一种基于联邦迁移学习的训练方法
By employing homomorphic encryption and noise protection to safeguard data privacy in federated transfer learning, and optimizing computational terms, the problem of excessive computation and communication in existing technologies is solved, thus achieving an efficient training process.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- FUZHOU QIYUAN INFORMATION TECHNOLOGY CO LTD
- Filing Date
- 2022-11-17
- Publication Date
- 2026-07-17
AI Technical Summary
In the existing federated transfer learning framework, the use of homomorphic encryption to protect data privacy by both parties results in excessive computation and communication, slow speed, and the need to convert to scalar multiplication operations, which also leads to huge amounts of communication and computation.
Data privacy is protected by using homomorphic encryption on one side and adding noise on the other side, so that the computation of one side is performed in plaintext, and the computation terms during the training process are optimized to reduce the computation and transmission of intermediate result matrices in the homomorphic encryption state.
This significantly reduces the computational and communication load of federated transfer learning, improves training efficiency, and the benefits far outweigh the impact of establishing two more rounds of communication connections.
Smart Images
Figure CN116049843B_ABST