Research On Key Technologies Of Federated Learning Optimization Training For Edge Computing

Artificial intelligence technology has achieved great success in application fields such as computer vision,natural language processing,and pattern recognition.But traditional machine learning methods often require data to be centralized in data centers.It has problems such as user privacy leakage and low communication efficiency.AI technologies face a lack of highquality data due to increasingly stringent data privacy legislation.Federated learning enables users to jointly model without sharing the original data.Because of its ability to protect data privacy,it has gradually become a research hotspot in the field of artificial intelligence.Starting from the above background,this thesis studies the key techniques of federated learning optimization training.In this thesis,the research work is mainly carried out from two aspects:heterogeneous optimization of federated learning participating devices and communication cost optimization of federated learning.In federated learning,edge devices have differences in communication capabilities,computing capabilities,and local data distribution.It affects the model training efficiency and model accuracy.Aiming at the above problems,this thesis proposes a federated learning optimization algorithm for participating in device heterogeneity.The proposed algorithm considers the data heterogeneity and system heterogeneity in the federated learning environment.It selects devices based on device data quality and computing power,improving the training efficiency and convergence accuracy of federated learning.The experimental results show that the proposed algorithm can improve the model training efficiency and model convergence accuracy.In federated learning,participants and the federated learning server need to complete multiple rounds of communication before the global model reaches the target accuracy.However,deep learning models may contain millions of parameters when updated.This results in high communication costs.Aiming at the above problems,this thesis proposes a federated learning communication overhead optimization algorithm.The proposed algorithm adaptively adjusts the local update rounds by sensing the training status of different training stages.It improves the final convergence accuracy of the model while ensuring the training efficiency of the federated learning model.The experimental results show that the proposed algorithm can greatly improve the convergence speed of the model.