A Collaborative Inference Method of Deep Neural Network Based on Device-Edge-Cloud Architecture
A deep neural network and inference method technology, applied in the field of deep neural network model acceleration and optimization, can solve the problem of incomparable edge server network environment and resource deployment, and achieve accelerated computing execution process, wide application range, reduced latency and The effect of energy consumption
Active Publication Date: 2022-05-06
ZHEJIANG LAB
View PDF3 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
But in general, the network environment and resource deployment of the edge server cannot be compared with the cloud system
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment
[0065] The core idea of the embodiment of the present invention is to use the heterogeneity of computing hardware resources of the terminal, edge and cloud nodes to solve the high real-time and low energy consumption requirements of the terminal in certain application scenarios, such as unmanned driving scenarios. Road feedback requires millisecond-level decision-making and reaction. The concrete effect of embodiment is with figure 2 Take the convolutional neural network as an example to illustrate:
[0066] After the data set reasoning evaluation on the terminal, edge side and cloud side, according to figure 2 The convolutional neural network model, the amount of layered data and the network delay are as follows:
[0067] The data volume of Conv1_1 is 3.2M, and the computing execution time of the terminal, edge, and cloud side are 4ms, 2ms, and 2ms respectively;
[0068] The data volume of Conv1_2 is 3.2M, and the computing execution time of the terminal, edge, and cloud...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention discloses a deep neural network collaborative reasoning method based on the device-edge-cloud architecture. The method accelerates the device-side reasoning speed by means of device-edge-cloud collaboration, and uses the deep model in artificial intelligence according to the layering of the neural network. Carry out segmentation, and send the computing tasks in the model reasoning process to the corresponding device side according to the network environment, resource quota and usage of the device, edge and cloud, and complete the whole process of reasoning. The invention discloses the overall framework of model segmentation and the algorithm components and principles used in the segmentation calculation tasks. Through the collaboration of the device, edge and cloud, the reasoning speed of the device can be accelerated, the real-time performance of the business scene can be improved, and the energy of the resource end can be reduced at the same time. consumption.
Description
technical field [0001] The invention belongs to the field of deep neural network model acceleration and optimization, and in particular relates to a deep neural network collaborative reasoning method based on a device-edge-cloud architecture. Background technique [0002] In recent years, deep learning has achieved great success in applications such as machine vision, natural language processing, and big data analysis. Through the method of deep learning, the performance in image classification and object recognition is better than traditional methods. However, the high accuracy of deep learning comes at the cost of high computational and memory demands for deep learning training and inference. Some trained deep neural network models have millions of parameters. During the inference process, the input data needs to undergo millions of calculations. High accuracy and high resource consumption are the hallmarks of deep learning. In resource-constrained edge scenarios, the d...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Patents(China)
IPC IPC(8): G06N3/04G06N3/08G06N5/04G06F9/50
CPCG06N3/08G06N5/04G06F9/5072G06N3/045Y02D10/00
Inventor 梁松涛高丰杨涛施佩琦汪明军郁善金王晓江
Owner ZHEJIANG LAB