English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Conventional unsupervised multi-source domain adaptation (UMDA) methodsassume all source domains can be accessed directly. This neglects theprivacy-preserving policy, that is, all the data and computations must be keptdecentralized. There exists three problems in this scenario: (1) Minimizing thedomain distance requires the pairwise calculation of the data from source andtarget domains, which is not accessible. (2) The communication cost and privacysecurity limit the application of UMDA methods (e.g., the domain adversarialtraining). (3) Since users have no authority to check the data quality, theirrelevant or malicious source domains are more likely to appear, which causesnegative transfer. In this study, we propose a privacy-preserving UMDA paradigmnamed Knowledge Distillation based Decentralized Domain Adaptation (KD3A),which performs domain adaptation through the knowledge distillation on modelsfrom different source domains. KD3A solves the above problems with threecomponents: (1) A multi-source knowledge distillation method named KnowledgeVote to learn high-quality domain consensus knowledge. (2) A dynamic weightingstrategy named Consensus Focus to identify both the malicious and irrelevantdomains. (3) A decentralized optimization strategy for domain distance namedBatchNorm MMD. The extensive experiments on DomainNet demonstrate that KD3A isrobust to the negative transfer and brings a 100x reduction of communicationcost compared with other decentralized UMDA methods. Moreover, our KD3Asignificantly outperforms state-of-the-art UMDA approaches.

KD3A: Unsupervised Multi-Source Decentralized Domain Adaptation via Knowledge Distillation