Deploying Graph Neural Networks in Wireless Networks: A Link Stability Viewpoint

TL;DR

This work tackles deploying GNNs in wireless networks where fading, noise, and energy constraints degrade training. It formulates a stochastic optimization to maximize the long-term average number of activated links $ ar N $ under the energy constraint $ rac{1}{T} extstyle\sum_t E_u(t)\le ar E_u^{ ext{max}}$, and uses Lyapunov optimization to convert it into a per-slot deterministic problem that balances link activation against energy usage via the drift-plus-penalty $ oldsymbol riangle_V(t) = ext{E}[-V N(t) + extstyleoldsymbol Z(t) extstyle E(t) | oldsymbol Z(t)] $. A greedy-based solver transforms the resulting convex feasibility subproblems in each time slot into a sequence of tractable steps, including sub-slot partitioning, link sorting by $C_{uv}(t)$, and convex feasibility checks with constraints of the form $|h_{uv}|^2 P_u(t)- extstyle\sum_{k eq u} |h_{kv}|^2 P_k(t) G geq G ilde{ u}^2$, where $G=2^{D_u/(B au^{ ext{max}})}-1$. The approach yields stable energy usage, more activated links, and improved GNN test accuracy compared with baselines, demonstrating the practicality of robust, resource-aware wireless GNN training. The framework enables decentralized, energy-constrained learning by coupling TDMA-like scheduling with convex optimization in each slot. Overall, it offers a rigorous method to sustain GNN performance in realistic wireless environments.

Abstract

As an emerging artificial intelligence technology, graph neural networks (GNNs) have exhibited promising performance across a wide range of graph-related applications. However, information exchanges among neighbor nodes in GNN pose new challenges in the resource-constrained scenario, especially in wireless systems. In practical wireless systems, the communication links among nodes are usually unreliable due to wireless fading and receiver noise, consequently resulting in performance degradation of GNNs. To improve the learning performance of GNNs, we aim to maximize the number of long-term average (LTA) communication links by the optimized power control under energy consumption constraints. Using the Lyapunov optimization method, we first transform the intractable long-term problem into a deterministic problem in each time slot by converting the long-term energy constraints into the objective function. In spite of this non-convex combinatorial optimization problem, we address this problem via equivalently solving a sequence of convex feasibility problems together with a greedy based solver. Simulation results demonstrate the superiority of our proposed scheme over the baselines.

Figures (3)

• Figure 1: Wireless graph neural network system. (a) Node transmission in the system. (b) Orthogonal access in the time division system, where a slot is divided into sub-slots, and each node can only receive or transmit information in each sub-slot.
• Figure 2: Average virtual queue backlogs with different $V$ values and the accumulative number of links on Cora, respectively.
• Figure 3: The test accuracy of our proposed algorithm and baseline algorithms. (a) and (b) show the test accuracy on Citeseer and Cora, respectively.