Unlock instant, AI-driven research and patent intelligence for your innovation.

Neural network model acceleration method and platform based on filter distribution

A neural network model and neural network technology are applied in the field of neural network model acceleration methods and platforms based on filter distribution, and can solve problems such as difficulties in neural network training.

Active Publication Date: 2021-03-26
ZHEJIANG LAB
View PDF2 Cites 5 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, due to the introduction of quantization errors, the training of neural networks is very difficult

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 acceleration method and platform based on filter distribution
  • Neural network model acceleration method and platform based on filter distribution
  • Neural network model acceleration method and platform based on filter distribution

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0037] The present invention will be further described below in conjunction with accompanying drawing.

[0038] The present invention considers the amplitude information of different channel filters and the correlation features between each other, and proposes a neural network model acceleration method based on filter distribution, and its overall structure is as follows figure 1 shown. The distance between the filters is introduced to reflect the distribution of the filters, and on this basis, a clipping criterion based on the filter average similarity score is designed, that is, with the continuous update of the network iterative training, according to the current channel filter Clipping criteria for distributed computing neural network models.

[0039] The neural network model acceleration method based on the filter distribution of the present invention, the whole process is as follows figure 2 As shown, it is divided into four steps: the first step is to define the prob...

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 acceleration method and platform based on filter distribution, introduces the distance between filters to reflect the distribution of the filters, and designs a novel clipping criterion based on the average similarity score of the filters on the basis, i.e., along with the continuous updating of network iterative training, a clipping criterion of theneural network model is calculated according to the distribution of the current channel filter. The method comprises the following steps of 1, defining a problem, and modeling a neural network convolution operation; 2, designing a neural network optimization target based on filter cutting; 3, calculating a filter similarity score based on a Minkowski distance; and 4, designing a filter cutting criterion.

Description

technical field [0001] The invention belongs to the application field of computer technology, and in particular relates to a neural network model acceleration method and platform based on filter distribution. Background technique [0002] Large-scale deep convolutional neural network models have achieved excellent performance in the field of computer image applications. However, due to the fact that the computing tasks in practical application scenarios must be completed under the conditions of limited resource supply, such as computing time, storage space, battery power, etc., Deploying a pre-trained model with a large number of parameters to a device with limited memory is a huge challenge. For example, the VGG-16 model has 138.3 million parameters, takes up more than 500MB of storage space, and requires 30.94 billion floating-point operations to process a single image. Classification. In the field of model compression, existing neural network quantization compression met...

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): G06N3/04G06N3/063G06N3/08
CPCG06N3/082G06N3/063G06N3/045Y02T10/40
Inventor 王宏升管淑祎
Owner ZHEJIANG LAB