Cryptography Based on Chaotic Neural Networks: A Critical Literature Review

Abstract

Chaotic neural network cryptography which combines the sensitivity and ergodicity of chaotic systems with the nonlinearity and parallelism of neural networks, and have all used to image encryption, pseudorandom number generation, and FPGA-based cryptosystems. Instead of this review synthesizes the field by idea do this by paper. The literature is generally in agreements that these systems have useful cryptographic properties, which are feasible on real-time hardware and share evaluation toolkit (NPCR, UACI, entropy, NIST SP 800-22). However, there is disagreement over whether passing statistical metrics establishes security.  Drawing on the cryptanalytic strand that has broken schemes which passed those metrics, we argue that a demonstrated attack outweighs favourable statistics, that claims of superiority over AES and of DNA-derived quantum resistance are minority positions unsupported by a security reduction, and that the field is moving toward maturity rather than expansion. The single most important open question is whether such a scheme can be given provable security reduced to an established hard problem, rather than statistical security alone.