Transfer Learning and Representations

Five questions. Choose one answer for each, then submit to see your score and explanations.

The practice target is 80%. You can retry as often as you like, and every lesson stays accessible.

You can practice as a guest. Sign in before starting an attempt to save your scores.

1. What is transfer learning?A. Moving test labels into the training setB. Copying an image to another folderC. Guaranteeing that two tasks have equal accuracyD. Reusing knowledge learned on one task or dataset for another

2. What does freezing a pretrained backbone mean?A. Preventing any forward passB. Replacing every activation with zeroC. Keeping its parameters unchanged during the current training stageD. Deleting its learned parameters

3. What is an embedding?A. Only a single final probabilityB. A learned vector representation of an inputC. An original image stored without any transformationD. A guarantee that every nearby vector has the same label

4. You have a small labeled dataset and a suitable pretrained image model. What is a sensible first baseline?A. Train a new task head with the backbone frozen, then evaluateB. Randomize every weight and remove validationC. Train only on the held-out test splitD. Assume the original output classes match your task

5. A partially fine-tuned model improves training accuracy but damages validation performance. What is a reasonable next step?A. Always unfreeze more layers and increase the learning rateB. Report only training accuracyC. Delete the previous baselineD. Compare a lower backbone learning rate or fewer unfrozen layers

Review this module from its first lesson →

Transfer Learning and Representations

Your result