Category-aware and reweighting-based multi-source domain adaptation algorithm
DOI:
CSTR:
Author:
Affiliation:

1.School of Mathematical Sciences and Information Engineering,Guizhou Minzu University,Guiyang 550025,China; 2.Guizhou Minzu University,Key Laboratory of Pattern Recognition and Intelligent System,Guiyang 550025,China

Clc Number:

TP181;TN911.7

Fund Project:

  • Article
  • |
  • Figures
  • |
  • Metrics
  • |
  • Reference
  • |
  • Related
  • |
  • Cited by
  • |
  • Materials
  • |
  • Comments
    Abstract:

    Multi-source domain adaptation is an important branch of transfer learning. Category shift, a prominent challenge in this field, stems from the mismatch between category distributions in the source and target domains. To address this problem, a category-aware and reweighting-based multi-source domain adaptation algorithm is proposed. The algorithm enhances positive transfer between similar categories through a category-aware strategy and introduces a reweighting moment matching strategy to reduce distribution differences at various levels. Additionally, adaptive weights are constructed using pseudo-labels to effectively mitigate the impact of category shift. Experimental results on the Digits-Five and Office-Caltech10 datasets show that the proposed algorithm achieves classification accuracies of 94.11% and 97.18%, respectively. These results indicate that the proposed algorithm significantly improves accuracy in scenarios with category shift compared to current typical multi-source domain adaptation algorithms.

    Reference
    Related
    Cited by
Get Citation
Share
Article Metrics
  • Abstract:
  • PDF:
  • HTML:
  • Cited by:
History
  • Received:
  • Revised:
  • Adopted:
  • Online: December 11,2024
  • Published:
Article QR Code