Transfer learning and domain adaptation¶
In transfer learning we adapt an network that was learned to perform well in task P1 to perform well in task P2. This is most useful if we have a little information about P2, and P1 provides an good basis of generalization.
useful in computer vision where we can share edge detectors
Domain adaptation¶
special case of transfer learning where the input distance between P1 and P2 differs.
Example:
sentiment analysis trained on news that is applied on reviews
One shot Learning and Zero shot learning¶
one shot learning we are given a single labeled item for transfer learning
zero shot we have no labeled item