

Transfer Learning: train on one src and test on one target ; different tasks; target domain accessed in trainning;
Domain Adaption: train on src(one or more) and test on one given target; target domain accessed in trainning; A special kind of Transfer learning.
Zero-shot learning: train a model and test it with unseen categories.
Domain Generalization: train on srcs and test on one unseen target; target is not accessed in trainning.
learn a representation function g : X → Z that maps the original input data x to some representation space Z, so that the representation distributions of the two domains become closer; → domain-invairant representations