Software, when put in service to perform some function, may not perform as desired. The failures may be due to errors, ambiguities, oversights or misinterpretations in the specifications. Carelessness or incompetence in writing code, inadequate testing or incorrect or unexpected usage of the software may contribute to the problems. These failures create an environment of uncertainty. Will the software fail or not? If so, when? Statistical modeling and data analysis procedures provide tools to investigate this phenomenon. In this talk, Prof. Keiller will present an overview of two types of statistical models (software reliability growth models): inter-failure time models and failure count models (the non-homogeneous Poisson Process models). He will also discuss procedures or analyzing and improving measures of the =93goodness of fit and the quality of prediction measures.

Biography

Peter Keiller joined the Department of Systems and Computer Science at Howard University in August 1994 as an assistant professor. He has 23 years of experience at IBM, concentrating in software application development and test, software quality analysis, capacity planning, and performance modeling. His research interests include software engineering, performance modeling, fault-tolerant programming and software reliability modeling. He is the co-developer of the Keiller-Littlewood software reliability model and the author of numerous research papers. Keiller graduated with the BSc degree in civil engineering and the MSc in computer science from Howard University; he holds a MSc and professional degree (applied scientist) in operations research and a DSc in engineering management from George Washington University.

CREST home page	News	Research	Seminar Series	Publications	Feedback