GNN-Mamba model evaluates collaborator trust with 94% accuracy

Selecting trustworthy collaborators is critical for complex multi-device tasks, but current methods struggle to combine diverse trust signals—historical behavior, current capability, and spatiotemporal dependencies. Botao Zhu and Xianbin Wang from IEEE ICC 2026 propose a hybrid GNN-Mamba model that addresses this by jointly modeling spatial and temporal trust dimensions. The Graph Neural Network (GNN) component fuses trust information across devices based on historical interaction graphs, capturing inter-device dependencies. Meanwhile, the Mamba-based temporal model—a state-space architecture known for handling long sequences efficiently—tracks both short-term fluctuations and long-term evolution of device trust over time. Additionally, task-specific resource trust is incorporated to reflect a device's practical capability under varying conditions.

Experimental results demonstrate that the proposed GM model significantly outperforms baseline approaches—including standalone GNNs, LSTMs, and transformers—in both accuracy and stability of trust evaluation. The model achieves a reported 94% accuracy in predicting trustworthy collaborators across simulated cooperative tasks. By combining spatial graph reasoning with efficient temporal modeling, the GNN-Mamba offers a practical solution for autonomous collaborator selection in domains like drone swarms, edge computing, and IoT networks. This work was submitted to arXiv in May 2026, and accepted for IEEE ICC 2026, signaling strong peer validation.