Survey

Deep transfer learning for image classification: a survey

Sometimes collecting large amounts of training data is infeasible:

1. Insufficient data because the data is very rare or there are issues with privacy
2. Expensive to collect and/or label data
3. The long tail distribution where a small number of classes are very frequent and thus easy to model, while many more are rare and thus hard to model

Why learn from a small number of training samples?

1. From a cognitive science perspective to attempt to mimic the human ability to learn general concepts from a small number of examples
2. Restraints on compute resources