Motivated by the urgent need for secure software, we construct new testing methods inspired by biology to improve current development life cycles. We connect probability theory with current testing technologies by formulating feedback-driven fuzzing in the language of stochastic processes. This mathematical model allows us to translate deep results from probability theory into algorithms for software testing. Exploring the full capabilities of our model leads us to the application of reinforcement learning methods, which turns out to be a fruitful new direction in software testing.

Chapter Contents:

• 13.1 Research challenge
• 13.2 The stochastic process of fuzzing
• 13.2.1 Fuzzing essentials
• 13.2.1.1 Fuzzing origins
• 13.2.1.2 Modern fuzzing
• 13.2.1.3 Generic architecture and processes
• 13.2.2 Markov decision processes
• 13.2.2.1 Policies and behavior
• 13.2.2.2 Value functions
• 13.2.3 Fuzzing as a Markov decision process
• 13.2.3.1 States
• 13.2.3.2 Actions
• 13.2.3.3 Rewards
• 13.3 Fuzzing with predefined behavior
• 13.3.1 Hunting bugs with Lévy flight foraging
• 13.3.1.1 Optimal foraging
• 13.3.1.2 Lévy flight hypothesis
• 13.3.1.3 Swarm behavior
• 13.3.2 Guiding a colony of Fuzzers with chemotaxis
• 13.3.2.1 Colonies with explorers
• 13.3.2.2 Chemotaxis
• 13.4 Fuzzing with learning behavior
• 13.5 Outlook
• 13.5.1 Hierarchies of learning agents
• 13.5.2 Alternative models
• 13.6 Conclusion
• References

Inspec keywords: stochastic processes; fuzzy set theory; learning (artificial intelligence); safety-critical software; program testing

Other keywords: life cycles development; reinforcement learning methods; feedback-driven fuzzing; software testing; mathematical model; nature-inspired fuzzing; probability theory; bugs hunting; stochastic processes

Subjects: Diagnostic, testing, debugging and evaluating systems; Combinatorial mathematics; Other topics in statistics; Data security; Knowledge engineering techniques; Software engineering techniques