Research On Intelligent Scheduling Mechanism Of In-network Cache For Edge Ensemble Learning

With the development of the Internet,the data traffic in the network presents an exponential growth trend,and the new data-intensive applications have higher and higher requirements for the data processing capacity of the edge network.The traditional cloud computing method is to upload the data generated by terminal devices to the cloud data center for data-intensive intelligent processing in the cloud data center.Considering the massive data transmission in this process,this centrali zed processing mode is limited by network bandwidth and network security,which will affect the quality and speed of data processing.In order to solve these problems,some data can be processed on the edge network by combining the computing storage resources of the edge network.Edge computing arises at the right moment and has become the core research hotspot in the data-intensive application field.Edge network nodes have limited computing and storage resources,which has become the key to restrict data processing and learning on edge network.In the process of mass data processing and learning,in order to ensure the data learning performance,it is necessary to schedule the computing and storage resources of edge network nodes reasonably.Ensemble learning is a common learning performance improvement method.Aiming at this learning method,this paper studies the cache intelligent scheduling mechanism in the network to ensure the performance and efficiency of the edge network ensemble learning.Ensemble diversity(differences between sub-models)is a key factor affecting ensemble learning performance.In this paper,we propose an intelligent scheduling mechanism of in-network cache for edge ensemble learning,which can utilize the storage resources on edge devices and the distributed computing method to realize the efficient collection,processing and analysis of user data.To be specific,firstly,the cached data of each edge node is compacted and recorded,and then the adaptive collaborative cache scheduling method is used to cache the differentiated data of each edge node according to the records.Secondly,the different cached data are used in each edge computing node to conduct distributed learning of sub-models,and the differentiated sub-model is obtained.Finally,the parameters of sub-models are uploaded to the central node to ensemble sub-models.The cache intelligent scheduling mechanism proposed in this paper can reasonably schedule the cache data of edge nodes according to the training data distribution of the model,improve the differences of sub-models,and then improve the performance of ensemble learning.At the same time,an edge ensemble learning framework is constructed based on the proposed collaborative cache scheduling mechanism,and the design and implementation of edge network discrete event generator and distributed deep learning module are mainly studied,thus forming a complete edge ensemble learning framework.To verify the performance of cache scheduling and edge ensemble learning,an edge ensemble learning simulation framework based on a discrete event emulator is designed and implemented by ensemble an advanced deep learning framework.On the simulation platform,an experimental strategy was developed for edge cache scheduling and ensemble learning,and the effectiveness of the proposed mechanism was verified based on four evaluation indexes(model accuracy,training latency,network data transmission overhead and cache hit ratio).Through comparative analysis,the mechanism proposed in this paper can effectively reduce the data transmission overhead and learning latency in the process of edge network ensemble learning while ensuring the accuracy of learning,which has certain theoretical and practical significance for the application of ensemble learning in edge networks.