Source-free Domain Adaptation Requires Penalized Diversity.
Laya Rafiee SevyeriIvaxi ShethFarhood FarahnakAlexandre SeeSamira Ebrahimi KahouThomas FevensMohammad HavaeiPublished in: CoRR (2023)
Keyphrases
- domain adaptation
- multiple sources
- target domain
- source domain
- cross domain
- labeled data
- transfer learning
- semi supervised
- training data
- manifold alignment
- semi supervised learning
- test data
- maximum likelihood
- source and target domains
- data sets
- covariate shift
- unlabeled data
- domain specific
- general purpose
- training samples
- co training
- sentiment classification
- feature selection
- data mining