Total points 3
1.
Question 1
Interpretation methods can be grouped based on (Check all that apply):

1 / 1 point

Whether they are complex or basic.


> Whether they are local or global.

Correct
You’ve got it. They can also be grouped by whether they generate interpretations which are local or global.


> Whether they are model specific, or model agnostic.

Correct
Keep it up! Model specific methods are limited to certain model types, while model agnostic methods are applied to any model after it is trained.


> Whether they are intrinsic or post-hoc.

Correct
Nice job! One way of grouping interpretability methods is by whether the model is intrinsically interpretable or treated as a black box, and external tools are used to analyze it.

2.
Question 2
One key aspect that helps improve interpretability is the presence of monotonic features.

1 / 1 point

> True


False

Correct
That's right! This matches our domain knowledge for many features in many kinds of problems. When we are trying to understand a model result, if the features are monotonic, it matches our intuition about the world's reality, which we are trying to model.

3.
Question 3
Many classic models are intrinsically interpretable models, such as the transparent, intuitive, and relatively easy to understand neural networks.

1 / 1 point

True


> False

Correct
Absolutely right! NNs’ complex architecture makes them “black boxes” when you try to interpret them.