Check your understanding of this module before moving on.
?
0/3
Checkpoint quiz 3 questions
A quick check on this module's key ideas before you move on
Question 01
Why is it risky to tune hyperparameters (like architecture or learning rate) by repeatedly checking performance on the same test set?
A is correct. If you keep adjusting your model based on test-set feedback, you're implicitly fitting your choices to that specific set, which means the test set no longer gives an unbiased estimate of how the model will perform on truly unseen data. A separate validation set should be used for tuning.
Question 02
What is a common pitfall when comparing two deep learning models trained with different random seeds, batch sizes, or numbers of epochs?
A is correct. Training is a stochastic process, and uncontrolled differences in seeds, batch size, or training length can shift results as much as or more than the actual change being studied. Fair comparisons require holding everything else as constant as possible.
Question 03
Why can a model that performs extremely well on a benchmark dataset still perform poorly in the real world?
A is correct. A model can latch onto patterns or biases specific to how a benchmark was collected, patterns that won't hold for the messier, more varied distribution of real-world inputs, leading to a gap between benchmark performance and real deployment performance.