A model training system and method

A model training and model technology, applied in the field of model training system, can solve the problems of long model update cycle, data leakage, low model accuracy, etc., achieve high model update efficiency, overcome low accuracy, and improve accuracy

Active Publication Date: 2021-07-20
ZHEJIANG DAHUA TECH CO LTD
View PDF3 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] The embodiment of the present invention provides a model training system and method to solve the problems of low model accuracy, risk of data leakage and long model update cycle in the prior art

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
  • A model training system and method
  • A model training system and method
  • A model training system and method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0051] figure 1 Model training system structure schematic diagram of the embodiment of the system according to the present invention comprises: 13, the front end of the front end of the apparatus 11 apparatus 11, the rear end of device 12 and storage device comprising a first weak initial model 111, the rear end of the apparatus 12 includes a second weak initial model 121, model 122 and initial intensity controller 123, comprising a set of training images 13, the storage device;

[0052] The first weak initial model 111, for receiving a first image for each of the outputs of the first image corresponding to a first recognition result to the controller 123;

[0053] The strong initial model 122, for receiving the second identification for each of the first image, a first image corresponding to the output of the result to the controller 123;

[0054] The controller 123, for each of the first image for determining a first identification of the first result and the second image is th...

Embodiment 2

[0068] On the basis of the above-described embodiments, the embodiment of the invention, the controller 123, particularly when the second number of images for identifying the training image set reaches a first preset threshold number, determining the current arrival model training time.

[0069]In order to ensure the accuracy of the second initial weaker model 121 obtained, in the embodiment of the present invention, the controller 123 determines the current arrival when the number of second images in the training image is reached to the first number of thresholds. Model training time. The second initial weaker model 121 is then trained. The reason why the training image is maintained in the training image, and the original training image is maintained, and the controller 123 is subsequently controlled to the image of the training image set to the second identification result output from the initial strength model 122 and the second initial weak model 121 output. The first identif...

Embodiment 3

[0074] In order to avoid problems with long training time, and reduce the redundancy of the training image set, on the basis of the above embodiments, in the embodiment of the present invention, the initial strength model 122 is also used to output each of the The fourth identification result of the second image is to the controller 123, the second initial weakening model 121 outputs a fifth recognition result of each of the second images to the controller 123;

[0075] The controller 123 is further configured to determine the fourth identification result of each of the second images and the first similarity of the fifth identification result, sort each first similarity, according to the first similarity Select the second number of second numbers of the second number, remove the selected second image, and the training image set is removed.

[0076] In the embodiment of the present invention, when the control device 123 determines that the initial strength model 122 and the second ...

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 model training system and method, which relate to the field of artificial intelligence of machine learning technology. For each first image, a controller judges that the difference between the first recognition result and the second recognition result of the first image satisfies a preset When required, save the first image to the training image set; when it is judged that the current model training time is reached, control the initial strong model and the second initial weak model, and obtain each second image in the training image set; receive the initial strong model Output the third recognition result of each second image in the training image set, use each third recognition result as supervision information, train the second initial weak model, and update the first initial weak model. The embodiments of the present invention improve model accuracy through online learning. It also avoids threats to the security of sample data transmitted in the network and the risk of data leakage. At the same time, the update of the model does not rely on manual intervention, so the update efficiency of the model is higher.

Description

Technical field [0001] The present invention relates to the artificial intelligence field of machine learning techniques, and in particular, there is a model training system and method. Background technique [0002] With the rapid development of artificial intelligence technology in recent years, model training is applied in more and more fields, such as in the field of image processing, pre-training a target classification model, using this model to classify the target objects in the image. [0003] In the prior art, a model training method is to use offline training models, and then deploy the training model to each front-end device in practical applications. The problem exists in this way is that the model is limited, and the accurate results cannot be obtained when there is an image in the training set in the scene. [0004] In order to solve some problems in offline model training, there is a method of offline-online combined model training in the prior art, and the samples ...

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 Patents(China)
IPC IPC(8): G06N20/00
Inventor 巫立峰
Owner ZHEJIANG DAHUA TECH 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