Joint Job Assignment and Resource Allocation for Multi-Job Wireless Federated Learning

Wireless Federated Learning (WFL) marks a significant evolution in edge artificial intelligence (AI), allowing for collaborative learning while preserving the privacy of edge devices. Recently, the enhanced data collection and storage capability of edge devices have precipitated a paradigm shift from training an individual model (single-job) to multiple AI models (multi-job). This shift poses a new challenge in maintaining high system performance while managing resource management among multiple jobs. In this work, we address the challenge of a multi-job WFL framework by optimizing its dual efficiency. We formulate a multi-objective optimization problem with the goal of minimizing energy consumption and execution time to enhance system efficiency, while ensuring that all AI jobs meet their predefined performance criteria, thereby guaranteeing learning efficiency. To solve the problem, we propose an algorithm that jointly optimizes job assignment and resource allocation, by considering the triple-heterogeneity of data, devices, and jobs in the multi-job WFL framework. Specifically, a matching game-based method is utilized to assign jobs to capable devices, considering their respective contributions and costs, while convex optimization techniques are employed to refine the resource allocation toward computing frequency and transmission power. The performance of the algorithm is evaluated through numerical simulations in terms of system and learning performance over multiple AI model training jobs, showing that our proposed algorithm ensures time and energy efficiency under the same learning performance constraints.