51. Interpretability and explainability

Inspect models with saliency methods, feature attribution, activation analysis, probes, counterfactuals, and example-based explanations. Match explanation methods to the audience and test whether an explanation is stable and faithful.