We present an extension to learning-based testing of systems for adversary-induced weaknesses that addresses the problem of repeated generation of known weaknesses. Our approach adds to the normally used fitness measure a component that computes the similarity of a test to known tests that revealed a weakness and uses this similarity to penalize new tests. We instantiated this idea to the testing of ad-hoc wireless networks using the IACL approach, more precisely to applications in precision agriculture, and our experiments show that our modification results in finding substantially different tests from the test(s) that we want to avoid.
Generated: 25/4/2017