Robot-oriented deep learning model segmentation method under cloud-edge-terminal framework

Abstract

The invention provides a robot-oriented deep learning model segmentation method under a cloud-edge-terminal framework, and belongs to the field of deep learning and distributed computing. According tothe method, firstly, a deep learning model is modeled into a directed acyclic graph, nodes of the directed acyclic graph represent deep learning model layers, and edges between the nodes represent data transmission between the deep learning model layers. Secondly, values are assigned to the nodes according to the processing time of the model layers on clouds, edges and terminals, and values are assigned to the edges between the nodes according to the transmission time of data between the model layers between the cloud edges, the edge terminals and the cloud terminals. Then, the nodes in the graph are layered by adopting a directed acyclic graph longest distance algorithm, and the nodes are processed layer by layer. For each node in one layer, a heuristic strategy is adopted to perform dynamic segmentation according to the input edge weight and the node weight of the node, the segmented deep learning model is distributed to cloud-edge-terminal computing equipment, and therefore cloud-edge-terminal distributed collaborative reasoning without precision losses is achieved.

Description

technical field [0001] The invention relates to the field of deep learning and distributed computing, in particular to a method for segmenting deep learning models under a robot-oriented cloud-edge-end architecture. Background technique [0002] In modern computer applications, deep learning models are widely used in many fields such as machine vision, natural language processing, and data mining. However, the deep learning model requires a lot of computing power, and it is difficult for a single machine to complete the inference process of the deep learning model within the specified time to meet the service level agreement; The geometric progression grows, and the cloud collaboration framework dominated by cloud computing is also difficult to meet the needs of data processing, resulting in data accumulation at the terminal, blocking during transmission, and slow return after processing in the cloud. Because the cloud-edge-device collaborative computing paradigm in edge co...

AI Technical Summary

Problems solved by technology

[0004] 1. The current technology does not make full use of the computing power of the terminal side, and only divides the deep learning model into two parts. After the data is generated on the terminal side, the data is directly uploaded to the side for processing, and the intermediate results are uploaded after the side processing Give the cloud side processing to produce the final result

[0005] 2. Existing methods cannot divide the deep learning model into three parts according to the processing time of the deep learning model layer and the transmission time of data

[0006] 3. When the computing power of the cloud-side device changes, or when the network environment changes, which affects the processing time of the deep learning model and the data transmission time between the cloud-side device, the current technology needs to re-segment the entire deep learning model, which consumes longer time

Images