Multi-Source Unsupervised Domain Adaptation with Prototype Aggregation

Multi-source domain adaptation (MSDA) plays an important role in industrial model generalization. Recent efforts regarding MSDA focus on enhancing multi-domain distributional alignment while omitting three issues, e.g., the class-level discrepancy quantification, the unavailability of noisy pseudo l...

Full description

Saved in:
Bibliographic Details
Main Authors: Min Huang, Zifeng Xie, Bo Sun, Ning Wang
Format: Article
Language:English
Published: MDPI AG 2025-02-01
Series:Mathematics
Subjects:
Online Access:https://www.mdpi.com/2227-7390/13/4/579
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850231750527025152
author Min Huang
Zifeng Xie
Bo Sun
Ning Wang
author_facet Min Huang
Zifeng Xie
Bo Sun
Ning Wang
author_sort Min Huang
collection DOAJ
description Multi-source domain adaptation (MSDA) plays an important role in industrial model generalization. Recent efforts regarding MSDA focus on enhancing multi-domain distributional alignment while omitting three issues, e.g., the class-level discrepancy quantification, the unavailability of noisy pseudo labels, and source transferability discrimination, potentially resulting in suboptimal adaption performance. Therefore, we address these issues by proposing a prototype aggregation method that models the discrepancy between source and target domains at the class and domain levels. Our method achieves domain adaptation based on a group of prototypes (i.e., representative feature embeddings). A similarity score-based strategy is designed to quantify the transferability of each domain. At the class level, our method quantifies class-specific cross-domain discrepancy according to reliable target pseudo labels. At the domain level, our method establishes distributional alignment between noisy pseudo-labeled target samples and the source domain prototypes. Therefore, adaptation at the class and domain levels establishes a complementary mechanism to obtain accurate predictions. The results on three standard benchmarks demonstrate that our method outperforms most state-of-the-art methods. In addition, we provide further elaboration of the proposed method in light of the interpretable results obtained from the analysis experiments.
format Article
id doaj-art-db1db44a295d4c17bc295fdb24cee5d6
institution OA Journals
issn 2227-7390
language English
publishDate 2025-02-01
publisher MDPI AG
record_format Article
series Mathematics
spelling doaj-art-db1db44a295d4c17bc295fdb24cee5d62025-08-20T02:03:25ZengMDPI AGMathematics2227-73902025-02-0113457910.3390/math13040579Multi-Source Unsupervised Domain Adaptation with Prototype AggregationMin Huang0Zifeng Xie1Bo Sun2Ning Wang3School of Software Engineering, South China University of Technology (SCUT), Guangzhou 510006, ChinaSchool of Software Engineering, South China University of Technology (SCUT), Guangzhou 510006, ChinaInstitute of International Services Outsourcing, Guangdong University of Foreign Studies, Guangzhou 510006, ChinaOperation and Maintenance Center of Information and Communication, CSG EHV Power Transmission Company, Guangzhou 510663, ChinaMulti-source domain adaptation (MSDA) plays an important role in industrial model generalization. Recent efforts regarding MSDA focus on enhancing multi-domain distributional alignment while omitting three issues, e.g., the class-level discrepancy quantification, the unavailability of noisy pseudo labels, and source transferability discrimination, potentially resulting in suboptimal adaption performance. Therefore, we address these issues by proposing a prototype aggregation method that models the discrepancy between source and target domains at the class and domain levels. Our method achieves domain adaptation based on a group of prototypes (i.e., representative feature embeddings). A similarity score-based strategy is designed to quantify the transferability of each domain. At the class level, our method quantifies class-specific cross-domain discrepancy according to reliable target pseudo labels. At the domain level, our method establishes distributional alignment between noisy pseudo-labeled target samples and the source domain prototypes. Therefore, adaptation at the class and domain levels establishes a complementary mechanism to obtain accurate predictions. The results on three standard benchmarks demonstrate that our method outperforms most state-of-the-art methods. In addition, we provide further elaboration of the proposed method in light of the interpretable results obtained from the analysis experiments.https://www.mdpi.com/2227-7390/13/4/579multiple sourcesdomain adaptationprototype learningprototype aggregation
spellingShingle Min Huang
Zifeng Xie
Bo Sun
Ning Wang
Multi-Source Unsupervised Domain Adaptation with Prototype Aggregation
Mathematics
multiple sources
domain adaptation
prototype learning
prototype aggregation
title Multi-Source Unsupervised Domain Adaptation with Prototype Aggregation
title_full Multi-Source Unsupervised Domain Adaptation with Prototype Aggregation
title_fullStr Multi-Source Unsupervised Domain Adaptation with Prototype Aggregation
title_full_unstemmed Multi-Source Unsupervised Domain Adaptation with Prototype Aggregation
title_short Multi-Source Unsupervised Domain Adaptation with Prototype Aggregation
title_sort multi source unsupervised domain adaptation with prototype aggregation
topic multiple sources
domain adaptation
prototype learning
prototype aggregation
url https://www.mdpi.com/2227-7390/13/4/579
work_keys_str_mv AT minhuang multisourceunsuperviseddomainadaptationwithprototypeaggregation
AT zifengxie multisourceunsuperviseddomainadaptationwithprototypeaggregation
AT bosun multisourceunsuperviseddomainadaptationwithprototypeaggregation
AT ningwang multisourceunsuperviseddomainadaptationwithprototypeaggregation