Domain adaptation is further classified as homogenous and heterogeneous DA . The source and target domains have similar or same dimensionality feature spaces in homogenous DA and different feature spaces in heterogeneous DA. Domain generalization can also be classified as homogenous and heterogeneous DG . In homogenous DG, label data is not available from the target domain, but the target label space is the same as the source label space. Whereas in heterogeneous DG, label spaces are different or disjoint at the source and target . Domain adaptation is classified as Supervised, Unsupervised, and Semi-supervised based on the availability of labels for the source domain.
Rahate, A., Walambe, R., Ramanna, S., & Kotecha, K. (2021). Multimodal Co-learning: Challenges, Applications with Datasets, Recent Advances and Future Directions. arXiv preprint arXiv:2107.13782.
DA aims to maximize the performance on a given target domain using existing training source domain(s). The difference between DA and DG is that DA has access to the target domain data while DG cannot see them during training. This makes DG more challenging than DA but more realistic and favorable in practical applications.
Wang, J., Lan, C., Liu, C., Ouyang, Y., Zeng, W., & Qin, T. (2021). Generalizing to Unseen Domains: A Survey on Domain Generalization. arXiv preprint arXiv:2103.03097.
Domain Generalization with MixStyle
Open Domain Generalization with Domain-Augmented Meta-Learning
Adaptive Methods for Real-World Domain Generalization
Heterogeneous Domain Generalization via Domain Mixup