Resilient Distributed Multi-task Estimation with Network Learning under Malicious Attacks

Network security plays a critical role in distributed estimation. When a multi-task distributed network is subjected to adversarial attacks, normal nodes are forced to both detect malicious nodes and learn cluster membership. The existing distributed algorithms, which mainly rely on some prior information and operate mechanically, remain insufficient and lack adaptability. To solve these problems, an adaptive distributed multi-task algorithm with cluster learning is proposed, which can make all normal nodes achieve resilience and secure estimation. In the proposed algorithm, an adaptive detection mechanism without any prior information is firstly proposed to identify the difference among nodes. Subsequently, a node queue strategy is designed to enable all normal nodes to obtain resilience and cluster learning. The convergence behavior of the proposed algorithm is analyzed, and then a convergence condition is derived to ensure the mean stability and mean-square convergence. Finally, based on different forms of data tampering and forgery, simulations consider three typical and distinct attacks. The results show that the proposed algorithm, without using any prior information, outperforms state-of-the-art algorithms, demonstrating the superior adaptability and effectiveness.