Byzantine attack defending method in decentralized federated learning based on block chain

Abstract

The invention discloses a Byzantine attack defending method in decentralized federated learning based on a block chain. The method mainly solves the problem that in the prior art, when a Byzantine attack occurs in decentralized federated learning, the calculation overhead and the communication overhead are too high. The method comprises the following implementation steps: 1) a local user obtains a pre-trained model and establishes a reputation contrast relationship; 2) a local user trains the model and then transmits the model, and then generates a signature message and broadcasts the signature message; (3) the local user verifies the signature message and then stores the signature message, and when a certain number of signature messages are stored, blocks are generated, consensus is conducted on the blocks through an improved PBFT consensus algorithm, and then uplink is conducted; and 4) updating the reputation contrast relationship of the transmission and broadcast behaviors of other local users by the local user, then adjusting the generation difficulty of the signature message, and repeating the execution process until the model converges. According to the method, the calculation overhead and the communication overhead in the prior art can be effectively reduced, and the Byzantine robustness of decentralized federated learning can be improved.

Description

technical field [0001] The invention belongs to the technical field of information security, and further relates to attack resistance, in particular to a method for defending a Byzantine attack in a decentralized federated learning based on a blockchain. It is used to resist Byzantine attacks in decentralized federated learning, reduce the amount of calculation and transmission scale of message data, and improve the security of decentralized federated learning. Background technique [0002] Decentralized federated learning does not require a central server, and adjacent users exchange and update models multiple times to ensure that the trained model has good generalization ability. Specifically, users participating in decentralized federated learning hold their own private data and obtain a pre-trained global model as a local model, use the private data to train the local model to obtain a local update model, and send the local update model to adjacent users , and receive t...

AI Technical Summary

Problems solved by technology

In the BytoChain framework proposed in this paper, the three-party cooperative training model of blockchain packaging nodes, data holders, and verifiers is used, and the anti-Byzantine consensus algorithm PoA is used to detect abnormal models, which enhances the ability of federated learning to resist Byzantine attacks. But its disadvantages are: in each round of training, the data holder needs to obtain the latest global model from the blockchain packaging node and send the local model to the verifier for inspection, thus generating a large amount of communication during the model transmission process, which increases the communication overhead

Images