E2: Entropy Discrimination and Energy Optimization for Source-free Universal Domain Adaptation

Meng Shen, Andy J. Ma*, Pong C. Yuen

*Corresponding author for this work

Research output: Chapter in book/report/conference proceedingConference proceedingpeer-review

Abstract

Universal domain adaptation (UniDA) transfers knowledge under both distribution and category shifts. Most UniDA methods accessible to source-domain data during model adaptation may result in privacy policy violation and source-data transfer inefficiency. To address this issue, we propose a novel source-free UniDA method coupling confidence-guided entropy discrimination and likelihood-induced energy optimization. The entropy-based separation of target-known and unknown classes is too conservative for known-class prediction. Thus, we derive the confidence-guided entropy by scaling the normalized prediction score with the known-class confidence, that more known-class samples are correctly predicted. Due to difficult estimation of the marginal distribution without source-domain data, we constrain the target-domain marginal distribution by maximizing (minimizing) the known (unknown)-class likelihood, which equals free energy optimization. Theoretically, the overall optimization amounts to decreasing and increasing internal energy of known and unknown classes in physics, respectively. Extensive experiments demonstrate the superiority of the proposed method.

Original languageEnglish
Title of host publicationProceedings - 2023 IEEE International Conference on Multimedia and Expo, ICME 2023
Place of PublicationBrisbane, Australia
PublisherIEEE
Pages2705-2710
Number of pages6
ISBN (Electronic)9781665468916
ISBN (Print)9781665468923
DOIs
Publication statusPublished - 10 Jul 2023
Event2023 IEEE International Conference on Multimedia and Expo, ICME 2023 - Brisbane Convention & Exhibition Centre, Brisbane, Australia
Duration: 10 Jul 202314 Jul 2023
https://www.2023.ieeeicme.org/index.php
https://www.2023.ieeeicme.org/program.php
https://ieeexplore.ieee.org/xpl/conhome/10219544/proceeding

Publication series

NameProceedings - IEEE International Conference on Multimedia and Expo
Volume2023-July
ISSN (Print)1945-7871
ISSN (Electronic)1945-788X

Conference

Conference2023 IEEE International Conference on Multimedia and Expo, ICME 2023
Country/TerritoryAustralia
CityBrisbane
Period10/07/2314/07/23
Internet address

Scopus Subject Areas

  • Computer Networks and Communications
  • Computer Science Applications

User-Defined Keywords

  • Confidence-guided Entropy
  • Energy
  • Source-free Domain Adaptation
  • Universal Domain Adaptation

Fingerprint

Dive into the research topics of 'E2: Entropy Discrimination and Energy Optimization for Source-free Universal Domain Adaptation'. Together they form a unique fingerprint.

Cite this