BE Computer Engineering (IOE, TU) Artificial Intelligence (IOE, CT 653) Question Paper 2078

Define an intelligent agent and explain the structure of a rational agent in terms of the PEAS (Performance measure, Environment, Actuators, Sensors) framework.

(a) Describe, with a suitable diagram, the working of a model-based reflex agent and a utility-based agent, and clearly distinguish between them. (8)

(b) Classify the task environment of an automated taxi driver agent along the dimensions: fully/partially observable, deterministic/stochastic, episodic/sequential, static/dynamic, and discrete/continuous. Justify each classification. (4)

Consider the following graph where each node represents a city and edge labels denote actual path costs. The straight-line distance heuristic h(n) to the goal G is given as: h(A)=10, h(B)=6, h(C)=4, h(D)=7, h(G)=0. Edges: A-B = 3, A-D = 5, B-C = 4, C-G = 5, D-G = 9.

(a) State the conditions for a heuristic to be admissible and consistent. Verify whether the given heuristic h(n) is admissible. (5)

(b) Apply the A* search algorithm to find the optimal path from A to G. Show the contents of the open and closed lists, and the f(n) = g(n) + h(n) values at each step. (8)

(c) Why is A* search guaranteed to be optimal when the heuristic is admissible? Explain briefly. (3)

(a) Represent the following statements in First Order Predicate Logic: "All students who study hard pass the exam. Ram is a student. Ram studies hard." Hence prove that "Ram passes the exam" using resolution refutation. Show the conversion to clausal (CNF) form and the unification steps. (8)

(b) Explain the role of unification and the Most General Unifier (MGU) in resolution-based inference, with an example. **(4)

(a) Draw the architecture of a feed-forward multilayer perceptron and explain how the backpropagation algorithm is used to train it. Derive the weight update rule for an output-layer weight using gradient descent. (8)

(b) Explain the role of an activation function in a neural network. Compare the sigmoid and ReLU activation functions, and state one advantage of each. **(4)

Compare Breadth-First Search (BFS) and Depth-First Search (DFS) on the basis of completeness, optimality, time complexity, and space complexity. Under what circumstances would you prefer Iterative Deepening Depth-First Search (IDDFS) over both, and why?

Explain semantic networks and frames as knowledge representation schemes. Represent the fact "A sparrow is a bird that can fly and has wings; Tweety is a sparrow" using a semantic network, and show how property inheritance applies to Tweety.

(a) Draw and explain the basic architecture of an expert system, clearly describing the function of the knowledge base and the inference engine. (5)

(b) Differentiate between forward chaining and backward chaining inference with a suitable example. **(3)

What is a heuristic function? Explain the working of hill-climbing search and discuss the problems of local maxima, plateau, and ridges. State one technique to overcome these problems.

Differentiate between supervised, unsupervised, and reinforcement learning with one application example of each. How does the problem of overfitting arise in supervised learning, and how can it be reduced?

List and briefly explain the major stages of Natural Language Processing (NLP): lexical (morphological) analysis, syntactic analysis, semantic analysis, and pragmatic analysis. Give one example illustrating ambiguity handled at the syntactic level.

Explain the Minimax algorithm for two-player games using a suitable game tree. How does alpha-beta pruning improve its efficiency? Illustrate which branches are pruned in a small example tree.

Convert the following First Order Logic sentences into Conjunctive Normal Form (CNF), showing the steps of eliminating implications, moving negations inward, skolemization, and dropping universal quantifiers:

(i) ∀x (Person(x) → ∃y (Loves(x, y)))

(ii) ∀x (Bird(x) ∧ ¬Penguin(x) → CanFly(x))