Blockchain-based federated learning local model ground truth discovery method

By embedding watermarking and trust management into blockchain federated learning, the problem of collusion attacks is solved, decentralized watermark verification and trust value management are achieved, and the fault tolerance performance and efficiency of the global model are improved.

CN117436134BActive Publication Date: 2026-07-17XIDIAN UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
XIDIAN UNIV
Filing Date
2022-07-08
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

In existing blockchain-based federated learning, collusion attacks harm the interests of honest trainers, reduce their participation, and lead to poor global model performance. Furthermore, there is a lack of research on the collusion problem between miner nodes and trainers.

Method used

By embedding a blockchain address as a watermark in the trainer's local model, and utilizing the decentralized technology and trust management of blockchain, miners verify the trainer's trust value, obtain the verification result and update the trust value, and finally aggregate the global model to achieve decentralized verification of the watermark and trust value management.

Benefits of technology

It effectively avoids single points of failure and collusion attacks, improves the system's fault tolerance and efficiency, protects the rights of honest trainers, and ensures the copyright and generalization ability of the global model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117436134B_ABST
    Figure CN117436134B_ABST
Patent Text Reader

Abstract

The application discloses a local model true value discovery method based on a blockchain federated learning, relates to the technical field of blockchains, and comprises the following steps: a task publisher publishes a task and an initial global model through a blockchain; a trainer participating in training downloads the task and the initial global model, and trains a local model with a watermark; the trainer participating in training submits the local model with the watermark to the blockchain; a miner verifies the trust value of the trainer participating in training and the local model, obtains a verification result, and broadcasts or sends the verification result to the task publisher through the blockchain; the task publisher issues a reward to the trainer according to the verification result; the miner obtains a right to keep accounts, the miner obtaining the right to keep accounts aggregates the verified local model into the global model, and updates the trust value of the trainer participating in training; and the miner obtaining the right to keep accounts broadcasts the packaged block, and adds the block in the blockchain after achieving global network consensus. The application can improve the verification efficiency of the whole system.
Need to check novelty before this filing date? Find Prior Art