Learning Curves and Diagnosis

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 does a validation learning curve help you assess?A. Whether the test set may be used repeatedly for tuningB. Performance on data not used for gradient updatesC. Only how fast the optimizer changes weightsD. The exact labels of future production examples

2. What pattern is consistent with high bias?A. Training and validation error are both highB. Training error is tiny and validation error much higherC. Both errors are always exactly zeroD. Only the file size is large

3. What pattern is consistent with high variance?A. Equally high errors caused by insufficient capacityB. A small deployment file by itselfC. A balanced number of examples per class by itselfD. Low training error with substantially worse validation error

4. You change the optimizer, augmentation, and architecture together and improve accuracy. What can you confidently conclude?A. All three changes must individually improve accuracyB. The old measurements are no longer usefulC. The combined setup improved, but the cause is not isolatedD. The optimizer alone caused the improvement

5. You select a configuration after many validation experiments. What should the untouched test set be used for?A. Replacing missing training labelsB. A final evaluation of the selected approachC. Choosing every new hyperparameter repeatedlyD. Updating gradients during training

Review this module from its first lesson →

Learning Curves and Diagnosis

Your result