Automated software vulnerability testing using deep learning methods

The paper provides a state of current technologies, which are used for automated vulnerability testing in software. Features of fuzzing technology (which is based on making many inputs with different mutated data) are also studied. The essence of the algorithm is the selection of input data, which are more likely to cause a failure or incorrect behavior of the software product. Deep learning algorithms are used to reduce the computational complexity of the testing process. Using a simple fuzzer and Deep Reinforcement Learning algorithm shows that the number of mutations required to detect vulnerabilities is reduced by 30%.