Correntropy-Based Multistage Resilient Distributed Estimation Over Adversarial Multitask Networks Against Noisy Interference

This article studies robust and resilient distributed estimation in adversarial multitask wireless sensor networks (WSNs) under input noise and output impulsive interference. Under these conditions, malicious attacks and noises degrade network estimation performance by tampering with node parameters and disrupting communication. To resolve these challenges, we propose the neighboring node resilience acquisition-based secure multitask diffusion bias-compensated least mean M-estimate (NRAS-MDBCLMM) algorithm. To eliminate the influence of input noises, the MD-LMM algorithm is augmented with a bias compensation term that can estimate noise variance in real-time without prior knowledge. To mitigate the impact of malicious attacks in the multitask network, we introduce a node task similarity-based multistage reliable neighbor node resilience acquisition. First, we propose a neighboring node relative state perception and detection method using generalized correntropy. Second, a resilience-driven distributed detection framework is also developed for dynamic malicious node reduction. Finally, the performance of the proposed algorithm is analyzed and derived. Simulations confirm that the proposed NRAS-MDBCLMM algorithm outperforms state-of-the-art multitask estimation methods in both complex attack scenarios and impulsive interference environments.