Statistics Seminars: Bayesian inference about design reliability from testing: some practical issues
9 March 2001 14:00 in CM221
"If we need to predict the unreliability due to unknown design faults (e.g. in software), a Bayesian approach seems the obvious choice. We may ask, then, why it is so little used in practice and why proposed uses are often unconvincing. I will describe some published exercises in Bayesian reasoning about software testing, including some success in uncovering plausible fallacies. The topics will include the value of software ""testability"", the use of Bayesian networks to describe complex probabilistic models, and some ideas about testing multiple-component systems. I will then discuss practical difficulties, concerning comprehension of the approach, agreement on the value of results, and the choice of priors. I will argue for a purist approach to the first two issues and a flexible one to the third."
room CM103 (seminar) 2.15 - 3.00 pm
Contact email@example.com for more information