week 3

Question 1

What is XAI (Explainable AI)?

Answer 1

XAI focuses on making AI models transparent and understandable, aiding debugging, improvement, and building trust.

Question 2

What are the two main approaches to achieving model understanding?

Answer 2

1. Build inherently explainable models (e.g., decision trees, linear regression). 2. Explain pre-built models in a post-hoc manner (e.g., LIME, SHAP).

Question 3

What are intrinsic methods in XAI?

Answer 3

Intrinsic methods are explanations built into the model itself, such as decision trees or linear regression models.

Question 4

What are post-hoc methods in XAI?

Answer 4

Post-hoc methods are explanations applied after the model is built, like LIME or SHAP, and can be model-agnostic.

Question 5

What is the difference between model-specific and model-agnostic methods?

Answer 5

Model-specific methods are tailored to particular model types, while model-agnostic methods can be applied to any model.

Question 6

What are some examples of model-specific techniques?

Answer 6

Techniques include ANOVA for statistical analysis, variable importance in random forests, and partial dependence plots.

Question 7

How does LIME explain model predictions?

Answer 7

LIME creates a simple, interpretable model around the specific instance being explained by perturbing feature values and fitting a local surrogate model.

Question 8

What is the main goal of LIME’s objective function?

Answer 8

To create a surrogate model that is both faithful to the original complex model and simple enough to be interpretable in the local neighborhood of the instance.

Question 9

What is SHAP and how does it work?

Answer 9

SHAP (SHapley Additive exPlanations) attributes model predictions to features using game theory principles, providing feature importance and direction of influence.

Question 10

What are local vs. global explanations?

Answer 10

Local explanations clarify why a specific instance was predicted, while global explanations elucidate the overall model behavior.

Question 11

Why is it important to evaluate explanations in XAI?

Answer 11

To ensure explanations accurately reflect model reasoning, meet stakeholder needs, and identify potential biases.

Question 12

What are key evaluation criteria for XAI explanations?

Answer 12

Fidelity, comprehensibility, sufficiency, and trustworthiness are critical criteria for evaluating explanations.

Question 13

What are the levels of evaluating explanations?

Answer 13

Evaluation can occur at the application level (real-world testing), human level (simplified tasks for laypersons), or function level (proxy tasks without humans).

Question 14

What properties make explanations effective?

Answer 14

Expressive power, translucency, portability, and algorithmic complexity are properties that affect explanation quality.

Question 15

What characteristics define good human-friendly explanations?

Answer 15

Good explanations are contrastive, causal, counterfactual, and tailored to the audience’s context.

Question 16

What is the importance of benchmarks in XAI?

Answer 16

Benchmarks enable systematic comparison of explanation methods and provide standardized datasets and evaluation metrics.

Question 17

What are some challenges in XAI evaluation?

Answer 17

Challenges include difficulty in assessing explanation quality and lack of principled guidelines for practitioners.

Question 18

What are the properties of individual explanations in XAI?

Answer 18

The properties include accuracy, fidelity, consistency, and stability, which ensure that explanations are reliable and useful.

Question 19

What does fidelity mean in the context of XAI explanations?

Answer 19

Fidelity refers to how closely the explanation approximates the model’s predictions, ensuring that the explanation accurately reflects the model’s behavior.

Question 20

What is the significance of consistency in individual explanations?

Answer 20

Consistency ensures that different models provide similar explanations for the same input data, promoting reliability across models.

Question 21

Why is stability important in individual explanations?

Answer 21

Stability means that similar instances should receive similar explanations, enhancing trust and reliability in model predictions.

Question 22

What does comprehensibility mean in evaluating XAI explanations?

Answer 22

Comprehensibility measures how easy it is for humans to understand the explanation, taking into account the target audience’s knowledge level.

Question 23

What is the purpose of evaluating explanations at the application level?

Answer 23

Evaluating at the application level involves real-world testing by end users to ensure that explanations are practical, useful, and understandable in actual settings.

Question 24

How are explanations evaluated at the human level?

Answer 24

At the human level, explanations are tested with simplified tasks for laypersons to assess their understandability and usability without requiring deep technical knowledge.

Question 25

What is the function level of explanation evaluation?

Answer 25

The function level uses proxy tasks, such as comparing explanation fidelity to model predictions on test datasets, without involving human subjects.

Question 26

What is expressive power in XAI explanations?

Answer 26

Expressive power refers to the ability of an explanation to succinctly capture the structure of the model’s behavior, often through decision rules or visual representations.

Question 27

What does translucency mean in the context of explanations?

Answer 27

Translucency describes the reliance of an explanation on the internal workings of the model, such as using feature weights in a linear model to clarify decision-making.

Question 28

How does portability apply to XAI explanations?

Answer 28

Portability is the ability of an explanation technique to be applied across different model types, such as SHAP values being used with both tree-based models and neural networks.

Question 29

What is algorithmic complexity in XAI evaluations?

Answer 29

Algorithmic complexity addresses the computational demands of generating explanations, which can impact the feasibility of using certain methods with large datasets or complex models.

Question 30

Why is it important to assess the trustworthiness of explanations?

Answer 30

Trustworthiness ensures that explanations are reliable, unbiased, and accurately reflect the model’s decision-making, fostering confidence in the model’s predictions.

Question 31

How does SHAP handle feature interaction?

Answer 31

SHAP captures interaction effects between features by attributing contributions to the prediction based on combinations of feature values, providing a more nuanced interpretation.

Question 32

What are contrastive explanations?

Answer 32

Contrastive explanations highlight the differences between outcomes, explaining why one scenario occurs over another by focusing on the differing features.

Question 33

What are counterfactual explanations?

Answer 33

Counterfactual explanations provide insights into what changes would be necessary to alter the outcome, helping to understand the conditions under which a different result would occur.

Question 34

What does it mean for an explanation to be causal?

Answer 34

Causal explanations focus on the key causes that directly impact the prediction, highlighting the most influential features driving the model’s decision.