Ordering for Hypothesis Testing and Beyond
Abstract
This effort will develop the general theoretical underpinnings needed to design and justify a different, more energy ef?cient, distributed processing approach, called ordering, for general cases which include those with statistically dependent observations. Our team had previously proposed ordering for cases with statistically independent observations from sensor to sensor, but we have recently uncovered evidence it can be expanded to statistically dependent observation cases. In the more general ordering approach, sensors are grouped into clusters for processing based on the statistical distribution of the sensor observations. Each cluster head will collect and summarize the observed data from their cluster. Then each cluster head will decide, in a distributed way, the relative value of the data from each cluster, such that the cluster head with the most informative data will transmit ?rst while the cluster heads with less informative data will transmit later. Using such an approach in cases with statistically independent observations, we have shown that it is often unnecessary to transmit the less informative data, thus saving transmission energy. The savings can be quanti?ed using analysis and appear impressive in signal detection problems with acceptable error probabilities. Here we will attempt to demonstrate similar ?ndings for cases with statistically dependent observations from sensor to senor. We will also investigate efficient distributed processing using alternative approaches to ordering for important applications including machine learning, cyber security and inference.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Oct 15, 2018
- Source ID
- W911NF1710331
Entities
People
- Rick Blum
Organizations
- Army Contracting Command
- Lehigh University
- United States Army