UNIVERSITY OF GHANA (All rights reserved)

Transcription

1 UNIVERSITY OF GHANA (All rights reserved) Faculty of Engineering Sciences Department of Computer Engineering First Semester 2011/2012 Academic Year CENG 409: Artificial Intelligence Final Exams (3 credits) Time: 3hours INSTRUCTION: ANSWER ALL QUESTIONS, Answer section A on the question paper and all other sections in the answer booklet. STUDENT ID#... SECTION A [15 MARKS] 1. Which of the following is true? (i) On average, neural networks have higher computational rates than conventional computers. (ii) Neural networks learn by example. (iii) Neural networks mimic the way the human brain works. (a) all of them are true (b) (ii) and (iii) are true (c) (i), (ii) and (iii) are true 2. Which of the following is true for neural networks? (i) The training time depends on the size of the network. (ii) Neural networks can be simulated on a conventional computer. (iii)artificial neurons are identical in operation to biological ones. (a) all of them are true. (b) (ii) is true. (c) (i) and (ii) are true. 3. What are the advantages of neural networks over conventional computers? (i) They have the ability to learn by example (ii) They are more fault tolerant (iii)they are more suited for real time operation due to their high 'computational' rates Page 1 of 6 Examiner: Robert Adjetey Sowah, PhD

2 (a) (i) and (ii) are true (b) (i) and (iii) are true (c) all of them are true 4. Which of the following is true? Single layer associative neural networks do not have the ability to: (i) perform pattern recognition (ii) find the parity of a picture (iii)determine whether two or more shapes in a picture are connected or not (a) (ii) and (iii) are true (b) (ii) is true (c) all of them are true 5. What is the role of the hidden layers in a neural network? What is the most popular algorithm for training a neural network? 7. Explain briefly the principle of this algorithm? Briefly enumerate and explain five (5) types of problems that can be solved with neural networks? Page 2 of 6 Examiner: Robert Adjetey Sowah, PhD

3 SECTION B (5 marks) (Introduction to Artificial Intelligence) 1. Define intelligence. [1 mark] 2. What are the different approaches in defining artificial intelligence? [2marks] 3. Suppose you design a machine to pass the Turing test. What are the capabilities such a machine must have? [2marks] 4. Discuss briefly the key salient features of bounded rationality? [2marks] 5. Based on your research findings in the homeworks about the Mars rover. [2marks] a. What are the percepts for this agent? b. Characterize the operating environment. c. What are the actions the agent can take? d. What sort of agent architecture do you think is most suitable for this agent? 6. What is learning in artificial agents? Enumerate the four (4) types of learning. [3marks] 7. With the aid of a diagram describe how an agent interacting with its environment learns. [3marks] 8. An agent is placed in an environment containing two slot machines, each of which costs $1 to play. The expected payoffs are p and q and the agent can play each machine as many times as it likes. Describe, in qualitative terms, how a rational agent should behave in each of the following cases: (a) p = 1.5 and q = 0.8. (b) p = 1.5 and q is unknown. (Hint: Remember your PEAS!) [4marks] 9. Give the initial state, goal test, successor function, and cost function for each of the following. Choose a formulation that is precise enough to be implemented. (a) A 3-foot-tall monkey is in a room where some bananas are suspended from the 8-foot ceiling. He would like to get the bananas. The room contains two stackable, movable, climbable 3-foot high crates. (b) You have a program that outputs the message ``illegal input line'' when fed a certain input file. You want to determine what line in the input file is causing the problem. (c) You have three jugs measuring 12 gallons, 8 gallons, and 3 gallons, and a water faucet. You need to measure out exactly one gallon. [6 marks] SECTION C (15 marks) (Neural Networks and Genetic Algorithms) Neural Networks 1. Describe a neural network that could be used to control the steering of a robotic car on a race track. How can it be trained? 2. A surveillance camera is used to observe the main hall of an official building. From the position of visitors in the camera frame, we want to be able to compute their position on the building map. Describe a neural network that could be used to achieve this task. How can the system be calibrated? 3. A mobile robot is equipped with an omni-directional range sensor and bumpers. We want to make it learn an obstacle avoidance behaviour. Describe how this can be addressed with reinforcement learning. Page 3 of 6 Examiner: Robert Adjetey Sowah, PhD

