One-time data-free transmission federated learning method and system in data privacy scene
By constructing an ensemble model in a data privacy scenario and optimizing the model weights using cosine similarity and generative adversarial networks, the problems of high communication overhead and data heterogeneity in federated learning are solved, thereby improving the model's performance and generalization ability on global data.
Patent Information
- Application Number
- CN202610144898.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-02
- Publication Date
- 2026-06-05
AI Technical Summary
Existing federated learning methods in data privacy scenarios suffer from problems such as high communication overhead, high computational complexity, and insufficient model generalization ability due to data heterogeneity. In particular, it is difficult to effectively integrate the unique information and important features of each local model in heterogeneous scenarios.
By constructing an ensemble model with a global perspective, using cosine similarity to evaluate local data differences, employing generative adversarial networks to generate high-quality data, and combining knowledge distillation techniques to optimize model weights, we can achieve one-time data-free federated learning in a data privacy scenario under single-round communication.
It improves the model's performance on global data, enhances communication efficiency, reduces data collection costs, strengthens the model's generalization ability, generates higher-quality data, and is suitable for real-world heterogeneous scenarios.