Use cases

In practice, which pretrained model should we reuse? Which layers should be frozen or fine-tuned? The answers to these questions depend on the similarity between the target task and the tasks that models have already been trained on, as well as the abundance of training samples for the new application.