Speaker

Doug Cochran
Department of Electrical Engineering

(Joint work with D. Sinno and Q. Li)

Title

"Sequences of Tests"

Abstract

The binary hypothesis test plays central roles in radar/sonar engineering, digital communications, and numerous other areas of engineering and science. Assessing whether a target is present or absent, whether the transmitted bit was a zero or a one, and whether the device being tested is good or faulty are all applications in which the same mathematical principle is employed: based on statistical knowledge about the data under each of the two hypotheses and prior information about the probabilities of the hypotheses, which hypothesis is more likely to be true given the observed data can be deduced using Bayes' rule.

This talk will focus on questions arising when a sequence of binary hypothesis tests is performed. In particular, Bayesian updating of the hypothesis probabilities between tests is scrutinized and argued to be the best thing to do only under special circumstances. The design of test sequences under conditions where a certain level of confidence absolutely, positively must be attained after N tests is also considered; occasional acts of desperation are seen to be justified.