Artificial Intelligence and Application Development Notes | B.Sc. (Information Technology) Semester 5 | Mumbai University | munotes

Contents

Module I Foundations and theoretical approach

1. What Artificial Intelligence Is 1
2. The Turing Test, Read Out of Turing's Own Paper 7
3. The Disciplines AI Was Built From 12
4. The State of the Art 17
5. The Risks and the Benefits of AI 21
6. Agents and Environments 25
7. Rationality, and What a Performance Measure Is 30
8. Omniscience, Learning and Autonomy 35
9. PEAS: Specifying a Task Environment 39
10. The Six Dimensions of a Task Environment 45
11. The Table-Driven Agent 51
12. Simple Reflex Agents 56
13. Model-Based Reflex Agents 61
14. Goal-Based Agents 66
15. Utility-Based Agents 71
16. Learning Agents 76
17. How an Agent Represents the World 81
18. Problem-Solving Agents 86
19. Example Problem: The 8-Puzzle 91
20. Example Problem: The 8-Queens Problem 97
21. Example Problem: Route Finding 102
22. The Water-Jug Problem 107
23. Nodes, the Frontier and the Reached Table 113
24. Best-First Search: The One Algorithm Behind All of Them 119
25. How a Search Strategy Is Measured 125
26. Breadth-First Search 131
27. Breadth-First Search on the 8-Puzzle 137
28. Uniform-Cost Search 143
29. Uniform-Cost Search from Arad to Bucharest 148
30. Depth-First Search 153
31. Depth-First Search on the Water Jugs 159
32. Depth-Limited and Iterative Deepening Search 165
33. The Uninformed Strategies Compared 171
34. Greedy Best-First Search 177
35. A* Search 183
36. Admissibility, Consistency, and Why A* Is Optimal 188
37. A* on the 8-Puzzle with Manhattan Distance 194
38. A* on the Road Map with Straight-Line Distance 200
39. Heuristic Functions: Where One Comes From 206
40. Comparing Two Heuristics: Dominance and the Branching Factor 212
41. Memory-Bounded Search: IDA* and Recursive Best-First Search 218
42. Acting Under Uncertainty 224
43. Basic Probability Notation 230
44. The Full Joint Distribution 236
45. Inference by Enumeration 241
46. Independence and Conditional Independence 247
47. Bayes Rule, and Using It for Diagnosis 253
48. The Base-Rate Fallacy 258
49. The Naive Bayes Classifier 263
50. Naive Bayes on Text, and the Zero-Frequency Problem 268
51. The Wumpus World Revisited 274
52. Representing Knowledge in an Uncertain Domain 280
53. The Burglary and Alarm Network 286
54. The Semantics of a Bayesian Network 292
55. Conditional Independence in a Network, and the Markov Blanket 298
56. Exact Inference by Enumeration 304
57. Variable Elimination 310
58. Approximate Inference: Rejection Sampling 316
59. Likelihood Weighting 322
60. Gibbs Sampling 328
61. Causal Networks 334
62. Doing This with a Library: pgmpy 339

Module II AI Application Development and Ethics

1. Forms of Learning 345
2. Supervised Learning 351
3. Bias, Variance and the Trade-Off Between Them 357
4. Learning Decision Trees 363
5. Entropy and Information Gain 369
6. Building a Decision Tree by Hand 375
7. Model Selection 381
8. Loss Functions and Regularisation 387
9. The Theory of Learning: PAC Learning 393
10. Linear Regression 399
11. Gradient Descent 405
12. Linear Classification and the Perceptron 411
13. Logistic Regression 417
14. Nonparametric Models and k-Nearest Neighbours 423
15. k-Nearest Neighbours in Practice 429
16. The Curse of Dimensionality 435
17. Locally Weighted Regression and Support Vector Machines 441
18. Ensemble Learning: Bagging 447
19. Random Forests 453
20. Boosting, AdaBoost and Stacking 459
21. Developing a Machine Learning System 465
22. What Machine Learning Is 471
23. Why Use Machine Learning 477
24. Types of Machine Learning System 482
25. Supervised and Unsupervised Learning 487
26. Semi-Supervised, Self-Supervised and Reinforcement Learning 492
27. Batch and Online Learning 497
28. Instance-Based and Model-Based Learning 503
29. The Main Challenges of Machine Learning 509
30. Insufficient Quantity of Training Data 514
31. Nonrepresentative Training Data 519
32. Poor-Quality Data 525
33. Irrelevant Features and Feature Engineering 531
34. Overfitting the Training Data 537
35. Underfitting the Training Data 543
36. Stepping Back: The Landscape in One Page 548
37. Testing and Validating 554
38. MU's Exercises on the Machine Learning Landscape, Answered 560
39. MNIST 565
40. Training a Binary Classifier 571
41. Performance Measures, and Why Accuracy Misleads 576
42. Measuring Accuracy Using Cross-Validation 582
43. The Confusion Matrix 586
44. Precision, Recall and the F1 Score 591
45. The Precision and Recall Trade-Off 596
46. The ROC Curve and the Area Under It 601
47. Multiclass Classification 607
48. Error Analysis 612
49. Multilabel Classification 618
50. Multioutput Classification 623
51. Training and Visualising a Decision Tree 628
52. Making Predictions with a Trained Tree 633
53. Estimating Class Probabilities 639
54. The CART Training Algorithm 644
55. The Computational Complexity of a Decision Tree 650
56. Gini Impurity or Entropy 656
57. Regularisation Hyperparameters 662
58. Decision Tree Regression 667
59. Instability 672
60. MU's Exercises on Decision Trees, Answered 677
61. The Limits of AI 683
62. Can Machines Really Think? 689
63. The Ethics of AI 694
64. Bias and Fairness in a Model 700
65. Accountability, Transparency and Safety 706
66. The Feed-Forward Neural Network 712
67. Backpropagation 718
68. Data Preprocessing in a Pipeline 725
69. An End-to-End Pipeline 731
70. Inference Integration: Serving a Trained Model 736
71. Scalability, Robustness and Data Drift 741
72. A Bias and Class-Imbalance Audit 747
73. The Applications MU Names 752
74. The Frameworks and Tools, and How to Choose 757
75. Answering This Paper 762