一种基于元学习的不平衡数据联邦学习方法和系统
By combining class-weighted and sample-weighted models with meta-learning in federated learning, the problems of data heterogeneity and imbalanced distribution are solved, and the robustness and accuracy of the global model are improved, especially in the ability to capture data from minority classes.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG LAB
- Filing Date
- 2023-04-25
- Publication Date
- 2026-07-17
AI Technical Summary
In federated learning, the imbalance in data distribution among clients makes it difficult for the model to effectively capture minority class information with a small number of samples. Existing methods are not applicable or have privacy issues when considering data imbalance, and cannot improve the performance of the global model in scenarios with heterogeneous and imbalanced data.
A meta-learning-based approach is adopted, which assigns different weights to local data by using class weighting models and sample weighting models on the client and server sides respectively, and updates the weighting model parameters on the server side using metadata for meta-learning, thereby ensuring the robustness of the global model under unbalanced data distribution.
It improves the performance of the global model in scenarios with heterogeneous and imbalanced data, enhances the model's ability to capture minority classes, and improves the model's robustness and accuracy.
Smart Images
Figure CN116628543B_ABST