Contrastively Smoothed Class Alignment for Unsupervised Domain Adaptation

Shuyang Dai*, Yu Cheng, Yizhe Zhang, Zhe Gan, Jingjing Liu, Lawrence Carin

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Recent unsupervised approaches to domain adaptation primarily focus on minimizing the gap between the source and the target domains through refining the feature generator, in order to learn a better alignment between the two domains. This minimization can be achieved via a domain classifier to detect target-domain features that are divergent from source-domain features. However, when optimizing via such domain-classification discrepancy, ambiguous target samples that are not smoothly distributed on the low-dimensional data manifold are often missed. To solve this issue, we propose a novel Contrastively Smoothed Class Alignment (CoSCA) model, that explicitly incorporates both intra- and inter-class domain discrepancy to better align ambiguous target samples with the source domain. CoSCA estimates the underlying label hypothesis of target samples, and simultaneously adapts their feature representations by optimizing a proposed contrastive loss. In addition, Maximum Mean Discrepancy (MMD) is utilized to directly match features between source and target samples for better global alignment. Experiments on several benchmark datasets demonstrate that CoSCAoutperforms state-of-the-art approaches for unsupervised domain adaptation by producing more discriminative features.

Original languageEnglish (US)
Title of host publicationComputer Vision – ACCV 2020 - 15th Asian Conference on Computer Vision, 2020, Revised Selected Papers
EditorsHiroshi Ishikawa, Cheng-Lin Liu, Tomas Pajdla, Jianbo Shi
PublisherSpringer Science and Business Media Deutschland GmbH
Pages268-283
Number of pages16
ISBN (Print)9783030695378
DOIs
StatePublished - 2021
Externally publishedYes
Event15th Asian Conference on Computer Vision, ACCV 2020 - Virtual, Online
Duration: Nov 30 2020Dec 4 2020

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12625 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference15th Asian Conference on Computer Vision, ACCV 2020
CityVirtual, Online
Period11/30/2012/4/20

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint Dive into the research topics of 'Contrastively Smoothed Class Alignment for Unsupervised Domain Adaptation'. Together they form a unique fingerprint.

Cite this