Domain Adaptation Quiz

Answers and explanations appear as you take the quiz.

1. What problem is domain adaptation primarily designed to solve?
   • Reducing the number of model parameters
   • Speeding up matrix multiplication on GPUs
   • Performance drop when test data differs from training data
   • Generating synthetic training labels automatically
2. In a domain-adversarial network, what does the gradient reversal layer accomplish?
   • It pushes the feature extractor to make source and target indistinguishable
   • It deletes the domain classifier during inference
   • It doubles the learning rate for faster training
   • It converts regression outputs into classification labels
3. What makes 'unsupervised' domain adaptation especially useful?
   • It requires no model training at all
   • It only works on image data
   • It eliminates the need for a source domain
   • It needs no labeled data in the target domain
4. Which scenario is a classic example of domain shift?
   • A model trained and tested on the exact same dataset
   • A tumor classifier trained on one scanner failing on a different scanner
   • A model that runs faster after quantization
   • A network with more layers than another network
5. What is a 'source-free' domain adaptation method designed to avoid?
   • Accessing the original source training data during adaptation
   • Using any neural network architecture
   • Producing any output predictions
   • Running on modern GPUs
6. A team starts adaptation from a broadly pretrained model. What should it still measure?
   • Only the parameter count
   • Only source training accuracy
   • Performance on the actual target data and task
   • Whether target evaluation can be skipped
7. Which DANN head supplies the objective to distinguish source examples from target examples?
   • The task label vocabulary
   • The domain classifier
   • The optimizer checkpoint file
   • A fixed output embedding dimension
8. Does gradient reversal mean changing a positive class label into a negative class label?
   • Yes; every task label is negated
   • Yes; all target samples are relabeled as source
   • No; it freezes every feature weight
   • No; it reverses the domain gradient into the feature extractor