Neuro-Symbolic Twins of Software Systems and Their Use in Solving Cybersecurity Problems

This paper explores the use of software system twins, represented as a combination of an algebraic model and a neural network. The application of this technology is examined both in intrusion detection systems and at the stage of assessing the reliability of cyber defense in software systems. The proposed approach can significantly enhance the accuracy of real-time cyberattack detection, improve resilience against adversarial attacks, and reduce false positives — a common challenge in detecting attacks with unknown semantics. The use of a neuro-symbolic twin at the stage of system preparation for operation has been analyzed, specifically, the procedure for detecting system vulnerabilities. Furthermore, the architecture of an intrusion detection system based on a neuro-symbolic twin is presented, featuring capabilities for monitoring incoming communication protocols and restoring the software environment. Examples of the technology’s implementation in blockchain environments and hardware security are provided.