Security analysis and new models on the intelligent symmetric key encryption

Data protection is achieved in modern cryptography by using encryption. Symmetric key cryptography is mainly responsible for the actual user data protection in various network protocols such as SSL/TLS and so on. The design of such encryption algorithms have always been one of the most important research targets, where heavy cryptanalysis works have been performed to evaluate the security margin. As a result, the research community is busy with fixing the security flaws based on the cryptanalysis results. Recently, the idea of building the automatic security protection scheme based on the neural network has been proposed. The encryption algorithm, which is a neural network is instead constructed by machine during the learning stage in an adversarial environment. This is a totally different approach compared with our current design principle, and could potentially change our understanding about how the (symmetric key) encryption works and what is the security requirement for the scheme. In this paper, we investigate the security of the underlined scheme which remains unexploited based on several statistical models. And furthermore, we strengthen the automatic encryption schemes by introducing much stronger adversaries. Our results showed that the security solutions based on the advanced deep learning techniques may start to play an important role in the future related directions.