基于联邦学习的模型训练处理方法、装置、设备及介质
By updating only the high-level feature parameter layer and freezing the low-level feature layer in federated learning, the problem of high computational resource consumption is solved, achieving the effects of privacy protection and reduced computational cost.
CN115358414BActive Publication Date: 2026-07-17PING AN TECH (SHENZHEN) CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PING AN TECH (SHENZHEN) CO LTD
- Filing Date
- 2022-07-28
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
In federated learning, existing technical solutions consume a lot of computation time and resources when defending against feature reasoning attacks using cryptographic methods, resulting in high computational costs.
Method used
In federated learning, only the parameter layers for learning high-level features are updated, while the parameter layers for low-level features are frozen. These low-level features do not participate in training or parameter uploading, thus avoiding feature inference attacks.
Benefits of technology
It effectively protects user privacy, reduces computing resource costs, and simultaneously achieves efficient completion of image classification and object detection tasks.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN115358414B_ABST
Abstract
本申请涉及人工智能技术领域，公开了一种基于联邦学习的模型训练处理方法、装置、设备和存储介质，用于减少计算成本。方法部分包括：将中央节点设备本地的初始网络模型，发送给各个子节点设备；获取各个子节点设备分别上传的目标参数，为各个子节点设备对初始网络模型进行训练后，得到的已训练网络模型的第一预设特征参数层的参数，其中，第一预设特征参数层为已训练网络模型中感受野大于预设值的网络层，已训练网络模型的第二预设特征参数层的参数被子节点设备冻结在本地，第二预设特征参数层为已训练网络模型中感受野小于或等于预设值的网络层；根据各个子节点设备分别上传的目标参数，更新中央节点设备本地的初始网络模型，得到目标网络模型。
Need to check novelty before this filing date? Find Prior Art