Course Objectives:  The main objectives of the course are to 

● Understand the importance of explainability in AI and its impact on stakeholders. 

● Explore different techniques and methods for making AI systems explainable. 

● Analyze the trade-offs between model complexity and interpretability. 

● Examine the ethical and societal implications of XAI. 

● Apply XAI techniques to real-world datasets and scenarios. 

Follow the material at: https://github.com/smsamspublications/interpretable-ml-book-Christoph-Molnar 

UNIT I:  Introduction to Explainable AI (XAI): Motivations for XAI, Importance of interpretability and transparency Techniques for XAI, Model-specific interpretability methods (e.g., decision trees, rulebased systems) Model-agnostic interpretability methods (e.g., LIME,SHAP) Post-hoc explanation techniques (e.g., feature importance, counterfactual explanation.

UNIT II: Interpretable Models: Linear models, Decision trees and rule-based systems Symbolic AI approaches, Interpretable Neural Networks, Sparse neural networks, Attention mechanisms, Layer-wise relevance propagation (LRP) 

UNIT III: Methods for Explainable AI: Partial Dependence Plot (PDP), Conformal Prediction, Individual Conditional Expectation (ICE), Feature Importance, Saliency Maps, Local Interpretable Model-Agnostic Explanations (LIME), SHAP, Integrated Gradient (IG), Explainability for Linear Models, Non-linear models and Deep Learning Models. 

UNIT IV: Evaluation of XAI Methods: Quantitative metrics for interpretability, Human centric evaluation methods, Ethical and Societal Implications of XAIB, is and fairness in interpretable AI, Trust and accountability in AI systems, Regulatory considerations.

UNIT V: Applications of XAI: Healthcare (e.g., medical diagnosis, personalized treatment) Finance (e.g., credit scoring, fraud detection), Autonomous systems (e.g., self-driving cars, drones). Explainability in Time Series Forecasting, Natural Language Processing, and Computer Vision 

Text Books: 

1. "Interpretable Machine Learning" by Christoph Molnar 

2. "Explainable AI: Interpreting, Explaining and Visualizing Deep Learning" by L. Liu and G. Hu 

Reference Books: 

1. "Interpretable Machine Learning: A Guide for Making Black Box Models Explainable" by Christoph Molnar 

2. "Explainable AI: Interpreting, Explaining and Visualizing Deep Learning" by L. Liu and G. Hu – 

3. "Explainable AI in Healthcare: Exploring Interpretable Models and Learning from Patient Data" edited by F. E. Elsayed and B. G. Stoecklin 

Online Resources: 1. https://christophm.github.io/interpretable-ml-book/    

UNIT-1                                         5-Mark Questions

Introduction to Explainable AI (XAI)

1. Define Explainable Artificial Intelligence (XAI) and explain its objectives. (L1)

2. Discuss the need for Explainable AI in modern machine learning systems. (L2)

3. Explain how XAI can improve decision-making in healthcare or banking applications. (L3)

Motivations for XAI

4. Explain the major motivations for Explainable AI. (L1)

5. Discuss the role of trust and transparency in the adoption of AI systems. (L2)

6. Illustrate how XAI helps reduce risks in high-stakes applications. (L3)

7. Explain how XAI supports accountability and fairness in AI systems. (L2)

8. Discuss the importance of regulatory compliance as motivation for XAI. (L2)

Importance of Interpretability and Transparency

9. Define interpretability and transparency in XAI. (L1)

10.  Differentiate between interpretability and transparency with examples. (L2)

11.  Explain the benefits of interpretability in machine learning models. (L2)

12.  Discuss how transparency improves trust in AI systems. (L2)

13.  Illustrate the importance of interpretability in medical diagnosis systems. (L3)

Model-Specific Interpretability Methods

14.  Explain the structure and work of a Decision Tree. (L1)

15.  Discuss the advantages and limitations of Decision Trees in XAI. (L2)

16.  Explain Rule-Based Systems with suitable examples. (L1)

17.  Differentiate between Decision Trees and Rule-Based Systems. (L2)

18.  Construct a Decision Tree for a simple loan approval problem. (L3)

19.  Apply IF-THEN rules to classify student performance. (L3)

Model-Agnostic Interpretability Methods

20.  Explain the concept of model-agnostic interpretability methods. (L1)

21.  Describe the working of LIME with a suitable example. (L1)

22.  Explain the principles of SHAP and its importance in XAI. (L1)

23.  Compare LIME and SHAP techniques. (L2)

24.  Discuss the advantages and limitations of LIME. (L2)

25.  Explain why SHAP provides more consistent explanations than LIME. (L2)

26.  Apply LIME to explain a spam email classification result. (L3)

27.  Apply SHAP to explain loan approval prediction. (L3)

Post-Hoc Explanation Techniques

28.  What are post-hoc explanation techniques? Explain their significance. (L1)

29.  Define feature importance and explain its role in model interpretation. (L1)

30.  Explain permutation feature importance with an example. (L1)

31.  Define counterfactual explanations and discuss their importance. (L1)

32.  Differentiate between feature importance and counterfactual explanations. (L2)

33.  Discuss the benefits and limitations of feature importance techniques. (L2)

34.  Explain how counterfactual explanations improve user understanding. (L2)

35.  Apply feature importance to explain a disease prediction model. (L3)

36.  Generate a counterfactual explanation for a rejected loan application. (L3)

2-Mark Questions

Introduction to Explainable AI (XAI)

