Multi-user deep neural network model segmentation and resource allocation optimization method in edge computing scene

Abstract

The invention discloses a multi-user deep neural network model segmentation and resource allocation optimization method in an edge computing scene, and the method comprises the steps: comprehensively analyzing the execution characteristics of a deep neural network model segmentation technology in an edge computing environment; modelling a combination optimization problem of deep neural network model segmentation and computing resource allocation on an edge server into a nonlinear integer programming problem, and further providing an iterative alternative optimization algorithm based on dynamic step length adjustment. The algorithm not only can efficiently solve the optimal solution of the problem in the polynomial time, but also has the characteristic of high robustness for various external influences in a real deployment scene.

Description

technical field [0001] The present invention relates to the technical fields of deep learning, edge computing and distributed computing, and more specifically, to a multi-user deep neural network model segmentation and resource allocation optimization method in an edge computing scenario. Background technique [0002] With the gradual popularization of 5G technology and the continuous development of technologies such as mobile artificial intelligence and the Internet of Things (IoT for short), the number of devices at the edge of the network has ushered in explosive growth. At the same time, terminal devices at the edge of the network are gradually transitioning from the role of consumers of smart applications to special nodes that are both consumers and producers, and continue to generate massive amounts of real-time data during operation. However, the traditional mobile cloud computing method is limited by the transmission bandwidth of the backbone network and the high tra...

AI Technical Summary

Problems solved by technology

[0004] The first idea is to optimize the model itself, which has the advantage of not requiring other additional equipment support, but it is also accompanied by several difficult problems: First, the accuracy of the model after compression is usually difficult to guarantee theoretically, and not all All models are suitable for compression; the second is that structure-based weight pruning cannot be applied to all models, and general weight pruning that is not based on structure will hinder high-performance parallel optimization at the hardware level

However, the existing methods are all considered from the perspective of a single user, usually assuming that the resources on the corresponding cloud server or edge server are static and fixed

However, the actual deployment scenarios usually involve resource-constrained multi-user scenarios, and the challenges of distributed deployment are more complex, and the existing methods fail to provide low-cost solutions.

Images