A privacy protection heterogeneous federated learning method and system based on image translation
By optimizing the generative model and data augmentation techniques through an image translation server, the problems of global model accuracy and privacy protection in heterogeneous federated learning are solved, achieving efficient data processing and security assurance under non-independent and identically distributed conditions.
Patent Information
- Application Number
- CN202510719780.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-30
- Publication Date
- 2026-05-19
- Estimated Expiration
- 2045-05-30
AI Technical Summary
In heterogeneous federated learning, existing methods lack data privacy protection capabilities, which limits the accuracy of the global model and degrades model performance when local user data is not independently and identically distributed.
The generative model is optimized by using an image translation server. The generative model is then used to perform data augmentation on the local data to make the data present a uniform distribution of category labels. The transformed data is then used to train the classification model. The separation design of the image translation server and the data classification server ensures privacy protection.
It improves the accuracy of the global model under non-independent and identically distributed conditions, while providing strong privacy protection capabilities, resisting gradient inversion attacks, and ensuring data security.
Smart Images
Figure CN120597324B_ABST