A New Science for Reliability
Abstract
Although software reliability estimation based on results of software testing has been the subject of decades of software reliability publications, software reliability prediction using measures from the design and development of software has lagged. software reliability prediction using measures from the design and development of software has lagged. Software reliability prediction from Rome Air Development Center (RADC) in the 1980s enjoyed some popularity and then fell into disuse. Such prediction attempts were built on multiple regression models which, at times, suffered in precision and accuracy. Subsequent use of experimental research proved elusive and industry use of software reliability prediction modeling is almost non-existent. However, in 2018 with the publication of Dr Judea Pearls book The Book of Why, a complete new scientific approach to gaining knowledge of cause-effect without controlled experimentation became practical. Indeed, the ability to take observational data to create causal graphs and then quantify direct, indirect, mediated and moderated causal effects represents a major paradigm shift in research and specifically, reliability modeling of software and humans, as well as in risk analysis and prognostics and health management. Causal learning goes beyond traditional correlation to distinguish spurious correlation versus cause and effect. While traditional statistical regression and most forms of machine learning depend on correlation and association, causal learning will enable intelligence approaching what is termed General AI.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2020
- Accession Number
- AD1110406
Entities
People
- Robert W. Stoddard
Organizations
- Carnegie Mellon University