A Decentralized Device Collaborative Deep Learning Inference Method

Abstract

The invention discloses a method for distributed deployment of a cache-based deep neural network on decentralized edge devices. The method first divides the neural network, and prunes the previous layer of neural network at the division, then calculates a part of the deep neural network in the task initiating device, transmits a small amount of intermediate results to other edge devices, and calculates the remaining part, In addition to caching and reusing the intermediate results of the edge device neural network, the cache can be shared between different devices, thereby reducing the delay of edge intelligent applications and reducing the performance requirements of the neural network on edge devices, especially when similar data are initiated on the edge side. When intelligent tasks are requested, the amount of repeated calculations can be reduced, the performance requirements of deep learning on devices can be reduced, and computing resources in edge scenarios can be fully utilized.

Description

technical field [0001] The invention relates to the field of artificial intelligence and edge computing in computer science, in particular to a deep learning inference method combining decentralized equipment collaborative computing and caching. Background technique [0002] As an emerging computing paradigm, edge computing aims to utilize the computing and communication resources of edge devices to meet the needs of users for real-time response to services, privacy and security, and computing autonomy. Driven by the rapid development of algorithms, computing power and big data, as the most active field in artificial intelligence, deep learning has made significant progress in many fields. With the development of the Internet of Things and cyber-physical systems (CPS), new applications such as autonomous driving, intelligent drone formation, and intelligent robot clusters have driven the integration of edge computing and artificial intelligence, and promoted the emergence an...

AI Technical Summary

Problems solved by technology

[0007] These methods can accelerate the deep learning model in a certain way, but they cannot make full use of the characteristics of the interconnection between edge devices, nor do they take into account that the input data of edge devices that are likely to perform intelligent tasks within a period of time is relatively similar or even repeated.

Images