Search Algorithms & Explainable AI

Question 1

What is the workflow for predicting the next word in text generation?

Answer 1

Tokenize text → convert to word indices → create sequences → train RNN model.

Question 2

What activation function is used in the output layer of the RNN for text generation?

Answer 2

Softmax.

Question 3

What is Actionability in AI?

Answer 3

Deep learning understands data; search algorithms act on it.

Question 4

What is the goal of search algorithms?

Answer 4

Plan a sequence of actions that takes us from a start state to a goal state.

Question 5

What is the Initial State in search algorithms?

Answer 5

Where the problem starts.

Question 6

What does State Space refer to?

Answer 6

All possible configurations.

Question 7

What are Actions in the context of search algorithms?

Answer 7

Choices you can make.

Question 8

What is a Goal Test?

Answer 8

Did we solve it?

Question 9

What is Path Cost?

Answer 9

Total cost of reaching a goal.

Question 10

What is a Search Tree?

Answer 10

Structure of all possible steps.

Question 11

What does Completeness mean in search algorithms?

Answer 11

Will it always find a solution if one exists?

Question 12

What is Optimality in search algorithms?

Answer 12

Will it find the best solution?

Question 13

What does Time Complexity measure?

Answer 13

How long does it take?

Question 14

What is Space Complexity?

Answer 14

How much memory does it use?

Question 15

What is Breadth-First Search (BFS)?

Answer 15

Explore level-by-level.

Question 16

Is Breadth-First Search (BFS) complete?

Answer 16

Yes.

Question 17

Is Breadth-First Search (BFS) optimal?

Answer 17

Yes, if all steps cost the same.

Question 18

What is Depth-First Search (DFS)?

Answer 18

Explore down a path, then backtrack.

Question 19

Is Depth-First Search (DFS) always complete?

Answer 19

No.

Question 20

Is Depth-First Search (DFS) optimal?

Answer 20

No.

Question 21

What does Uniform Cost Search (UCS) do?

Answer 21

Always picks the lowest-cost path next.

Question 22

Is Uniform Cost Search (UCS) complete?

Answer 22

Yes.

Question 23

Is Uniform Cost Search (UCS) optimal?

Answer 23

Yes.

Question 24

What is Greedy Search?

Answer 24

Uses h(x) (estimated distance to goal) and picks the closest-looking node.

Question 25

Is Greedy Search optimal?

Answer 25

No.

Question 26

Is Greedy Search always complete?

Answer 26

No.

Question 27

What does A* Search algorithm use?

Answer 27

f(x) = g(x) + h(x), where g(x) = cost from start and h(x) = estimated cost to goal.

Question 28

Is A* Search complete?

Answer 28

Yes.

Question 29

Is A* Search optimal?

Answer 29

Yes.

Question 30

What is the objective of the 8-Puzzle?

Answer 30

Slide tiles to reach the goal state.

Question 31

What is Monte Carlo Dropout?

Answer 31

Train model with dropout and keep dropout on at test time to run predictions multiple times.

Question 32

What does Explainable AI (XAI) do?

Answer 32

Makes AI decisions interpretable.

Question 33

What is Grad-CAM used for?

Answer 33

Highlights which image regions influenced the decision.

Question 34

What does Grad-CAM compute?

Answer 34

Gradient of predicted class with respect to the last convolutional layer.

Question 35

What are some applications of Grad-CAM?

Answer 35

* Medical imaging * Object detection * Debugging deep models

Question 36

What does AI stand for?

Answer 36

Systems that act intelligently.

Question 37

Define Machine Learning (ML).

Answer 37

Learn from data without being explicitly programmed.

Question 38

What is Deep Learning (DL)?

Answer 38

ML + Neural networks with many layers.

Question 39

What is Data Mining?

Answer 39

Finding patterns in large datasets.

Question 40

List common issues in data cleaning.

Answer 40

* Missing values * Duplicate data * Noise and outliers * Inconsistencies

Question 41

What are some fixes for data cleaning issues?

Answer 41

* Drop/merge/estimate missing or bad data * Visualize with summaries (histograms, boxplots, etc.)

Question 42

What is the k-Nearest Neighbours (k-NN) method?

Answer 42

Find k closest labelled data points and assign the most common class (majority vote).

Question 43

What is the formula for Linear Regression?

Answer 43

Y = b + mX.

Question 44

What is Gradient Descent used for in Linear Regression?

Answer 44

To minimize loss.

Question 45

Define Unsupervised Learning.

Answer 45

Learning from data without labeled responses.

Question 46

What is Partition-based clustering?

Answer 46

Group data into k clusters (e.g., k-means).

Question 47

What does Hierarchical clustering do?

Answer 47

Build a tree of nested clusters (dendrograms).

Question 48

What is Representation Learning?

Answer 48

Learns features automatically from raw data.

Question 49

What are Fully Connected Networks (FCN)?

Answer 49

A type of neural network architecture.

Question 50

What are CNNs used for?

Answer 50

Vision tasks (e.g., Faster R-CNN, U-Net, Mask R-CNN).

Question 51

What are RNNs/LSTMs primarily used for?

Answer 51

Language processing.

Question 52

What is Backpropagation?

Answer 52

Updates weights using error gradients.

Question 53

What does Batch Norm do?

Answer 53

Stabilizes training.

Question 54

What is Regularisation in the context of neural networks?

Answer 54

Prevents overfitting (e.g., Dropout).

Question 55

List types of Loss Functions.

Answer 55

* Binary Classification: binary_crossentropy * Multiclass: categorical_crossentropy * Regression: mean_squared_error

Question 56

What is Grid Search?

Answer 56

An exhaustive but slow optimization method.

Question 57

What is Random Search?

Answer 57

An efficient optimization method for large spaces.

Question 58

What is Faster R-CNN used for?

Answer 58

Object detection.

Question 59

What is U-Net used for?

Answer 59

Semantic segmentation.

Question 60

What does Mask R-CNN do?

Answer 60

Instance segmentation (detect + pixel mask).

Question 61

What does Bag of Words (BoW) do?

Answer 61

Counts words, ignores order.

Question 62

What does TF-IDF stand for?

Answer 62

Term Frequency–Inverse Document Frequency.

Question 63

What is an N-Gram?

Answer 63

Predicts next word using previous N-1 words.

Question 64

True or False: N-Grams can handle long-term dependencies.

Answer 64

False.

Question 65

What can be used to handle long-term dependencies in NLP?

Answer 65

RNNs / LSTMs.

Question 66

What does BFS stand for in search algorithms?

Answer 66

Breadth-First Search.

Question 67

Is BFS complete and optimal?

Answer 67

Yes (equal cost only).

Question 68

Is DFS complete and optimal?

Answer 68

No.

Question 69

What does UCS stand for?

Answer 69

Uniform Cost Search.

Question 70

Is UCS complete and optimal?

Answer 70

Yes.

Question 71

What is the A* algorithm known for?

Answer 71

Best combo of cost + heuristic.

Question 72

What is Dropout Monte Carlo used for?

Answer 72

Keep dropout ON at test time to estimate uncertainty.

Question 73

What does Grad-CAM do?

Answer 73

Highlights image regions used in CNN decisions.

Question 74

What should you understand for exam prep regarding architectures?

Answer 74

Each architecture’s use case.

Question 75

What should you compare for exam prep in NLP?

Answer 75

BoW vs TF-IDF vs Word Embeddings.

Question 76

What metrics should you practice for ML evaluation?

Answer 76

Confusion matrices, F1, AUC, etc.