Meta-Learning-Driven GFlowNets for 3D Directional Modulation in Mobile Wireless Systems

In our prior work we have proposed the use of GFlowNets, a generative AI (GenAI) framework, for designing a secure communication system comprising a time-modulated intelligent reflecting surface (TM-IRS). However, GFlowNet-based approaches assume static environments, limiting their applicability in mobile wireless networks. In this paper, we proposes a novel Meta-GFlowNet framework that achieves rapid adaptation to dynamic conditions using model-agnostic meta-learning. As the communication user is moving, the framework learns a direction-general prior across user directions via inner trajectory-balance updates and outer meta-updates, enabling quick convergence to new user directions. The approach requires no labeled data, employing a pseudo-supervised consistency objective derived from the learned reward by GFlowNet and the actual sum-rate reward of the TM-IRS system. Simulation results show that the proposed method attains faster adaptation and higher secrecy performance than retrained GFlowNets, offering an efficient GenAI framework for dynamic wireless environments. Although the scenario considered here focuses on directional modulation-based physical-layer security, the proposed framework can also be applied to other mobile wireless systems, such as joint sensing-communication networks, that utilize GFlowNets.