6. Advanced Topics

Interpretability — Quiz

Test your understanding of interpretability with 5 practice questions.

Practice Questions

Question 1

Which of the following interpretability methods is most suitable for explaining a specific prediction of a complex NLP model by creating a locally faithful, interpretable surrogate model?

Question 2

In the context of responsible AI, when an NLP model's predictions are used to make high-stakes decisions (e.g., loan applications, medical diagnoses), which of the following is the most critical aspect of 'explaining model predictions responsibly'?

Question 3

Consider a Transformer-based NLP model. If an attention head consistently assigns high attention scores to the token 'UNK' (unknown token) across various sentences, what is the most plausible interpretation of this behavior?

Question 4

A researcher is designing a probing task to determine if a pre-trained language model encodes information about the grammatical number (singular/plural) of nouns. Which of the following would be the most effective design for this probing task?

Question 5

Which of the following interpretability techniques is most effective for identifying the minimal set of input features (e.g., words or phrases) that, if removed, would most significantly change an NLP model's prediction for a given instance, thereby revealing critical dependencies?