A lighthouse postcard archive has only a few dozen labeled captions but can start from a network already trained on a huge public text collection and adapt the last stages. What is that approach?
Select an answer to reveal the explanation.
Short Explanation
A few dozen labeled captions, and a network already trained on a huge public collection. That is transfer learning: reuse those weights and adapt. Training from scratch or writing a kernel is not that reuse.
Full Explanation
Transfer learning starts from weights learned on related data and adapts them to the local task. That is the usual associate choice when labeled examples are few. Training from scratch, inventing a new architecture, or writing a kernel is not that reuse.