Dynamical ensemble learning with model-friendly classifiers for domain adaptation

摘要: In the domain adaptation research, which recently becomes one of most important research directions in machine learning, source and target domains are with different underlying distributions. this paper, we propose an ensemble learning framework for adaptation. Owing to distribution differences between domains, weights final model sensitive examples. As a result, our method aims dynamically assign test examples by making use additional classifiers called model-friendly classifiers. The can judge base models predict well on specific example. Finally, give favorable experiments, firstly testify need dynamical based adaptation, then compare other classical methods real datasets. experimental results show that learn performing domain.