massachusetts Institute of Technology