HDCCSR: Software Self-Awareness Using Dynamic Analysis and Markov Models

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

Long-lived, autonomous software systems require ongoing self- assessment, which implies a comparison between expected and actual stimuli, as a prerequisite to self-diagnosis and repair. This project addresses the use of dynamic analysis and machine learning techniques for behavior classification. The proposed work will - address fundamental research and education issues in dependable software-based computing systems by exploring how machine learning techniques can most effectively be applied to improve the self-assessment process; - develop research products in the form of prototype tools or methodologies, applying Markov models and cluster analysis to the assessment of program behavior data; - provide dependability attributes that are suitable for measuring the impact of the research products, such as the extent to which a set of Markov models successfully describes the behavior of a software system; - provide empirical evaluation/validation of the research products together with associated test data to validate the effectiveness of the approach The proposed research will provide empirical data on the use of Markov models to encode behavioral models, and methodologies and infrastructure for use in performing further experimentation. The resulting techniques for provisioning software with behavior models to aid in self-awareness will promote research on development of real-time systems.