基于联邦学习的边缘智能协同计算与隐私保护方法

In federated learning, each terminal transmits the updated model parameters instead of the original data to the server, which becomes the key technology to guarantee data security in edge computing. On this basis, a Federated Learning Based Edge Computing (FLBEC) method was proposed to preserve the users' privacy, while reducing the terminals' expense for federated learning. A system framework for edge computing based on federated learning was designed and a mechanism for privacy preserving was proposed. The learning time and energy consumption for terminals were analyzed, and the study target to preserve the users' privacy and reduce the learning time and energy consumption on the promise of guaranteeing accuracy was presented. The federated learning method was improved from the perspectives of participant selecting, local update and global aggregation. Comparative experiments were conducted to validate that there was a large amount of reduction on time and energy consumption for the majority of terminals in FLBEC by meeting the accuracy standards, which could abate the expense for federated learning and indicate the superiority of FLBEC.