LIME Local Explanations Quiz

Answers and explanations appear as you take the quiz.

1. What does LIME fit around the selected prediction?
   • A replacement model guaranteed to match globally
   • A causal model of the observed real-world outcome
   • A simple interpretable surrogate approximating the original model locally
   • A single global feature ranking shared by all examples
2. Why are nearby perturbations weighted more heavily in a local explanation?
   • To make every perturbed feature equally common
   • To focus the surrogate on behavior near the instance
   • To guarantee perturbations match the true data distribution
   • To give distant decision regions equal influence on the local fit
3. A LIME explanation changes sharply when the sampling seed changes. What should the analyst do?
   • Report only the explanation with the clearest feature signs
   • Assume the original predictor's outputs must also change between runs
   • Treat sampling variation as proof of a causal relationship
   • Assess stability under reasonable settings before relying on it
4. Why can removing words independently create a poor text neighborhood?
   • The resulting sentences may be unnatural
   • The original word order is guaranteed to remain semantically equivalent
   • Independent removals guarantee samples from the true language distribution
   • Local weighting automatically restores missing grammatical relationships
5. What does a LIME feature contribution establish?
   • The real-world cause of the outcome
   • Global feature importance for every example
   • A local association in a surrogate explanation, not causation
   • That a prediction is correct
6. Which diagnostic directly checks whether a local surrogate represents the original model near an instance?
   • Measure the predictor's global training accuracy alone
   • Measure surrogate fidelity in the weighted neighborhood
   • Count only the number of selected surrogate features
   • Inspect only the surrogate coefficient signs
7. For an image explanation, what can change which regions appear important?
   • Only the display order of identical highlighted regions
   • Only the text used to name an unchanged region
   • Only the image file's name when pixels stay fixed
   • The segmentation used as interpretable units
8. Which conclusion is appropriately limited after inspecting one LIME explanation?
   • This surrogate approximates this model near this selected input under stated settings
   • The model uses features identically for all cases
   • The highlighted factor caused the real outcome
   • The whole model is fair and trustworthy