Ordering for Hypothesis Testing and Beyond
Abstract
In this project, we have developed novel distributed processing approaches which can reduce the number of required communications in many applications. The distributed processing approaches that are popular with the research community require a very large number of communications and this has a number of disadvantages. The Army has many applications which employ wireless communications between battery powered nodes. In these applications, communications are typically one of the largest sources of energy usage. Thus, these excessive communications tend to drain the batteries and this can have serious negative impact on battlefield superiority. The latency for typical computing units to send data over a standard wired network connection is around 2500 times larger than that for accessing data in its own main memory. Reducing the number of communications can provide much faster results. This project seeks to develop methods which would allow the number of communications to be dramatically reduced with no significant loss in performance when compared to the performance of the best full-communication approaches. Hypothesis testing problems (hypothesis H0 versus H1) which attempt to decide which of two known probability density functions produced a given set of observations are of considerable interest in many Army applications like communications and sensor networking. Our team had previously developed an impressive new approach to reduce communications in distributed processing focused on solving such hypothesis testing problems. Our team originally developed this approach, called ordering, for the case where the observations were statistically independent from sensor to sensor conditioned on the hypothesis. A major objective for this project was to extend our approach using ordering combined with an alternative approach called censoring, to cases with dependent observations. We have successfully done this.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 19, 2021
- Accession Number
- AD1204414
Entities
People
- Rick Blum
Organizations
- Lehigh University