Distributed Deep Learning System In Fog Computing Environment

As a core technology supporting modern intelligent applications,deep learning has been widely used in computer vision,speech recognition,and natural language processing due to its high accuracy and reliability.However,due to its huge computing overhead,traditional deep learning applications are often deployed on cloud computing centers to perform inference.Under this computing mode,a large amount of data is transmitted to the cloud center through a high-latency wide area network,which often causes a large communication delay.Fog computing,as a derivative technology of cloud computing,can effectively solve problems such as excessively long response time of the deep learning application and excessive pressure on the network transmission bandwidth by sinking the calculation from the center to the edge device.Based on the characteristics of the fog computing environment,this paper designs a solution for the implementation of deep learning models on edge devices,and completes the design and implementation of distributed deep learning systems in the fog computing environment.In order to deploy deep learning models to edge devices with limited resources,this paper proposes a distributed convolutional network search algorithm based on the Genetic-CNN algorithm.This algorithm searches for convolutional network models containing branch structures in order to deploy the models to Collaborative inference is performed on multiple edge devices to reduce the computing requirements of deep learning models for edge devices and to ensure the accuracy of the model as much as possible.At the same time,by adding computational complexity and structural constraints to the fitness function in the evolution process,the computational overhead of the model is reduced,and the model's representation ability is improved.Further,this paper also designs deployment algorithm of a distributed deep learning model based on genetic algorithms,in order to deploy the above model to a suitable computing device to reduce the response time of the calculation.Specifically,this article first introduces some research advances in deep learning applications in the fog computing environment,and then resolves key issues such as the design and deployment of deep learning models in the fog computing environment.Then based on this,the distributed deep learning described in this article is discussed.The functional requirements of the system designed an overall implementation plan,including the overall architecture,functional modules,and module interactions.The design and implementation of the core modules were explained in detail.Finally,a test environment was set up to complete the system's functional tests and give There are deficiencies and improvement directions.