Influence factor analysis method and device for neural nodes of convolutional neural network

A convolutional neural network and neural node technology, applied in the field of influence factor analysis of neural nodes, can solve problems such as limited technical strength of users

Inactive Publication Date: 2020-02-18
CHENGDU SEFON SOFTWARE CO LTD
View PDF0 Cites 5 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a method and device for analyzing the influencing factors of the neural nodes of the convolutional neural network, which solves the problem that the existing open framework can only carry out upper-level packaging research and development, and the user is limited in technical strength and lacks real needs. A standard method for visualizing popular graphs in depth with different applications

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
  • Influence factor analysis method and device for neural nodes of convolutional neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0087] A method for analyzing influence factors of neural nodes of a convolutional neural network, comprising the following steps:

[0088] Analyze the neural nodes of the neural network, mark the distribution of each neural node, obtain the connection between the neural node and the layer, and the connection between the neural node and the neural node;

[0089] Determine the central node, according to the connection between the neural node and the layer, the connection between the neural node and the neural node, use the PageRank expansion algorithm to follow the IV index, Gini index index, entropy index, information gain index, Pearson correlation index Gaussian mixture model performing iterative training on the influence factor of each neural node to the central node by at least two indicators in the index, the intra-layer distance index and the inter-layer distance index;

[0090] When the change of all indicators within a unit time is less than the threshold, stop the ite...

Embodiment 2

[0104] On the basis of Embodiment 1, this embodiment further includes performing data preprocessing on the data of the neural nodes before analyzing the neural nodes of the neural network.

[0105] Further, the data preprocessing includes data preparation and data feature engineering. In order for machine learning algorithms to achieve optimal accuracy on datasets, data preprocessing is essential. It is characterized in that data is sorted and integrated to facilitate subsequent unified association. The data preprocessing module is divided into two steps, one is data preparation, and the other is data feature engineering. The following process is used:

[0106] In data preparation, the operation of receiving data transmitted from multiple sites, data cleaning enables the data to obtain the correct shape (shape) and quality (quality) for analysis; this operation in turn includes many different functions, such as:

[0107] 1. Basic functions, including: selection, filtering, de...

Embodiment 3

[0127] In this embodiment, further on the basis of Embodiment 1, the neural node impact analysis model corresponding to the central node also includes:

[0128] The neural node impact analysis model obtained by using the PageRank extension algorithm is combined with other algorithms for further unsupervised integrated learning;

[0129] The weights of the PageRank extension algorithm and other algorithms are obtained through integrated learning, and the optimal association model that is superior to the neural node impact analysis model is obtained.

[0130] Further, the other algorithms include at least one of PCA principal component analysis algorithm and self-encoding algorithm.

[0131] Subsequent selections include PCA principal component analysis, self-encoding and other learners for unsupervised integrated learning. By training the weight of the integrated learner and combining the advantages of the three learners, a weight analysis better than PageRank is obtained to ac...

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 an influence factor analysis method and device for neural nodes of a convolutional neural network. The invention provides a convolutional neural network visualization algorithmbased on a PageRank. A general module required in most convolutional neural network training processes is provided; wherein the functions include parameter adjustment, important node selection, calculation amount reduction, normalization processing, important influence factor analysis, input and output result flow direction and the like. And the method can be integrated with any standardized convolutional neural network and various indexes for evaluating the important association degree of the nodes, so that high-efficiency and high-precision association analysis is realized, and the existingdata is utilized to the maximum extent. The problems of blindness and uninterpretability of a result during parameter adjustment of a traditional convolutional neural network and low interpretability, low efficiency, time waste and the like of model adjustment and optimization depending on global traversal of all parameter combinations and algorithm parameter adjustment experience are solved, andthe interpretability of the convolutional neural network and the accuracy of the model are further improved.

Description

technical field [0001] The present invention relates to the field of big data, in particular to a method and device for analyzing influence factors of neural nodes of a convolutional neural network. Background technique [0002] First of all, in the training process of the existing convolutional neural network, although the convolutional neural network has a powerful feature extraction function for input images or other types of data, it can be used for beginners and project executors who have no experience in project adjustment. For workers, the inexplicability of the output results is difficult to get a good home, and it was once considered a violent algorithm, so there is a great demand for the interpretability of the output results. [0003] The parameter adjustment of the existing convolutional neural network is often based on the violent combination of model parameters, and the model combination parameters are randomly selected as the final result, or blindly adjusted ...

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
CPCG06N3/08G06N3/045
Inventor 查文宇张艳清王伟才潘小东殷腾蛟王纯斌赵神州
Owner CHENGDU SEFON SOFTWARE CO LTD
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap