Federal multi-source domain adaptation method and system based on shadow model

Abstract

The invention discloses a federated multi-source domain adaptation method and system based on a shadow model. The method comprises the following steps: designing a federated multi-source domain adaptation network model based on a shadow model, carrying out decentralized pre-training on different source domain models, and obtaining K decentralized trained source domain models; calculating inter-class variances of output results of different source domain models by obtaining possibility output corresponding to target domain samples, and obtaining contribution proportions of different source domain models to a target domain through normalization processing; obtaining a pseudo label of each target domain sample through weighted calculation, and endowing the target samples with the pseudo labels. carrying out label noise learning on the pseudo label target samples by combining collaborative teaching and confusion matching, carrying out iteration in sequence until network TA and TB converge, and realizing unsupervised domain adaptation under the condition that source domain data, network parameters, and learning gradients are all unknown. The method provided by the invention effectively promotes network training on the premise of ensuring the privacy security of users, and has great social significance.

Description

technical field [0001] The invention belongs to the field of computer vision technology of deep learning, and in particular relates to a federated multi-source domain adaptation method and system based on a shadow model. Background technique [0002] In recent years, research in computer vision has been widely used in real life. However, the implementation process is severely constrained by a large amount of labeled sample data, and labeled data requires a lot of manpower, material and financial resources. To address the problem of insufficient data labels in applications, great progress has been made in the related research of unsupervised domain adaptation. Most of the existing methods are based on the premise of simultaneously obtaining labeled source domain data and unlabeled target domain data. Traditional unsupervised multi-source domain adaptation (UMDA) methods assume direct access to all source domain datasets. But this assumption ignores the user's privacy prote...

AI Technical Summary

Problems solved by technology

[0004] The technical problem to be solved by the present invention is to provide a federated multi-source domain adaptation method and system based on a shadow model to solve the source domain attention federation based on black-box attacks that have not been involved in the prior art. Domain adaptation method, to make up for the limitations of existing federated domain adaptation ignoring different domain differences between different source domains and target domains

Images