Differentiated noise adding method and system in federated learning gradient exchange

Abstract

The invention discloses a differentiated noise adding method, medium and system in federated learning gradient exchange, and the method comprises the steps that a plurality of data parties obtain corresponding training data sets, and train a deep learning model according to the corresponding training data sets, so as to update the gradient of the deep learning model; each data party performs hierarchical processing on the corresponding gradient, calculates a two-norm corresponding to each layer of gradient, adds noise to each layer of gradient according to the two-norm, and sends the noise-added gradient to the central server; the central server aggregates the noise-added gradients and sends the aggregated gradients to each data party, so that each data party updates a local deep learningmodel according to the aggregated gradients; and according to the method, the privacy protection strength in the data exchange process of federated learning can be improved, and meanwhile, compared with an encryption algorithm in traditional federated learning, the overhead of data transmission can be reduced.

Description

technical field [0001] The present invention relates to the technical field of deep learning, in particular to a differential noise adding method in federated learning gradient exchange, a computer-readable storage medium, and a differentiated noise adding system in federated learning gradient exchange. Background technique [0002] In the process of data sharing and distributed deep learning, there are often problems of data privacy leakage. In order to solve this problem, federated learning methods are often used to reduce privacy leakage during data exchange. [0003] In related technologies, in the process of using the gradient update algorithm, there is often a problem of low privacy protection; in order to enhance the strength of privacy protection, the homomorphic encryption method is usually used, and this method will be used in deep learning scenarios. It will make the already huge gradient data expand again, which in turn will bring huge data transmission overhead...

AI Technical Summary

Problems solved by technology

[0003] In related technologies, in the process of using the gradient update algorithm, there is often a problem of low privacy protection; in order to enhance the strength of privacy protection, the homomorphic encryption method is usually used, and this method will be used in deep learning scenarios. It will make the already huge gradient data expand again, which will bring huge data transmission overhead

Images