Training method and apparatus of network models, equipment, and storage medium

A network model and training method technology, applied in the field of deep learning, can solve the problems of under-fitting, over-fitting of high-level features of parameter transfer paths, and unsatisfactory processing results, and achieve the effect of high accuracy and thorough training.

Inactive Publication Date: 2018-01-26
GUANGZHOU SHIYUAN ELECTRONICS CO LTD
View PDF2 Cites 14 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The existing technology usually adds one or more loss functions after the feature layer of the network for training, but during the training process, due to the long parameter transmission path, the high-level features are over-fitted and the middle-level features are under-fitted, resulting in The training of the entire network is not thorough enough, which leads to the inability to update the parameters of the middle layer of the deep network learning model network, which leads to unsatisfactory processing results of the trained model in actual face recognition

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
  • Training method and apparatus of network models, equipment, and storage medium
  • Training method and apparatus of network models, equipment, and storage medium
  • Training method and apparatus of network models, equipment, and storage medium

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0030] Figure 1a It is a flow chart of a network model training method provided by Embodiment 1 of the present invention. This embodiment can adapt to the situation where the parameter transmission path in the network model is too long and needs to optimize the network model. This method can be provided by the embodiment of the present invention A network model training device is implemented, and the device can be implemented in software and / or hardware. refer to Figure 1a , the method may specifically include the following steps:

[0031] S110. When the first network model reaches the preset update stop condition, determine the first target network model according to the update result of the first network model, and insert after the preset pooling layer in the first target network model The subsequent loss network layer is used to determine the second network model.

[0032] Specifically, during the training process of the first network model, a preset update stop conditio...

Embodiment 2

[0049] Figure 2a It is a flow chart of a method for training a network model provided by Embodiment 2 of the present invention. This embodiment is implemented on the basis of the foregoing embodiments. refer to Figure 2a , the method may specifically include the following steps:

[0050] S210. Input the picture to be trained into the first network model for training, and update the first network model according to the training result.

[0051] Wherein, the picture to be trained is input into the first network model for training, and the first network model is updated according to the training result. Optionally, the pictures to be trained may be pictures in a specific scene, and the specific scene may be a VTM (Video Teller Machine, remote teller machine), member identification in a jewelry store, and the like. Collect face photos in a specific scene, use the camera to collect video pictures, and store them in the computer system through network transmission and data line...

Embodiment 3

[0070] image 3 It is a flow chart of a network model training method provided by Embodiment 3 of the present invention. On the basis of the above embodiments, this embodiment applies the relay loss function and the global loss function to the second The parameters of the network model are updated to obtain the updated second network model" and optimized. refer to image 3 , the method may specifically include the following steps:

[0071] S310. When the first network model reaches the preset update stop condition, determine the first target network model according to the update result of the first network model, and insert it after the preset pooling layer in the first target network model The subsequent loss network layer is used to determine the second network model.

[0072] S320. Determine a relay loss function corresponding to the relay loss network layer according to the second network model and the relay loss network layer.

[0073] S330. Determine a global loss fu...

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 embodiment of the invention discloses a training method and apparatus of network models, equipment, and a storage medium. The training method of network models includes the steps: when a first network model achieves the preset update stopping condition, according to the update result of the first network model, determining the first target network model, and inserting a relay loss network layer in the preset pooling layer in the first target network model so as to determine the second network model; according to the second network model and the relay loss network layer, determining the relay loss function corresponding to the relay loss network layer; according to the second network model and the relay loss function, determining the global loss function of the second network model; andapplying the relay loss function and the global loss function to update the parameters of the second network model so as to obtain the updated second network model. The training method and apparatusof network models can solve the problem that during the network model training process, the high-layer characteristics are over fitting while the low layer characteristics are poor fitting, so that the network model training can become more thoroughly and has higher accuracy.

Description

technical field [0001] The present invention relates to the field of deep learning, in particular to a network model training method, device, equipment and storage medium. Background technique [0002] Existing face recognition models are usually trained based on deep learning algorithm models, and the quality of deep learning algorithm model training affects the results of face recognition. [0003] The current deep learning model for face recognition basically adds one or more loss function layers after the highest layer (feature layer) of the network to train and update the parameters of the deep learning network model. The existing technology usually adds one or more loss functions after the feature layer of the network for training, but during the training process, due to the long parameter transmission path, the high-level features are over-fitted and the middle-level features are under-fitted, resulting in The training of the entire network is not thorough enough, wh...

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): G06K9/00G06N3/04G06N3/08
Inventor 张玉兵
Owner GUANGZHOU SHIYUAN ELECTRONICS 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