Federal learning-oriented cross-chain consensus method and system

Abstract

The invention relates to a federated learning-oriented cross-chain consensus method and system. The method at least comprises the following steps of: performing single-chain federated learning in a cluster and counting local update information; sending the update consensus information to the second federation to perform cross-cluster gradient exchange; receiving a confirmation result of a cross-cluster gradient update consensus fed back by the second federation; and updating the local model based on the confirmation result. After the updating consensus is achieved, reward and punishment are carried out according to the contribution of the cluster representatives, so that the cluster representatives in the computing nodes are stimulated to vote honest, and participants can actively assist in model updating.

Description

technical field [0001] The invention relates to the technical field of federated learning, in particular to a federated learning-oriented cross-chain consensus method and system. Background technique [0002] In recent years, with the emphasis on privacy and security, federated learning has been widely used in fields that have extremely high requirements for data privacy, such as finance, medical care, insurance, credit information, services, autonomous driving, and indoor positioning. For example, patent CN 112418520A discloses a credit card transaction risk prediction method based on federated learning for transaction risk prediction in the financial field. Patent CN112201342A discloses a federated learning-based medical aided diagnosis method, device, device and storage medium for medical aided diagnosis in the medical field; patent CN 112446791A discloses a federated learning-based auto insurance scoring method, device, device and storage medium for scoring in the insur...

AI Technical Summary

Problems solved by technology

[0004] For this collaborative learning architecture based on the central server, there are currently four main challenges: (1) When the coordination server crashes, the federated learning of the participants will be terminated

(2) A malicious coordination server can reversely infer the distribution of the original data through the update information provided by each node, and there is a risk of privacy leakage

(3) Some malicious participants can poison the global model by submitting inferior update parameters

(4) Due to the lack of incentive mechanism, each participant has no motivation to actively assist in model update

Since all participants need to broadcast and communicate with each other, the communication frequency required to reach a consensus will increase with the number of participants, resulting in high communication overhead and low consensus efficiency

Images