Data privacy protection method and system based on adaptive adjustment of weights

By introducing a client-selective knowledge distillation strategy into federated learning and dynamically adjusting the distillation loss weights, the problems of model generalization and convergence speed under non-independent and identically distributed data are solved, and more efficient model training is achieved.

CN115495771BActive Publication Date: 2026-06-09INST OF COMPUTING TECH CHINESE ACAD OF SCI
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
INST OF COMPUTING TECH CHINESE ACAD OF SCI
Filing Date
2022-07-06
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing federated learning methods cannot adaptively adjust the update direction of the local model when faced with non-independent and identically distributed data, resulting in poor model generalization and slow convergence speed. Furthermore, they fail to effectively consider the performance changes of the global model and the performance differences between categories and samples.

Method used

A client-selective knowledge distillation strategy is introduced, which dynamically adjusts the distillation loss weights through the category confidence matrix and sample confidence matrix of the global model, adaptively guiding the local model training process, selectively retaining global knowledge and learning from local data.

Benefits of technology

It improves the model's generalization and convergence speed, reduces the number of federated communication rounds, and ensures that the local model does not deviate from the global model while learning new knowledge.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115495771B_ABST
    Figure CN115495771B_ABST
Patent Text Reader

Abstract

The application provides a data privacy protection method and system based on adaptive weight adjustment, solves the problems of model performance decline and slow convergence speed caused by non-independent and identically distributed data, and belongs to the technical field of federated learning application. The method comprises the following steps: at the beginning of each round of federated communication, the server side evaluates the credibility of the global model at the category level by using an auxiliary data set, and downloads the credibility matrix and the global model parameters to the clients participating in the round of federation; the client evaluates the credibility of the global model at the sample level according to the local private data set, and uses the category credibility and the sample credibility for weighting when performing knowledge distillation, dynamically guides the training process of the local model, and uploads the updated local model parameters to the server side; and the server side aggregates the local model parameters to update the global model.
Need to check novelty before this filing date? Find Prior Art