Unlock instant, AI-driven research and patent intelligence for your innovation.

Online customization method and system for client deep learning

A deep learning and client-side technology, applied in transmission systems, instruments, software design, etc., can solve the problems of deep learning models such as large size, large network bandwidth, and occupation

Active Publication Date: 2019-04-19
PEKING UNIV
View PDF2 Cites 12 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The size of the deep learning model is usually large, thus occupying a large amount of network bandwidth

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
  • Online customization method and system for client deep learning
  • Online customization method and system for client deep learning
  • Online customization method and system for client deep learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0056] refer to figure 1 , which shows a flow chart of the steps of an online customization method for client-side deep learning in an embodiment of the present invention, the method includes a server and at least one terminal, and the specific steps include:

[0057] Step 101, using the public data set to train the preset machine learning model to obtain the public model.

[0058] In the embodiment of the present invention, the server is responsible for pre-training to obtain a public model with parameters correctly initialized. First of all, it is necessary to artificially select a public dataset that is semantically closer to the data generated by users. The acquisition of public datasets can be obtained through direct download or web crawler. For example: obtain the ImageNet data set for image classification through direct download; use web crawlers to obtain corpus data on the Twitter website. The acquired public data needs to undergo a complete cleaning and preprocessin...

Embodiment 2

[0069] refer to figure 2 , which shows a flow chart of the steps of an online customization method for client-side deep learning in an embodiment of the present invention, the method includes a server and at least one terminal, and the specific steps include:

[0070] Step 201, the server acquires a public data set, preprocesses the public data set, uses the public data set to train a preset machine learning model, and obtains a public model.

[0071] In the embodiment of the present invention, the selection and preprocessing of the public data set need to be as consistent as possible with the semantics of the original prediction task. For example: in the input word prediction function in input method applications, public data sets can be obtained through web crawlers, such as Twitter corpus and BBC News corpus, but the models trained by the two are quite different: obviously, the former is more It is close to the user's daily input habits. In fact, many users themselves use...

Embodiment 3

[0086] refer to Figure 4 , the present invention discloses an online customization method for client deep learning, the method includes a server and multiple clients, specifically including:

[0087] In the embodiment of the present invention, on the server side, the preset machine model is pre-trained using the public data set to obtain a public model with parameters correctly initialized, and the public model is sent to different clients, and the On the client side, use different personalized data and preset observation output to perform personalized training on the public model to predict the public model, adjust the parameters of the public model, and obtain a customized model.

[0088] The embodiment of the present invention adopts the method of server-client cooperative training, and uses a large number of public data sets for pre-training on the server side with strong computing resources, and adjusts the parameters of the model to a suitable position; then sends the m...

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 an online customization method and system for client deep learning, and the system comprises a server and at least one client. The method comprises the steps that the server utilizes a public data set to train a preset machine learning model to obtain a public model; The client acquires the public model from the server; The client acquires personalized data of the user; Andthe client trains the public model by using the user personalized data to obtain a customized model. According to the method, a server-client cooperative training mode is adopted, a large number of public data sets are firstly used for pre-training at a server with high computing resources, and parameters of a model are adjusted to proper positions; And then the model is issued to each mobile device, and customized training is carried out by using locally generated data. In the prediction and customized training process of the client side, training data does not need to be uploaded, and therefore it is guaranteed that privacy information cannot be leaked.

Description

technical field [0001] The present invention relates to the field of software technology, in particular to an online customization method and system for client deep learning. Background technique [0002] Deep learning is a machine learning algorithm and an important branch of artificial intelligence. From rapid development to practical application, in just a few years, deep learning has subverted the algorithm design ideas in many fields such as speech recognition, image classification, and text understanding, and gradually formed an end-to-end ( end-to-end) model, and then directly output a new model to get the final result. [0003] On the mobile platform, deep learning technology has also been widely used. For example, relying on deep learning in smart glasses for high-accuracy face recognition; using deep learning in smartphones for precise text translation, etc. Existing application solutions usually train a general model on a cloud server, embed the model into the ...

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): G06F8/20G06K9/62H04L29/06
CPCG06F8/20H04L67/01G06F18/214
Inventor 黄罡刘譞哲徐梦炜马郓
Owner PEKING UNIV