Hyperparameter Tuning Quiz

Answers and explanations appear as you take the quiz.

1. What distinguishes a hyperparameter from a regular model parameter?
   • Hyperparameters are set before training and are not learned directly from the data
   • Hyperparameters are learned automatically by gradient descent during training
   • Hyperparameters only exist in linear models
   • There is no real difference between them
2. Which is commonly considered the single most impactful hyperparameter to tune in deep learning?
   • The random seed
   • The learning rate
   • The file format of the dataset
   • The name of the optimizer variable
3. Why does random search often outperform grid search for hyperparameter tuning?
   • It always tries more total combinations
   • It does not require a validation set
   • It samples the space more efficiently, finding good values without exhaustively covering every grid point
   • It learns the weights and hyperparameters at the same time
4. How does Bayesian optimization choose which hyperparameter configuration to try next?
   • It picks configurations completely at random every time
   • It always picks the configuration with the most layers
   • It copies the previous trial's settings exactly
   • It builds a model of which settings look promising and balances exploring uncertainty with exploiting good results
5. Why must the final test set remain untouched during hyperparameter tuning?
   • Because tuning against it would leak information and give an overoptimistic estimate of real performance
   • Because the test set is used to compute the weights
   • Because the test set is always smaller than the validation set
   • Because hyperparameters are stored in the test set
6. What problem does an early-stopping scheme like Hyperband solve in hyperparameter tuning?
   • It removes the need for a learning rate
   • It saves compute by killing underperforming trials early and giving resources to promising ones
   • It guarantees the model will not overfit
   • It converts hyperparameters into learnable parameters
7. In Bayesian hyperparameter optimization, what role does the probabilistic surrogate model play?
   • It replaces training every candidate with a guaranteed exact score
   • It updates model weights during the final test
   • It estimates how promising configurations may be and guides the next trial
   • It labels the training examples
8. Why should tuning trials be compared on a validation set rather than the final test set?
   • The test set cannot contain labels
   • The validation set is always larger
   • The model cannot predict on a test set
   • Repeated test-set feedback would leak information and make the final estimate overoptimistic