Cryptography using artificial intelligence

• Published in: 2015 International Joint Conference on Neural Networks (IJCNN) pp. 1 - 6
• Main Authors: Blackledge, Jonathan, Bezobrazov, Sergei, Tobin, Paul
• Format: Conference Proceeding Journal Article
• Language: English
• Published: IEEE 01.07.2015
• Subjects: Algorithms; artificial intelligence; artificial neural networks; coding and encryption; Complexity theory; Cryptography; Electronics; Encryption; Entropy; evolutionary computing; Lyapunov exponents; multiple algorithms; Neural networks; Neurons; Noise; personalised encryption engines
• ISSN: 2161-4393; 2161-4407
• DOI: 10.1109/IJCNN.2015.7280536

This paper presents and discusses a method of generating encryption algorithms using neural networks and evolutionary computing. Based on the application of natural noise sources obtained from data that can include atmospheric noise (generated by radio emissions due to lightening, for example), radioactive decay, electronic noise and so on, we `teach' a system to approximate the input noise with the aim of generating an output nonlinear function. This output is then treated as an iterator which is subjected to a range of tests to check for potential cryptographic strength in terms of metric such as a (relatively) large positive Lyapunov exponent, high information entropy, a high cycle length and key diffusion characteristics, for example. This approach provides the potential for generating an unlimited number of unique Pseudo Random Number Generator (PRNG) that can be used on a 1-to-1 basis. Typical applications include the encryption of data before it is uploaded onto the Cloud by a user that is provided with a personalized encryption algorithm rather than just a personal key using a `known algorithm' that may be subject to a `known algorithm attack' and/or is `open' to the very authorities who are promoting its use.