Neural network collaborative reasoning method for multi-access edge computing system

Abstract

The invention discloses a neural network collaborative reasoning method for a multi-access edge computing system, and belongs to the field of edge computing. In order to solve the problems that in a multi-access edge computing system, computing resources and power supply energy of access equipment are limited, and complex model reasoning is difficult to complete only by itself, a collaborative reasoning method for segmenting a neural network model and respectively putting the neural network model locally and executing the neural network model by a server is designed; a lightweight auto-encoder and a quantization technology are used to compress intermediate features, the amount of data needing to be transmitted through a wireless channel is reduced, and a deep reinforcement learning method is used to provide a model segmentation strategy, a wireless channel selection strategy and a transmitting power setting strategy for each access device. The edge server is fully utilized to assist the access device to complete the reasoning task under limited channel resources. According to the method, the average reasoning time delay and energy consumption of each access device can be effectively reduced, and support is provided for deployment of an intelligent application based on a complex model on a mobile device.

Description

technical field [0001] The invention relates to a neural network inference method, in particular to a neural network collaborative inference method for a multi-access edge computing system, belonging to the field of edge computing. Background technique [0002] With the rapid development of deep learning technology in recent years, Deep Neural Network (DNN) has achieved performance close to or surpassing human performance in many computer vision and natural language processing tasks. With the improvement of hardware performance of edge devices such as smartphones and watches, smart applications based on DNN models have become an indispensable part of mobile devices. However, mobile devices have far fewer computing resources than desktop computers, and are constrained by limited battery power to run only some applications that use simple DNN models. Cloud computing technology, which offloads computing tasks to cloud servers, can effectively improve the ability of mobile devi...

AI Technical Summary

Problems solved by technology

[0004] In view of the fact that it is difficult for the MEC system to directly combine with the cooperative reasoning method to assist the access device to complete the complex DNN model reasoning task, the main purpose of the present invention is to propose a neural network collaborative reasoning method for multi-access edge computing systems to realize the MEC system Collaborative reasoning in , reducing the reasoning delay and energy consumption of the user equipment

Images