Method for improving quality of deep learning data set and interpretability of model based on mesoscience guidance

A deep learning and data set technology, applied in the field of artificial intelligence, can solve problems such as poor model interpretability, and achieve the effect of improving representativeness and effectiveness, reducing the amount of data, and good interpretability

Active Publication Date: 2019-12-03
INST OF PROCESS ENG CHINESE ACAD OF SCI
View PDF1 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to propose a method for improving the quality of deep learning data sets and model interpretability under the guidance of mesoscience in view of the shortcomings of the current deep learning technology that requires a large number of samples in the data set and poor interpretability of the obtained model.

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 improving quality of deep learning data set and interpretability of model based on mesoscience guidance

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0019] Example 1: Recognition of the watershed and its average void ratio of a 3D gas-solid flow system based on a 2D simulation image

[0020] First, the unit of the gas-solid flow system is determined to be a single solid particle, and the problem to be dealt with only includes one system, and the mesoscale is the particle aggregate scale. There are two types of extreme mesoscale structures in this system: one is that the size of particle agglomerates is as large as the system size, that is, a fixed bed, and the corresponding dominant mechanism is that the average void ratio tends to the minimum value; the other is that the size of particle agglomerates is as small as The cell size, ie dilute phase transport, corresponds to the dominant mechanism where the rate of energy expenditure for suspending and transporting particles within a unit volume of the bed tends towards a minimum. The intermediate area between fixed bed and dilute phase transport is fluidized bed, whose stabi...

Embodiment 2

[0021] Example 2: Recognition of the watershed and its average void ratio of the 3D gas-solid flow system based on the 2D experimental image

[0022]First, the unit of the gas-solid flow system is determined to be a single solid particle, and the problem to be dealt with only includes one system, and the mesoscale is the particle aggregate scale. There are two types of extreme mesoscale structures in this system: one is that the size of particle agglomerates is as large as the system size, that is, a fixed bed, and the corresponding dominant mechanism is that the average void ratio tends to the minimum value; the other is that the size of particle agglomerates is as small as The cell size, ie dilute phase transport, corresponds to the dominant mechanism where the rate of energy expenditure for suspending and transporting particles within a unit volume of the bed tends towards a minimum. The intermediate area between fixed bed and dilute phase transport is fluidized bed, whose ...

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 belongs to the field of artificial intelligence, and relates to a method for improving the quality of a deep learning data set and the interpretability of a model by applying a scientific methodology. The implementation process of the invention is as follows: physical systems are defined, a mesoscale model of each physical system is established, a complete physical model of a processed object is established, association between object features and physical attributes is established, a one-dimensional label is established for the object features, a training set and a test set areestablished, deep learning is used for modeling, and the training model is used for predicting and explaining the training model. Compared with an existing deep learning method, due to the fact that the method is associated with the physical model of the processed object, the method has the advantages of being small in required data set, good in interpretability of the built model and the like.

Description

technical field [0001] The invention belongs to the field of artificial intelligence and relates to a method for improving the quality of deep learning data sets and model interpretability by applying mesoscience methodology. Background technique [0002] Deep learning technology is developed from artificial neural network technology. Based on artificial neural network technology, this technology improves the network configuration and calculation algorithm by greatly increasing the number of nodes and hidden layers, and greatly improves the ability to model complex objects. However, due to the increase in the number of nodes and hidden layers and the diversity of network configurations, the number of variables has increased sharply and the complexity of the mathematical model has increased. While improving the accuracy of the model, deep learning also greatly increases the number of samples in the data set. , resulting in a significant increase in modeling time overhead and...

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): G06N3/04G06N3/08G06K9/62G06F17/50
CPCG06N3/08G06N3/045G06F18/214
Inventor 郭力黄文来李静海
Owner INST OF PROCESS ENG CHINESE ACAD OF SCI
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