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

Convolutional neural network algorithm based on echo state network classification

A technology of echo state network and convolutional neural network, which is applied in the field of signal processing and can solve problems such as high training time cost, overfitting, and occupancy

Active Publication Date: 2019-08-13
SOUTHEAST UNIV
View PDF2 Cites 6 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, the current CNN algorithm generally has the problem of high training time cost, which makes it take up a lot of computing resources; at the same time, CNN does not perform well on small sample data sets, and is prone to overfitting

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
  • Convolutional neural network algorithm based on echo state network classification
  • Convolutional neural network algorithm based on echo state network classification
  • Convolutional neural network algorithm based on echo state network classification

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0059] This simulation experiment is carried out on a server with a main frequency of 2.5GHz, 12 cores, a CPU model of Intel Xeon E5-2678v3, and a memory of 64GB, using MATLAB R2016b as the algorithm editor.

[0060] Here, the number of convolution kernels of the two convolutional layers C2 and C4 of CNN and E-CNN is set to 6 and 16, and the size is 5×5; the modes of the two downsampling layers P3 and P5 are both MEAN, and the sampling area Both are 2×2; the activation function is sigmoid function, and the learning rate is set to 1. E-CNN's reserve pool size N R =1000, select tanh as the activation function of the reserve pool state, select the linear output function as the output activation function, and select the regularization parameter λ=1×e -7 , the idling of the reserve pool is no longer set here.

[0061] The CNN model parameters pre-trained on the CIFAR-10 dataset are used as the initial values ​​of the experiment. Zero pad around the 28*28 size image to make it a ...

Embodiment 2

[0069] This simulation experiment is carried out on a server with a main frequency of 2.5GHz, 12 cores, a CPU model of Intel Xeon E5-2678v3, and a memory of 64GB, using MATLAB R2016b as the algorithm editor.

[0070] Here, the number of convolution kernels of the two convolutional layers C2 and C4 of CNN and E-CNN is set to 6 and 16, and the size is 5×5; the modes of the two downsampling layers P3 and P5 are both MEAN, and the sampling area Both are 2×2; the activation function is sigmoid function, and the learning rate is set to 1. E-CNN's reserve pool size N R =1000, select tanh as the activation function of the reserve pool state, select the linear output function as the output activation function, and select the regularization parameter λ=1×e -7 , the idling of the reserve pool is no longer set here.

[0071] The CNN model parameters pre-trained on the CIFAR-10 dataset are used as the initial values ​​of the experiment. Zero pad around the 28*28 size image to make it a ...

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 convolutional neural network algorithm based on echo state network classification. The method mainly comprises the steps that an ESN is used for replacing a full connection layer in a CNN model, a new residual error iteration formula is derived, on the basis, the model continues to use a CNN back propagation algorithm to train hidden layer parameters of the CNN, and a linear regression rule is used for training the output weight of the ESN. Simulation experiments on a MNIST handwritten digital recognition data set, a Fathion MNIST object recognition data set and an ORLface recognition data set prove the feasibility of the method. meanwhile, the experimental result reflects that the model not only retains the CNN multi-level feature extraction capability, but alsoreduces the training time of the algorithm and improves the performance of the algorithm on a small sample data set by introducing an ESN module.

Description

technical field [0001] The invention relates to the technical field of signal processing, in particular to a convolutional neural network algorithm based on echo state network classification. Background technique [0002] Convolution Neural Network (CNN) is a deep neural network that can automatically extract multi-level features of images. Because of its characteristics of weight sharing, sparse connections and pooling operations, it is widely used in target detection, fields such as face recognition and natural language processing. However, the current CNN algorithm generally has the problem of high training time cost, which makes it take up a lot of computing resources; at the same time, CNN does not perform well on small sample data sets, and is prone to overfitting. How to solve these two problems is the current research focus of CNN optimization algorithm. Contents of the invention [0003] In order to solve the above problems, the present invention provides a conv...

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): G06N3/04G06N3/08
CPCG06N3/08G06N3/084G06N3/045
Inventor 夏亦犁王新裴文江
Owner SOUTHEAST UNIV
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