Blockchain-based federated modeling method, device, equipment and storage medium

Abstract

The present invention discloses a blockchain-based federated modeling method, device, device, and storage medium. The method includes: when the number of training clients detected reaches a preset number, the training initiator corresponds to The client information releases configuration information to each training client; the training initiator uploads the model to be trained to the main chain; the training initiator determines the target model based on the aggregation gradient and the model to be trained. The invention realizes the modeling of federated learning through the block chain. Under the premise of protecting the privacy of federated learning data, it has no impact on the accuracy of federated learning, improves the training effect and model accuracy of federated learning, and does not need to modify the gradient in transmission. Modification of model parameters such as gradients achieves a balance between privacy protection of model parameters such as gradients and model convergence or model accuracy; it can completely prevent information leakage and improve the security of data samples in federated learning.

Description

technical field [0001] The present invention relates to the technical field of federated learning, in particular to a blockchain-based federated modeling method, device, equipment and storage medium. Background technique [0002] Federated learning disassembles centralized machine learning into distributed machine learning, distributes machine learning tasks to terminal devices for learning, and then aggregates gradient results generated by machine learning to achieve the purpose of protecting the privacy of terminal devices. [0003] However, due to the unreadability of the gradient results generated by machine learning and the fact that the results may hide private information, there is a problem of privacy leakage in the federated learning mechanism. For example, in many training scenarios, the training network has hundreds or thousands of layers, and it is difficult for client users without a machine learning background to understand the specific role of each layer of th...

AI Technical Summary

Problems solved by technology

[0009] The main purpose of the present invention is to provide a blockchain-based federated modeling method, device, device, and storage medium, aiming to solve the problem of difficulty in achieving gradient privacy protection and model convergence or model accuracy in existing federated learning. balanced technical issues

Images