Neural network model real-time automatic quantification method and real-time automatic quantification system

A neural network model and network model technology, which is applied in the field of real-time automatic quantification of neural network models based on embedded AI accelerators, and real-time automatic quantification of neural network models. problems, such as real-time reasoning of type equipment, difficulty in training high-precision network models, etc.

Pending Publication Date: 2021-03-05
SENSLAB INC
View PDF0 Cites 8 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Therefore, limited by the resources of the embedded end, it is difficult to train a high-precision network model on the embedded end. In addition, even if a high-precision network model is trained, it is difficult to meet the real-time reasoning of embedded devices at the inference end of the network. It will aggravate the power consumption and heat dissipation of embedded devices

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 more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Neural network model real-time automatic quantification method and real-time automatic quantification system
  • Neural network model real-time automatic quantification method and real-time automatic quantification system
  • Neural network model real-time automatic quantification method and real-time automatic quantification system

Examples

Experimental program
Comparison scheme
Effect test

no. 1 example ；

[0086] refer to figure 1 As shown, the present invention provides a kind of real-time automatic quantification method of its neural network model based on embedded AI accelerator, comprising the following steps:

[0087] S1, conduct embedded AI neural network training on the PC side, build a deep learning neural network on the PC side, and train the input floating-point network model of the embedded AI model;

[0088] S2, the quantization of the PC-side network model, which quantifies the floating-point network model into an embedded-side fixed-point network model;

[0089] S3, the embedded AI accelerator automatically quantifies in real time, preprocesses the data that needs to be quantified, and realizes all acceleration operators of each layer of the model network through hardware mode;

[0090] S4, Embedded AI hardware deployment on the embedded side and neural network model transplantation on the embedded side, transplanting the neural network model to the built AI hardw...

no. 2 example ；

[0092] The present invention provides a kind of real-time automatic quantification method of its neural network model based on embedded AI accelerator, comprising the following steps:

[0093] S1, carry out the embedded AI neural network training on the PC end, build the deep learning neural network on the PC end, train the floating-point network model of the embedded AI model input, and include the following sub-steps when implementing step S1;

[0094] S1.1, according to the specific scenario of the embedded AI application at the embedded end, analyze the requirements of the embedded AI application, and collect the data sets required for network training;

[0095] S1.2, build the floating-point network model of the embedded AI model of the deep learning neural network training input on the PC side;

[0096]S2, PC-side network model quantification, the floating-point network model is quantified into an embedded-side fixed-point network model, and the implementation of step S2...

no. 3 example ；

[0111] refer to figure 2 As shown, the present invention provides a kind of real-time automatic quantification method of its neural network model based on embedded AI accelerator, comprising the following steps:

[0112] S1, carry out the embedded AI neural network training on the PC end, build the deep learning neural network on the PC end, train the floating-point network model of the embedded AI model input, and include the following sub-steps when implementing step S1;

[0113] S1.1, according to the specific scenario of the embedded AI application at the embedded end, analyze the requirements of the embedded AI application, and collect the data sets required for network training;

[0114] S1.2, build the floating-point network model of the embedded AI model of the deep learning neural network training input on the PC side;

[0115] S1.3. Evaluate the trained floating-point network model on the PC side, and output the floating-point network model to the network model qua...

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

No PUM Login to view more

Abstract

The invention discloses a neural network model real-time automatic quantification method, which is based on an embedded AI accelerator, and comprises the following steps: carrying out embedded AI neural network training at a PC end, establishing a PC end deep learning neural network, and training an input floating point network model of an embedded AI model; quantizing the floating point network model into an embedded end fixed point network model; preprocessing data needing to be quantized, and realizing all acceleration operators of each layer of the model network through a hardware mode; deploying embedded AI hardware of the embedded end and transplanting the neural network model of the embedded end, and transplanting the neural network model of the built AI hardware platform. The invention further discloses a neural network model real-time automatic quantification system. According to the invention, algorithm acceleration is realized based on an embedded AI accelerator hardware mode, the storage occupied space of a neural network model can be reduced, the operation of the neural network model can be accelerated, the computing power of embedded equipment can be improved, the operation power consumption can be reduced, and the effective deployment of the embedded AI technology can be realized.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence, in particular to a real-time automatic quantification method of a neural network model based on an embedded AI accelerator. The invention also relates to a real-time automatic quantification system of a neural network model based on an embedded AI accelerator. Background technique [0002] With the rapid development of artificial intelligence technology, the current embedded development tends to be more intelligent. In recent years, with the rapid development of mobile Internet and IOT, embedded AI technology is becoming more and more popular, and the application field of embedded AI is becoming more and more popular. Embedded AI products continue to permeate into daily life. From portable smart phones and smart tablets to smart air conditioners, smart sweeping robots, smart high-definition TVs, smart refrigerators, smart set-top boxes in the home, to smart industrial production,...

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
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06N3/08G06N3/063G06K9/62G06F7/483
CPCG06N3/08G06N3/063G06F7/483G06F18/214
Inventor 缪冉
Owner SENSLAB INC
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap