| With the deep integration of the Internet of Vehicles and artificial intelligence,more and more artificial intelligence applications have landed in the Internet of Vehicles,such as pedestrian detection,fatigue monitoring,and automatic parking.In the artificial intelligence application of the Internet of Vehicles,machine learning technology is mainly used,and the application of machine learning technology requires the training of the model first.In traditional machine learning model training schemes,model training is generally performed on a central server.This requires a large amount of user data to be transmitted to the central server,and data processing and model training are performed on the server.This centralized training method can effectively improve the performance of machine learning algorithms,but the collection process may lead to the leakage of user privacy and collecting a large amount of data is very difficult.In view of the above background,this article is mainly based on the research of federated learning algorithms,designing and implementing a joint learning system oriented to the edge intelligence of the Internet of Vehicles.In this system,the vehicle terminal obtains the global model from the server and uses local data for model training,and then uploads the trained model to the server.The server aggregates the models uploaded by the vehicle to obtain a new global model.In the process of joint training,the vehicle terminal needs to upload the model multiple times.Multiple model uploads will consume a lot of bandwidth resources and affect the overall joint training progress.Therefore,this paper also conducts research on model compression algorithms to achieve compression of model parameters in the process of joint learning of the system,improve the upload speed of the model on the vehicle side,and speed up the global joint learning.This article will elaborate on the design and implementation of the joint learning system,test the system and analyze the result.The main tasks completed in this paper are as follows:(1)Based on the Python and the Pytorch,we build the vehicle-mounted platform in the joint learning system.And realizing the basic functions of model acquisition,model training,and model upload.(2)Based on the Java and Spring Boot,we build the server platform in the joint learning system.And realizing the basic functions of global training management,model distribution and model integration.(3)Research and apply model compression algorithms,and design joint learning optimization schemes under heterogeneous communication resources to achieve effective compression of model parameters during joint learning.(4)Perform functional testing and overall testing of the system,focus on testing the application effect of model compression technology in the joint learning system,and perform a detailed analysis of the test results... |
