Adaptive Resilience via Knowledge Distillation: Safeguarding Systems from Leader Missing Attacks

The resilience of leader-follower structures has been a prominent topic in both academic and industrial research. Traditional studies have largely focused on maintaining follower consistency assuming the leader remains completely functional, neglecting the potential system paralysis caused by an attack on the leader. Recently, knowledge distillation has emerged as a promising approach to mitigate issues when leaders go offline, transferring knowledge from a large teacher model to a smaller student model. In this paper, we introduce a novel resilient framework utilizing Long Short-Term Memory (LSTM) neural networks for knowledge distillation, designed to respond to scenarios where central coordination is compromised due to an attack, leading to system paralysis. The key concept is to deploy LSTM neural networks on agents, allowing them to learn and internalize the rules governing the system under the leader's guidance. In the event of the leader going offline, these agents can continue performing tasks autonomously, using the learned rules to prevent system paralysis.