4 Genetic Algorithms 1. Explain the effect of selection, crossover and mutation in evolutionary computation. a. How is the population affected by the use of each one of these operators? 2. Consider the problem below and discuss how you would find a solution using evolutionary computation. Discuss the representation of the chromosome, the fitness function (objective function), and whether any special mutation and cross-over operators would be required. Justify your approach. You are given a list of 100 items, each with a weight and a utility value. The problem is to select an optimal set of items from the list, up to 6 items total, such that the weight is less than 20 pounds, and the utility is maximized. SECTION D (25 MARKS) (Search Problem Formulation and Solution) 2. Consider the tree shown below. The numbers on the arcs are the arc lengths. Assume that the nodes are expanded in alphabetical order when no other order is specified by the search, and that the goal is state G. No visited or expanded lists are used. What order would the states be expanded by each type of search? Stop when you expand G. Write only the sequence of states expanded by each search. Search Type Breadth First Depth First Progressive Deepening Search Uniform Cost Search List of states Page 4 of 6 Examiner: Robert Adjetey Sowah, PhD

5 3. Consider the graph shown below where the numbers on the links are link costs and the numbers next to the states are heuristic estimates. Note that the arcs are undirected. Let A be the start state and G be the goal state. Simulate A* search with a strict expanded list on this graph. At each step, show the path to the state of the node that s being expanded, the length of that path, the total estimated cost of the path (actual + heuristic), and the current value of the expanded list (as a list of states). You are welcome to use scratch paper or the back of the exam pages to simulate the search. However, please transcribe (only) the information requested into the table given below. Path to State Expanded Length of Path Total Estimated Cost Expanded List A 0 5 (A) 4. Consider the problem of constructing crossword puzzles: fitting words into a grid of intersecting horizontal and vertical squares. Assume that a list of words (i.e., a dictionary) is provided, and that the task is to fill in the squares using any subset of this list. Go through a complete goal and problem formulation for this domain, and choose a search strategy to solve it. Specify the heuristic function, if you think one is needed. Page 5 of 6 Examiner: Robert Adjetey Sowah, PhD

6 SECTION E (10 MARKS) (First Order Logic and Knowledge Representation) The process of writing down logical descriptions of situation and events enables formal conclusions to be drawn based on the statement of facts. Consider the case of knowing that the following statements are true: Does this mean that the lecture will be good? If today is sunny, Tomas will be happy; If Tomas is happy, the lecture will be good; And today is sunny. 1. How many variables are inherent in the above problem statement? 2. How many possible in total are valid for the above number of variables? 3. Write the set of interpretations that make the original sentences true. 4. Construct a truth table and check the interpretation for the sentences SECTION F (15 MARKS) (Decision Trees) The following is a modification of one of the examples presented in class. There are two candidate cars C1 and C2. Each can either be of good quality or bad quality. There are two possible tests. T1 on C1 costs $100. T2 on C2 costs $50. C1 costs $1500, which is $500 below market value. If C1 is of bad quality, the repair cost is $700 C2 costs $1150, which is $250 below market value. If C2 is of bad quality, the repair costs $150. The chance that car C1 is of good quality is 70%. The chance that car C2 is of good quality is 80%. Test T1 on C1 will confirm good quality with probability 80% and bad quality with probability 65%. Test T2 on C2 will confirm good quality with probability 75% and bad quality with probability 70%.The buyer must buy one of the two cars and can perform T1, T2, T1 and T2 simultaneously, T1 followed by T2, or T2 followed by T1. The total cost of performing the two tests simultaneously is $125 (this includes the cost of the T1 and T2) and the total cost of performing one test followed by the other is $175 (again including the cost of performing T1 and T2). 1) Build a decision tree for this problem. Only draw the branch that corresponds to T1 being performed first. Note: You do not have to repeatedly draw similar branches. Just draw a representative branch and explain what the other branches might be that you did not end up drawing. Page 6 of 6 Examiner: Robert Adjetey Sowah, PhD

7

8

9

10

11

12