An Analysis of Automated Software Diversity Using Unstructured Text Analytics

Automated software diversity promises to reduce an attacker's ability to reuse exploits across application instances. However, many questions remain regarding the efficacy of and application of software diversity. In particular, researchers have observed a lack of robust metrics to compare diversity strategies. Our work represents a step toward such metrics by using common methods from unstructured text analysis to differentiate strategies. Our investigation is agnostic to particular diversity strategies, and we analyze several methods of generating feature vectors from diversified binaries and comparing the resulting clusters in a high-dimensional space.