Few-Shot Learning Quiz

Answers and explanations appear as you take the quiz.

1. In an N-way K-shot classification episode, what does K count?
   • The number of classes
   • Labeled support examples per class
   • All parameters in the model
   • The number of optimization epochs
2. When a large language model does in-context few-shot learning from a prompt, what happens to its weights?
   • They are fine-tuned slightly on the prompt examples
   • They are reset to a task-specific checkpoint
   • They stay unchanged; the examples only condition the prediction
   • They are averaged with the example embeddings
3. What is the key difference between zero-shot and few-shot prompting?
   • Zero-shot provides no worked examples; few-shot provides a few
   • Zero-shot uses a smaller model than few-shot
   • Zero-shot always performs better than few-shot
   • Zero-shot requires labeled training data while few-shot does not
4. How can prior learning contribute to a few-shot task?
   • It guarantees every arbitrary demonstration yields the intended rule
   • It requires copying each example as every final answer
   • It necessarily retrains all network weights during prompting
   • It can supply useful learned patterns or representations for interpreting the small support set
5. Which describes a metric-based few-shot method like a prototypical network?
   • It writes example labels into a database for lookup
   • It generates synthetic examples until it has thousands
   • It compares a new sample to the average embedding of each class's few examples and picks the nearest
   • It uses reinforcement learning from human feedback
6. A practitioner finds that simply reordering the examples in a few-shot prompt changes accuracy. What does this illustrate?
   • A hardware bug in the GPU
   • That few-shot performance can be sensitive to example order and formatting
   • That the model is overfitting to the test set
   • That zero-shot would be impossible
7. A class has embedded support points (0,0) and (2,0). What arithmetic-mean prototype does the stated prototypical method form?
   • (2,0)
   • (0,2)
   • (2,2)
   • (1,0)
8. Which distinction defines support and query roles in the original prototypical-network episode?
   • Support labels construct class prototypes; separate query items evaluate classification
   • Query labels must be copied into every support prototype
   • Support items contain no labels by definition
   • The two sets must always contain exactly the same sampled examples