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

Neural network model-oriented filter distribution perception training acceleration method and platform

A neural network model and filter technology, applied in biological neural network models, neural learning methods, neural architectures, etc., can solve the problem of not reflecting filter correlation characteristics, and achieve the effect of speeding up the training and inference process.

Inactive Publication Date: 2021-03-26
ZHEJIANG LAB
View PDF2 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Since the filter norm only simulates the amplitude information of the filter, it does not reflect the correlation characteristics between the filters

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-oriented filter distribution perception training acceleration method and platform
  • Neural network model-oriented filter distribution perception training acceleration method and platform
  • Neural network model-oriented filter distribution perception training acceleration method and platform

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0043] The present invention will be further described below in conjunction with accompanying drawing.

[0044] Such as figure 1 As shown, the neural network model-oriented filter distribution perception training acceleration method of the present invention introduces the pruning criterion based on the filter relationship, and with the continuous update of the network iterative training, calculates the distribution of the neural network model according to the distribution of the current channel filter. Pruning criteria, constructing a meta-filter pruning framework that can adaptively select the most appropriate pruning criteria as the filter distribution changes.

[0045] Such as figure 2 As shown, the neural network model-oriented filter distribution perception training acceleration method of the present invention is divided into three steps: the first step is to define the meta-process state; the second step is to select the clipping criterion; the third step is to clip th...

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-oriented filter distribution perception training acceleration method and platform, and the method comprises the steps: introducing a pruning criterion based on a filter relation, constructing a meta-filter pruning framework which can adaptively select the most suitable pruning criterion along with the change of filter distribution, and considering thecorrelation between filters; in the network training process, the clipping criterion of the filter is adaptively switched along with the change of filter distribution; the cut model can accelerate the training and inference process of more visual tasks.

Description

technical field [0001] The invention belongs to the application field of computer technology, and in particular relates to a neural network model-oriented filter distribution perception training acceleration method and platform. Background technique [0002] Existing channel pruning methods usually directly cut out unimportant channels according to preset pruning criteria. This method has two limitations. First, the correlation between filters of different channels is ignored. Filters of different channels usually work together to make accurate predictions in a cooperative manner. Second, the clipping criterion remains unchanged during training. As the network is updated during each iteration, the distribution of filters for different channels is constantly changing. The update of the clipping criterion should also be consistent with the distribution of the current state of the filter. Finally, most of the existing filter clipping criteria are the following strategy: if...

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
IPC IPC(8): G06N3/04G06N3/063G06N3/08
CPCG06N3/063G06N3/082G06N3/045
Inventor 王宏升
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

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com