OPTIMIZING INTELLIGENT EDGE COMPUTING RESOURCE SCHEDULING BASED ON FEDERATED LEARNING

作者：Hanzhe Li, Shiji Zhou, Bo Yuan, Mingxuan Zhang · 发表于：Online (Weston, Conn.) · 年份：2024 · DOI：10.60087/jklst.vol3.n3.p.235-260 · 被引用次数：19

This study proposes a novel federated learning framework for optimizing intelligent edge computing resource scheduling. The framework addresses the challenges of device heterogeneity, non-IID data distribution, and communication overhead in edge environments. We introduce an adaptive client selection mechanism considering computational capabilities, energy status, and data quality. A personalized model training approach is implemented to handle non-IID data effectively using multi-task learning and local batch normalization layers. The framework incorporates efficient model aggregation techniques and communication-efficient updates to reduce bandwidth consumption. The privacy policy, including the difference between privacy and collective security, has been integrated to improve data protection. We develop scheduling problems based on multi-objective optimization, combining the best in computing and communication while updating local and global guidelines. Extensive testing on a wide range of data shows that the framework is superior regarding connection speed, resource utilization, and model performance. The proposed method achieves a 15% improvement in model accuracy and a 40% reduction in communication overhead compared to learning state-of-the-art algorithms. Case studies in intelligent city traffic prediction and healthcare IoT validate the framework's effectiveness in real-world scenarios, showcasing its scalability and adaptability to varying network conditions and cli...