1. What is Explainable Artificial Intelligence (XAI)? (L1)

2. What is meant by explainability in AI? (L1)

3. Defining a black-box model. (L1)

4. Why is XAI important in AI systems? (L2)

5. How does XAI improve user trust? (L2)

6. Give one real-world application of XAI. (L3)

Motivations for XAI

7. List any two motivations for XAI. (L1)

8. What is accountability in AI? (L1)

9. What is fairness in AI systems? (L1)

10.  Why is trust important in AI adoption? (L2)

11.  How does XAI help detect bias? (L2)

12.  Explain one situation where XAI improves transparency. (L3)

Interpretability and Transparency

13.  Define interpretability. (L1)

14.  Define transparency. (L1)

15.  What is an interpretable model? (L1)

16.  Differentiate between interpretability and transparency. (L2)

17.  Why is interpretability important in healthcare applications? (L2)

18.  Illustrate a transparent AI decision-making process. (L3)

Decision Trees and Rule-Based Systems

19.  What is a Decision Tree? (L1)

20.  What is a root node? (L1)

21.  What is a leaf node? (L1)

22.  Defining a Rule-Based System. (L1)

23.  What is an IF-THEN rule? (L1)

24.  Why are Decision Trees considered interpretable? (L2)

25.  How do Rule-Based Systems support explainability? (L2)

26.  Construct a simple IF-THEN rule for loan approval. (L3)

27.  Apply a Decision Tree to classify a student's results. (L3)

LIME and SHAP

28.  What does LIME stand for? (L1)

29.  What does SHAP stand for? (L1)

30.  Define a model-agnostic explanation method. (L1)

31.  Which technique is based on game theory? (L1)

32.  What are Shapley values? (L1)

33.  Explain the purpose of LIME. (L2)

34.  How does SHAP explain feature contributions? (L2)

35.  Difference between LIME and SHAP. (L2)

36.  Apply LIME to explain text classification results. (L3)

37.  Apply SHAP to explain a credit scoring model. (L3)

Post-Hoc Explanation Techniques

38.  What is a post-hoc explanation? (L1)

39.  Define feature importance. (L1)

40.  What is permutation feature importance? (L1)

41.  Define a counterfactual explanation. (L1)

42.  What is a local explanation? (L1)

43.  How does feature importance help understand model behavior? (L2)

44.  Why are counterfactual explanations useful? (L2)

45.  Differentiate between local and global explanations. (L2)

46.  Provide a counterfactual explanation for loan rejection. (L3)

47.  Apply feature importance to explain a disease prediction outcome. (L3)

Comprehensive Questions

48.  Differentiate between model-specific and model-agnostic interpretability methods. (L2)

49.  How does XAI support trustworthy AI systems? (L2)

50.  Explain one application of SHAP in healthcare. (L3)

51.  Why is transparency important in financial AI systems? (L2)

52.  How can feature importance be used for model debugging? (L3)

53.  Compare Decision Trees and LIME as explanation techniques. (L2)

54.  Explain the role of XAI in reducing bias. (L2)

55.  Apply counterfactual explanations to improve customer understanding of AI decisions. (L3)

56.  Illustrate the use of SHAP for explaining loan approval predictions. (L3)

UNIT-2                                          5-Mark Questions

Interpretable Models: Linear Models

1. Explain linear models as interpretable machine learning models with a suitable example. (L1)

2. Discuss the advantages and limitations of linear models in Explainable AI. (L2)

3. Illustrate how feature coefficients in a linear regression model help interpret predictions. (L3)

Decision Trees and Rule-Based Systems

4. Explain the structure and work of a Decision Tree with an example. (L1)

5. Differentiate between Decision Trees and Rule-Based Systems. (L2)

6. Construct a simple Decision Tree for predicting student performance and explaining the prediction process. (L3)

7. Explain Rule-Based Systems and their role in achieving interpretability. (L1)

8. Discuss the advantages of IF-THEN rules in explainable decision-making systems. (L2)

Symbolic AI Approaches

9. What is Symbolic AI? Explain its key characteristics and applications. (L1)

10.  Discuss how symbolic reasoning contributes to explainability in AI systems. (L2)

11.  Apply symbolic rules to solve a simple medical diagnosis problem. (L3)

Interpretable Neural Networks

12.  What are Interpretable Neural Networks? Explain their importance in XAI. (L1)

13.  Discuss the characteristics of interpretable neural networks compared with traditional deep neural networks. (L2)

