Discriminative adversarial domain generalization with meta-learning based cross-domain validation

The generalization capability of machine learning models, which refers to generalizing the knowledge for an “unseen” domain via from one or multiple seen domain(s), is great importance develop and deploy applications in real-world conditions. Domain Generalization (DG) techniques aim enhance such where learnt feature representation classifier are two crucial factors improve make decisions. In this paper, we propose Discriminative Adversarial (DADG) with meta-learning-based cross-domain validation. Our proposed framework tries learn a domain-invariant source domains generalize it unseen domains. It contains main components that work synergistically build domain-generalized Deep Neural Network (DNN) model: (i) discriminative adversarial learning, proactively learns generalized on “seen” domains, (ii) meta-learning based cross validation, simulates train/test shift applying training process. experimental evaluation, comprehensive comparison has been made among our approach other existing approaches three benchmark datasets. results shown DADG consistently outperforms strong baseline DeepAll, DG algorithms most evaluation cases.