Bayesian guided diagnosis of software failures
Lisa J. Burnell · 1996
There is strong motivation for developing methodologies that support modifying and maintaining legacy software. We present a new approach, Bayesian-guided diagnosis, to diagnose software failures caused by errors in the program logic. Bayesian-guided diagnosis adopts the premise that, realistically, perfect and complete knowledge is rarely available, and even when it is, the resources required to collect or generate it are often too costly. As the primary case study, we examine in detail the problem of algorithmic program debugging and the DAACS (dump analysis and consulting system) prototype. Demonstrating the generality of the system, we also show other applications of the approach, most notably to debugging AI-based planning systems.