Complex Networks Disintegration Based on Learning Automata

Abstract

Complex network disintegration stands as a paramount challenge within network science, playing a pivotal role in the mitigation of malicious network behaviour. Beyond its defensive role, it offers a strategy with broader applicability, encompassing risk prediction for networks with positive attributes. Complex networks, deeply rooted in graph theory, serve as a fundamental modelling framework across diverse problem domains, ranging from social networks, communications, and telecommunications to security, power distribution, information transmission, and even weather analysis with geographical implications. Yet, the disintegration of real-world networks carries tangible costs, necessitating the development of cost-effective methods a pressing concern when confronting such networks. Additionally, real-world networks often exhibit heterogeneity, mandating practical considerations in proposed solutions. Traditionally, complex network disintegration has relied on graph theory-based algorithms and heuristic methods. Recent years, however, have witnessed the incorporation of learning algorithms that engage dynamically with complex environments. Reinforcement learning, owing to its interactive nature with the environment, emerges as a well-suited methodology. Moreover, this paper introduces an innovative approach leveraging the Learning Automata algorithm to enhance existing disintegration strategies. This research explores the central role of complex network disintegration, bridging conventional graph theory techniques with cutting-edge reinforcement learning methods. The outcome is a more comprehensive and adaptable framework for addressing real-world network challenges, spanning defence against malicious networks with the optimized cost in unknown networks.