Research On Federated Learning Model And Algorithm Based On Mobile Edge Computing

With the rapid development of Mobile Internet,a large number of mobile applications,such as Social Networking,We-Media,e-Commerce and so on,have gradually integrated into people's lives and become an inseparable part of daily life.This also leads to the rapid growth of the amount of data living on the edge of the network.Massive data transmission and storage bring huge pressure to the transmission bandwidth and cloud storage.The traditional cloud centric centralized data storage and processing scheme is becoming more and more infeasible,which brings a new data processing method called Mobile Edge Computing(MEC).Mobile edge computing is a distributed data processing method.When Mobile Edge Computing nodes cooperate in data processing,the problem of data privacy leakage may occur.With consumers paying more and more attention to privacy protection and the introduction of data privacy protection laws,the development of mobile edge computing is facing great challenges.Federated Learning provides a machine learning protocol for collaboration and security,which can be used as the operating system of Mobile Edge Computing.Federated Learning can be used between Mobile Edge Computing nodes to train AI models under the premise of protecting data privacy.NGBoost(Natural Gradient Boosting)is a new data processing method,which can make probability prediction and be easy to train.It has excellent performance on small dataset,but it does not support distributed environment.Based on this,this paper studied the Vertical Federated NGBoost and Horizontal Federated NGBoost in Mobile Edge Computing.In the era of Internet,multi-modal data,such as image,text,video and audio,has become the main carrier of information dissemination and an important medium for human beings to perceive and understand the world.Therefore,multi-modal data processing technology has become a hot research issue.If we use the traditional method to deal with multi-modal data,we must use different neural networks for different modal data,which will greatly increase the complexity of the system and be not conducive to the actual production environment.Moreover,in the traditional data processing methods,we need to gather all the modal data together,which will lead to serious data privacy leakage.Therefore,in this paper,we also studied the Multi-Modal Data Classification Model based on Mobile Edge Computing.Based on this,the research work of this paper are mainly reflected in the following three aspects:First of all,we studied the Vertical Federated NGBoost Model based on Mobile Edge Computing,studied how to use the Vertical Federated Learning Technology to jointly train the NGBoost model between Mobile Edge Computing nodes,and studied how to use the trained model to predict the input.Simulation experiments were carried out to verify the effectiveness of the proposed Vertical Federated NGBoost Model.Then,we studied the Horizontal Federated NGBoost Model based on Mobile Edge Computing,studied how to use the Horizontal Federated Learning Technology to jointly train the NGBoost Model between Mobile Edge Computing nodes,and studied how to use the trained model to predict the input.Simulation experiments were carried out to verify the effectiveness of the proposed Horizontal Federated NGBoost Model.Finally,we studied the Multi-Modal Data Classification System based on Mobile Edge Computing.The research is divided into the following four parts.Firstly,the model architecture of Multi-Modal Data Classification System based on Mobile Edge Computing was studied.Secondly,we studied how to preprocess the dataset.Thirdly,how to realize the joint training of multi-modal data classification model between Mobile Edge Computing nodes using Federated Learning Technology was studied.Finally,simulation experiments were carried out to verify the effectiveness of the scheme.