Al-Azhar University 4th year

(Girls branch) )‫( رياضة بحته وعلوم الحاسب‬

Faculty of Science 18-June-2025 (2nd term)
Mathematics Time: 3 hrs.
Department Full mark: 80
Final examination in Artificial Intelligence (CS404)

Question 1: Choose the correct answer [30 marks] [1.5 marks/point]

1. What is the heuristic function of greedy best-first search?

a) F(n)! =h(n) b) f(n) < h(n) c) f(n) = h(n) d) f(n)=h(n) +g(n)
2. Which is used to improve the agents performance?
a) Perceiving b) Learning c) Observing d) None of the mentioned
3. Heuristic function h(n) is ____
a) Lowest path cost b) Estimated cost of cheapest path from root to goal node
c) Average path cost d) Cheapest path from root to goal node
4. It is a formal representation for world states.
a) Environment b) performance measure
c) Actions d) sensors
5. Choosing actions based on maximizing overall satisfaction or utility, rather than just achieving
goals
a) Goal-based agent b) simple reflex agent
c) Utility-based agent d) Model-based agent
6. Dividing the input dataset into logical groups
a) Dimensionality reduction b) clustering
c) Anomaly detection d) supervised learning
7. Method of deletion that is suffered to the maximum loss of information, where rows containing
missing values are deleted.
a) Pairwise b) Filter data c) spectral similarity d) listwise
8. A random error or variance in a measured variable
a) Noise b) unknown data c) pairwise d) listwise
9. determines the reduction in entropy while transforming the dataset.
a) Missing value ratio b) Chi-square test
c) Information gain d) fisher score
10. It transforms the original feature values x into a scaled range typically between 0, 1.
a) Standard scaler b) Mean normalization c) MinMax scaler d) Robust scaler
11. The number of neurons in the output layer of a classifier depends on:
a) Number of hidden layers b) Number of output classes
c) Learning rate d) Number of training samples
12. What is ReLU’s main advantage over Sigmoid or Tanh?
a) Less memory b) Always differentiable
c) Prevents vanishing gradients d) Better accuracy for binary output
13. Increasing the number of hidden neurons in MLP usually:
a) Increases capacity to learn complex functions b) Reduces training time

1
Al-Azhar University 4th year
(Girls branch) )‫( رياضة بحته وعلوم الحاسب‬
Faculty of Science 18-June-2025 (2nd term)
Mathematics Time: 3 hrs.
Department Full mark: 80
Final examination in Artificial Intelligence (CS404)

c) Guarantees better accuracy d) Avoids overfitting

14. Which of the following is not a hyperparameter?
a) Learning rate b) Batch size c) Weight gradients d) Number of epochs
15. If an MLP has 3 layers (input, hidden, output), how many weight matrices are there?
a) 1 b) 2 c) 3 d) Depends on the activation
16. What does dropout do?
a) Randomly disables neurons during training b) Reduces input size
c) Prevents vanishing gradients d) Prunes network permanently
17. What is the output of the ReLU function for input -3?
a) -3 b) 0.5 c) 0 d) 1
18. What is epoch in training?
a) Time between batches b) One full pass over the dataset
c) Weight update per input d) A layer in the model
19. In MLP, weights are usually:
a) Random integers b) Random small floating values
c) All initialized to zero d) Extracted from data
20. What is the main goal of an optimization algorithm in deep learning?
a) Increase the number of neurons b) Minimize the loss function
c) Increase training time d) Add more hidden layers

Question 2: Mark true () or false (X) [10 marks] [1 mark/point]

1. The cost-based agent depends on the goal.

2. GBFS is complete and not optimal.
3. Supervised learning depends on labeled data to train the models.
4. Reinforcement learning algorithms continuously learns from the environment in an interactive
fashion.
5. Categorical features have a finite number of possibilities.
6. Anomaly is a pattern in the data that does not conform to the expected behavior.
7. Features should correlated with the target and among themselves.
8. If gradient descent is working properly, then j(ɵ) should decrease after every iteration.
9. The sigmoid activation function outputs values between 0 and 1.
10. The ReLU function outputs negative values for negative inputs.

Question 3: Write the scientific term [10 marks] [2 marks/point]

1.The process of removing detail from a given state representation of state space
2.Which search strategy is also called blind search

2
Al-Azhar University 4th year
(Girls branch) )‫( رياضة بحته وعلوم الحاسب‬
Faculty of Science 18-June-2025 (2nd term)
Mathematics Time: 3 hrs.
Department Full mark: 80
Final examination in Artificial Intelligence (CS404)

3.The algorithm applied on game tree for making decision of win/lose

4.The type of uninformed search algorithms that reduces search time by searching from both ends
(initial state and goal)
5.The component added to tree search to solve the loopy path problem and produced graph search
Question 4: Answer the following questions [30 marks] [3 marks/point]

1. What are the fields of AI does the computer need to pass the total Turing test?
2. Specify the task environment PEAS for Autonomous Taxi?
3. State the properties of Sudoku environment?
4. Mention the types of supervised Feature Selection?
5. Write the cost function for linear regression and write the gradient descent update rule
for logistic regression?
6. Given the input vector X = [1, 0.5], and weights W = [0.4, -0.7], and bias b = 0.1,
calculate the output of a neuron using sigmoid activation.
7. What is overfitting? How can it be reduced in MLPs?
8. Define an activation function. Give two examples and explain their purpose.
9. Considering the following search graph, find out the pruned nodes using the Alpha-Beta
Pruning strategy.

10. Use graph searches to find paths from S to G in the following graph

Good Luck
Prof. Abeer S. Desuky, Dr. Shereen Abdel-Aal, Dr. Esraa Alaa