Method, device, equipment and storage medium for improving model generalization

By calculating and optimizing the equality score and dimensional equalization constraint loss value of the subsample set in the face recognition model, the problem of low generalization of the model in the case of big data is solved, and better feature distribution uniformity and generalization ability are achieved.

CN114519398BActive Publication Date: 2025-03-28PING AN TECH (SHENZHEN) CO LTD
2 Cites 0 Cited by

Patent Information

Application Number
CN202210161345.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-02-22
Publication Date
2025-03-28
Estimated Expiration
2042-02-22

AI Technical Summary

Technical Problem

In the case of big data, the generalization of the face recognition model is low, resulting in a decrease in accuracy in complex scenarios.

Method used

By obtaining the image set, normalizing and dividing it into multiple subsample sets, the preset image recognition model extracts the feature vector set, calculates the equality score of the feature vector set of each subsample set, calculates the dimensional equalization constraint loss value, and uses this loss value to optimize the model parameters until the trained image recognition model is obtained.

Benefits of technology

Through the optimization of the dimensional equalization constraint loss value, the feature vectors output by the model are more uniformly distributed in high-dimensional space, which improves the distinction between all samples in high-dimensional space, enhances the generalization ability of the model, and can better classify on unseen data.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The present invention relates to artificial intelligence technology, and discloses a method for improving the generalization of a model, including: obtaining a picture set, normalizing the picture set to obtain a sample set, and dividing the sample set into N sub-sample sets; extracting a feature vector set of each sub-sample set using a preset picture recognition model, and calculating a balance score of the feature vector set of each sub-sample set; calculating a dimensional balance constraint loss value using the balance score; optimizing the loss value of the picture recognition model using the dimensional balance constraint loss value to obtain an updated loss value, and iteratively optimizing the parameters of the picture recognition model using the updated loss value until a trained picture recognition model is obtained. In addition, the present invention also relates to blockchain technology, and the picture set can be stored in a node of the blockchain. The present invention also proposes a device, an electronic device, and a storage medium for improving the generalization of a model. The present invention can improve the generalization of a model.
Need to check novelty before this filing date? Find Prior Art