New MASL Algorithm Cuts AIGC Service Delays in MEC Networks

A team of researchers has introduced a novel game-theoretic approach to optimize scheduling for AI-generated content (AIGC) at the network edge. Their paper, published on arXiv and accepted by the IEEE Internet of Things Journal, tackles the challenge of deploying generative diffusion models (GDMs) on mobile edge computing (MEC) servers. These servers must balance latency and accuracy requirements from multiple users with heterogeneous demands. The authors formulate a Joint Communication Association and Computation Offloading (JCACO) game where each user selects its access point, edge server, and number of inference steps to minimize overall service completion time while meeting accuracy constraints.

The key contribution is a distributed Multi-Agent Stochastic Learning (MASL) algorithm that provably converges to a Nash equilibrium. Unlike traditional best-response schemes, MASL does not require knowledge of other players' strategies or global network information, making it fully distributed and adaptive to dynamic environments. The authors prove that the JCACO game is a potential game under both complete and stochastic information, guaranteeing equilibrium existence. They also provide a strict theoretical convergence analysis using ordinary differential equations (ODEs).

Simulation results demonstrate that MASL significantly reduces service completion time compared to benchmark methods while satisfying accuracy constraints. This is critical for real-world MEC-enabled AIGC networks, such as smart city applications where real-time content generation must meet strict latency requirements. The fully distributed nature of MASL makes it scalable and robust to network changes, promising practical deployment.

By combining game theory with stochastic learning, this work advances the efficient operation of edge AI systems. It addresses a key bottleneck in AIGC—the need for low-latency computation near the user—without requiring centralized control. As generative AI continues to proliferate at the edge, algorithms like MASL will be essential for delivering responsive and accurate services.