14.  Explain how an interpretable neural network can be used in a healthcare application. (L3)

Sparse Neural Networks

15.  Define Sparse Neural Networks and explain their working principles. (L1)

16.  Discuss how sparsity improves model interpretability and efficiency. (L2)

17.  Explain a practical scenario where Sparse Neural Networks enhance explainability. (L3)

Attention Mechanisms

18.  Explain the concept of Attention Mechanisms in neural networks. (L1)

19.  Discuss how attention weights improve interpretability in deep learning models. (L2)

20.  Illustrate the role of attention mechanisms in machine translation or text summarization. (L3)

Layer-Wise Relevance Propagation (LRP)

21.  What is Layer-Wise Relevance Propagation (LRP)? Explain its purpose in XAI. (L1)

22.  Discuss the working principle of LRP in explaining neural network predictions. (L2)

23.  Apply LRP to explain the classification of a medical image by a neural network. (L3)

2-Mark Questions

Interpretable Models: Linear Models

1. What is a linear model? (L1)

2. Give one example of a linear model. (L1)

3. What is a coefficient in a linear model? (L1)

4. Why are linear models considered interpretable? (L2)

5. How can feature coefficients help explain predictions? (L2)

6. Explain a simple application of a linear model in prediction. (L3)

Decision Trees and Rule-Based Systems

7. What is a Decision Tree? (L1)

8. What is a root node in a Decision Tree? (L1)

9. What is a leaf node in a Decision Tree? (L1)

10.  Defining a Rule-Based System. (L1)

11.  What is an IF-THEN rule? (L1)

12.  Why are Decision Trees considered interpretable? (L2)

13.  How do Rule-Based Systems improve explainability? (L2)

14.  Differentiate between a Decision Tree and a Rule-Based System. (L2)

15.  Construct a simple IF-THEN rule for loan approval. (L3)

16.  Apply a Decision Tree to classify a student's performance. (L3)

Symbolic AI Approaches

17.  Define Symbolic AI. (L1)

18.  What is symbolic reasoning? (L1)

19.  What are knowledge-based systems? (L1)

20.  Why is Symbolic AI considered explainable? (L2)

21.  How do symbols represent knowledge in Symbolic AI? (L2)

22.  Apply a symbolic rule to diagnose a disease symptom. (L3)

Interpretable Neural Networks

23.  What is an Interpretable Neural Network? (L1)

24.  Why are traditional neural networks called black-box models? (L1)

25.  State one advantage of Interpretable Neural Networks. (L1)

26.  How do Interpretable Neural Networks support XAI? (L2)

27.  Explain the need for interpretability in neural networks. (L2)

28.  Suggest a use case of an Interpretable Neural Network in healthcare. (L3)

Sparse Neural Networks

29.  What is a Sparse Neural Network? (L1)

30.  What is sparsity in neural networks? (L1)

31.  State one benefit of Sparse Neural Networks. (L1)

32.  How does sparsity improve interpretability? (L2)

33.  Why are Sparse Neural Networks computationally efficient? (L2)

34.  Apply Sparse Neural Networks to a feature-selection problem. (L3)

Attention Mechanisms

35.  What is Attention Mechanism? (L1)

36.  What are attention weights? (L1)

37.  Name one application for Attention Mechanisms. (L1)

38.  How does attention improve model interpretability? (L2)

39.  Explain the role of attention in sequence modeling. (L2)

40.  Illustrate the use of attention in a language translation task. (L3)

Layer-Wise Relevance Propagation (LRP)

41.  What does LRP stand for? (L1)

42.  Define Layer-Wise Relevance Propagation. (L1)

43.  What is the purpose of LRP in XAI? (L1)

44.  How does LRP explain neural network predictions? (L2)

45.  Why is LRP useful for image classification models? (L2)

46.  Apply LRP to identify important pixels in an image classification task. (L3)

Mixed Comprehensive Questions

47.  Differentiate between Linear Models and Decision Trees. (L2)

48.  Compare Symbolic AI and Neural Network approaches. (L2)

49.  How do Attention Mechanisms contribute to Explainable AI? (L2)

50.  Explain one application for LRP in healthcare. (L3)

51.  How can Sparse Neural Networks improve transparency in AI? (L2)

52.  Give a real-world example of a Rule-Based System. (L3)

53.  How does Symbolic AI support knowledge representation? (L2)

54.  Why are interpretable models preferred in high-stakes applications? (L2)

55.  Apply an attention-based model to explain a text classification result. (L3)

56.  Illustrate how LRP can be used to validate a neural network decision. (L3)