基于AC-GAN和动态概率调度的可靠联邦学习方法及系统
By using AC-GAN for data cleaning and augmentation in wireless federated learning, combined with CKKS encryption and dynamic probability scheduling, the problems of noisy labels and malicious users are solved, improving data quality and network security, and enhancing the accuracy and communication efficiency of the model.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NORTHWEST A & F UNIV
- Filing Date
- 2024-08-27
- Publication Date
- 2026-07-17
AI Technical Summary
In wireless federated learning, noise labels, non-independent and identically distributed data, and malicious users can lead to reduced model accuracy, privacy leaks, and security threats, affecting model convergence speed and the reliability of the global model.
The AC-GAN model is used for data cleaning and enhancement. Combined with CKKS homomorphic encryption and dynamic probabilistic scheduling strategies, key pairs are generated through a key generation center to identify and filter malicious users, ensuring data security and the reliability of model aggregation.
It improves data quality, mitigates the impact of non-independent and identically distributed data, prevents leakage of user model information and malicious attacks, enhances the accuracy of model training and network security, and optimizes communication efficiency and the convergence speed of the global model.
Smart Images
Figure CN119167226B_ABST