Involves transferring knowledge from a source domain to a closely related target domain. This proves valuable when labeled data in the target domain is scarce or absent.
Occurs when the source and target domains align, but the tasks differ. The pretrained model, already familiar with the source data, speeds up learning for new functions.