Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Neural network structured progressive pruning method and system

A neural network and network structure technology, applied in the field of computer vision, can solve the problems of cumbersome steps and long processing time, and achieve the effects of simple operation, reduced processing time and high performance

Pending Publication Date: 2021-10-19
ZHEJIANG LAB
View PDF2 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

This pruning method is cumbersome and requires a long processing time

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 structured progressive pruning method and system
  • Neural network structured progressive pruning method and system
  • Neural network structured progressive pruning method and system

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0033] Such as figure 1 As shown, a neural network structured progressive pruning method in this embodiment includes the following steps:

[0034] Step S1: Set the pruning rate, pruning standard and number of training cycles of the neural network for each layer of the neural network;

[0035] Step S2: Input pictures to train the neural network. Within a certain training period, the pruning rate of each layer of the neural network gradually increases from zero to the pruning rate set in step S1, and select the pruning rate of each layer of the neural network according to the pruning standard determined in step S1. The redundant information of , temporarily set the value of the redundant information to 0;

[0036] Step S3: After reaching the set clipping rate, remove the redundant information with a value of 0 processed in step S2 from each layer of the neural network, and reconstruct the original neural network layer;

[0037] Step S4: After the neural network is reconstructe...

Embodiment 2

[0056] Embodiments of the present invention also provide a neural network structured progressive pruning system, such as Figure 4 As shown, the system includes a parameter setting module 21 , a progressive pruning module 22 , a network reconstruction module 23 and a continuous training module 24 .

[0057] Wherein the parameter setting module 21 is used to set the clipping rate of each layer of the neural network, the pruning standard and the number of neural network training cycles; the progressive pruning module 22 is used to input pictures to train the neural network. The pruning rate of each layer of the network is gradually increased from zero to the pruning rate set by the parameter setting module 21, and according to the pruning standard determined by the parameter setting module 21, the redundant information of each layer of the neural network is selected, and the redundant information value is temporarily set to is 0; the network reconstruction module 23 is used to r...

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 relates to the field of computer vision, and relates to a neural network structured progressive pruning method and system, and the method comprises the steps: S1, setting the cutting rate, pruning standard and neural network training cycle number of each layer of a neural network; S2, inputting a picture to train a neural network, gradually increasing the cutting rate of each layer from zero to a set cutting rate in a certain training period, determining redundant information of each layer according to a pruning standard, and setting the redundant information as 0; S3, after a set cutting rate is reached, removing redundant information in the neural network, and reconstructing an original network layer; and S4, after the neural network is reconstructed, continuing training until a set neural network training period is reached. The method is simple in operation and few in steps, the purpose of pruning can be achieved in a normal neural network training process, and a fine adjustment process after pruning is not needed, so that the processing time can be greatly shortened, and compared with the prior art, higher performance can be obtained while a higher cutting rate is achieved.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a neural network structured progressive pruning method and system. Background technique [0002] At present, neural networks have achieved good performance in the field of computer vision, especially in image classification and object detection. It even exceeds the recognition ability of human beings. However, although neural networks achieve high performance, these neural networks usually have high floating-point operations and storage capacity. For example, when an image with an input size of 224×224 is input, ResNet-50 requires 4.1B floating-point operations and The amount of parameters is 25.6MB. The huge amount of calculation and storage requires the operating platform to have higher computing resources and more storage resources. Therefore, these neural networks with good performance cannot be deployed to resource-constrained platforms such as mobile phones and embedded pl...

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): G06N3/08G06N3/04
CPCG06N3/082G06N3/045
Inventor 唐乾坤徐晓刚何鹏飞朱岳江曹卫强王军
Owner ZHEJIANG LAB
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products