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

Method for quantizing weight by channels

A technology of quantizing weights and splitting channels, which is applied in the field of neural network acceleration, can solve the problems of reducing model accuracy and insufficient utilization of low-bit data, and achieve the effects of improving utilization, increasing convergence speed, and fully utilizing

Pending Publication Date: 2021-12-07
合肥君正科技有限公司
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0010] This application proposes a method for quantizing weights by channels, which aims to overcome the defects in the above-mentioned prior art, solve the problem of insufficient utilization of low-bit data when quantizing existing low-bit models, and concentrate on a small number of values, reducing the model The problem of precision

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
  • Method for quantizing weight by channels
  • Method for quantizing weight by channels
  • Method for quantizing weight by channels

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0032] In order to understand the technical content and advantages of the present invention more clearly, the present invention will be further described in detail in conjunction with the accompanying drawings.

[0033] Such as figure 1 As shown, a method for sub-channel quantization weights of the present invention specifically includes the following steps:

[0034] S1, convolutional neural network training: train the model with a full-precision algorithm. The full-precision algorithm is an image classification algorithm based on Resnet-50 as a neural network structure to obtain a network for target classification, that is, to obtain relevant parameters in the model reasoning process. , the relevant parameters include the weight of the convolution, the bias of the BiasAdd operator, the gamma, beta, mean and variance of the BatchNormal operator;

[0035] S2, fine-tuning the quantized model:

[0036] S2.1, for the model obtained from S1, quantify the weight according to the 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 provides a method for quantizing weight by channels, which is characterized in that the weight is quantized according to the number of output channels of a model, when the weight of a convolutional neural network is four-dimensional (height, width, input_channel and output_channel), the extreme values of other three-dimensional data are respectively counted according to the output channel, then the data are quantized into low bits, and the data are quantized according to the distribution characteristic of each channel. The invention aims to overcome the defects in the prior art and solve the problem that the precision of the model is reduced due to the fact that low-bit data is not fully utilized and concentrated in a small number of numerical values when the existing low-bit model is quantized.

Description

technical field [0001] The invention relates to the technical field of neural network acceleration, in particular to a method for channel-by-channel quantization of weights. Background technique [0002] In recent years, with the rapid development of science and technology, the era of big data has arrived. Deep learning uses deep neural network (DNN) as a model, and has achieved remarkable results in many key areas of artificial intelligence, such as image recognition, reinforcement learning, and semantic analysis. As a typical DNN structure, convolutional neural network (CNN) can effectively extract hidden layer features of images and accurately classify images. It has been widely used in the field of image recognition and detection in recent years. [0003] In particular, weights are quantized according to the global extremum: first obtain the extremum of the entire weight from the weights and then quantize the weights to low bits according to this value. [0004] Howeve...

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): G06N3/08G06N3/04G06F17/18
CPCG06N3/08G06F17/18G06N3/045Y02D10/00
Inventor 张东
Owner 合肥君正科技有限公司
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