EMPRESS: Extracting and Mining of Probabilistic Event Structures

The field of specification mining aims at extracting and reconstructing models from existing software systems - models that would be precise enough to serve as specifications. Most of the current approaches abstract software behaviour through models concisely representing all of the valid sequences of events that may happen during execution. However, we do not see how likely such sequences are. In EMPRESS, we want to extract probabilistic models in which transitions between events are labelled with probabilities: ``After open(), 99 % of executions are followed by read(), whereas 1 % end up in an exception.'' Such probabilistic models will increase the accuracy and effectiveness of several software engineering activities: In runtime monitoring and debugging, such models uncover anomalies during execution that predict and diagnose faulty behaviours. In software testing, probabilistic models allow to focus on likely behaviour, where defects will have the greatest impact, as well as unlikely behaviour, where yet undiscovered defects may loom. Finally, in security and reliability, probabilistic models capture deviations from normal behaviour, unveiling faulty and malicious behaviours. The central research challenges will be (1.) to efficiently extract accurate probabilistic models from software systems, and (2.) to find the appropriate abstraction levels for the targeted software engineering activities.