Federal learning task credible supervision and scheduling method and device based on state channel

Abstract

The invention provides a federal learning task credible supervision and scheduling method and device based on a state channel; the method comprises the steps: carrying out the uplink of key data through an on-chain and off-chain data storage structure for data sharing, so as to verify the integrity of the data; establishing a fine-grained data access control model combined with a node reputation value on the basis of a federated learning data sharing scene of an intelligent contract so as to ensure the identity of the node; enabling a credible supervision mechanism of a model training process based on a state channel to establish a state channel between participants performing the model training process, and completing the transactions generated in the model training process are transferred to a chain. According to the method, a permission chain is introduced to design a hybrid federated learning architecture, identity information and resource information of nodes are stored in a block chain in a user agent registration mode; besides, by designing an on-chain and off-chain data storage structure, data authenticity is ensured, data credibility check is realized, and data storage credibility in a federated learning process is ensured.

Description

technical field [0001] The invention relates to the technical field of trusted supervision, in particular to a method and device for trusted supervision and scheduling of federated learning tasks based on state channels. Background technique [0002] With the development and application of big data and Internet of Things technologies, network data has exploded, and edge devices have broken through the shackles of communication delay and bandwidth, and widely participate in distributed applications. Distributed intelligent systems based on edge computing technology are increasing rapidly, and enterprises obtain a large amount of valuable data. In order to fully tap the value of data, avoid data islands, collaboratively share scattered data, and enhance data value transfer, it has become an inevitable trend of network development. But at the same time, users are increasingly concerned about privacy issues. Privacy protection related clauses and laws have been promulgated one ...

AI Technical Summary

Problems solved by technology

[0004] However, due to the difficulty in solving the trustworthiness problem among cross-domain multi-stakeholders, the application of federated learning technology is mostly limited to a single business system, and the application of cross-domain data sharing is less, which makes it difficult to truly realize the value of data

Images