Domain Adaptation
Domain Adaptation is about learning to adapt to test data even when the distribution is different from the training data.
This paper proposes a domain adversarial learning method that can achieve domain adaptation.
Architecture
The architecture consists of 3 parts; The Feature Extractor, Class Predictor(This part depends on the computer vision task), and the Domain Classifier.
The training steps are the following.
- The Feature Extractor and the Class Predictor will be trained with the source domain(training dataset).
- The Features Extractor and the Domain Classifier(predict which domain the input belongs to) will be trained with both the source domain and the target domain(Test set).
- Calculate the loss for both heads to learn a feature extractor that can understand both domains.
