Week 7

Question 1

Interpretability

Answer 1

The degree to which a human can understand the cause of a decision and consistently predict the result of a model.

Question 2

Explainable

Answer 2

When feature values of instances can be related to model prediction in a humanly understandable way.

Question 3

What levels of interpretability / explainability does Molnar state?

Answer 3

Interpretability is at a global level of the model, explainability is concerned with an individual prediction.

Question 4

Intrinsically interpretable models

Answer 4

Provide all means necessary for the decisions explanation. Is interpretable due to simplicity, has all info itself.

Question 5

Model agnostic explainable AI method

Answer 5

Explains any model, no matter the type

Question 6

Model specific explainable AI method

Answer 6

Accesses and uses the model internals.

Question 7

Counterfactual explanation

Answer 7

The one that is the closest to the instance that we’re trying to predict, with minimal changes, that gives a different prediction.

Question 8

CNN (abbreviation)

Answer 8

Convolutional neural network

Question 9

Expressive power

Answer 9

What is the structure of the explanations?
EG. is it ‘if-then’, a tree, natural language…

Question 10

Translucency

Answer 10

Describes how much the explanation method relies on looking into the machine learning model

Question 11

Portability (property of explanation method)

Answer 11

Measures the range of machine learning models with which the explanation can be used.

Question 12

Algorithmic complexity (property of explanation method)

Answer 12

The computational complexity of the explanation method.

Question 13

Accuracy (property of explanation)

Answer 13

How well does an explanation predict unseen data?

Question 14

Fidelity (property of explanation)

Answer 14

How well does the explanation approximate the prediction of the black box model?

Question 15

Certainty (property of explanation)

Answer 15

Does the explanation reflect the certainty of the machine learning model?

Question 16

Comprehensibility (property of an explanation)

Answer 16

How well do humans understand the explanations? How convincing are they?

Question 17

How is fidelity measured?

Answer 17

Objectively

Question 18

How is plausability measured?

Answer 18

By comprehensibility: it requires a user study.

Question 19

How is simulatability measured?

Answer 19

By measuring the degree that a human can calculate or predict the model’s outcome, given the explanation.

Question 20

What should a good explanation be?

Answer 20

• Contrastive
• Selective
• Social
• Truthful
• General and probable
• Consistent with prior beliefs.

Question 21

Black-box models

Answer 21

Require post-hoc explanations, cannot have perfect fidelity.

Question 22

What are the two families of interpretable models that we’re focusing on?

Answer 22

1. Linear models
2. Decision trees and decision rules

Question 23

Give the hierachy of linear models from big to small:

Answer 23

1. Generalized additive models
2. Generalized linear models
3. Linear models
4. Scoring systems

Question 24

Scoring system

Answer 24

A specialised type of linear model that gives an integer in a range as output.

Question 25

Generalized linear model

Question 26

Multivariate linear model

Question 27

Multivariate polynomial model

Question 28

Linearity

Answer 28

f(x+y) = f(x) + f(y)
and
f(cx) = c f(x)

Question 29

Homoscedasticity

Answer 29

Constant variance

Question 30

Name four assumptions for linear modeling:

Answer 30

1. Normality of the target variable
2. Homoscedasticity
3. Independent instance distribution
4. Absence of multicollinearity
5. Linearity

Question 31

Multicollinearity

Answer 31

When there are pairs of strongly correlated features in the data, so coloms are correlated.

Question 32

What does it mean when we have a modular view in the interpretation of linear models?

Answer 32

We assume all remaining feature values are fixed, so a change in a particular feature will be reflected in the outcome.

Question 33

Numerical feature weight (in the interpretation of linear models)

Answer 33

When all other features are constant, it is the change in outcome when the feature weight value is increased by one unit.

Question 34

Binary feature weight (in interpretation of linear models)

Answer 34

The contribution to the model outcome of the feature when it is set to one.

Question 35

Categorical feature with L categories (interpretation of linear models)

Question 36

One-hot-encoding

Question 37

Feature effect

Answer 37

Multiplication of the estimated weight and the normalized feature value.

Question 38

How can we model nonlinear component functions f_j (x_j) in GAM models?

Answer 38

We learn them greedyli and use splines.

Question 39

Splines

Question 40

Indicator functions

Answer 40

Are combinary. Give an output with a condition, so that output is the case if the condition is met. Otherwise another output will be given.