CS663/362 Artificial Intelligence

Autumn 2024-25

General Instructions for Labs

Students enrolled in the course are expected to work on programming problems assigned for
laboratories in groups. Groups once formed would remain fixed for the entire duration of the
course. Evaluation of lab assignments should primarily be taken as a two-way feedback, both for
the students and the instructor.

A group can have a maximum of 4 students from the same section.

Attending laboratory sessions is a must and absence during laboratory sessions without prior
intimation to the course instructor (only genuine cases) will attract 0 credit for that particular
session.

Feel free to contact me or the TAs in case you have any query or confusion regarding
submissions.

Course Instructor:
Pratik Shah
pratik@iiitvadodara.ac.in

Evaluation Policy

Component 1: 35%
A. Attendance and participation - 15%
B. Peer Assessment - 20%
a. Four student groups will be identified per week to solve, discuss and explain the
respective laboratory problem for the coming week to the respective sections.
b. Each group will have 45 minutes of pre-assigned slots during the laboratory
hours. Each group will explain their solution with the codes within the first 30
minutes followed by 15 minutes of discussion/ question answer on the solution/
codes with the section.
c. The peers/ students present during the lab will assess the performance of each
group by assigning the score on the scale of 0-5 for the following
i. Quality of explanation and clarity of exposition and discussion
ii. Quality of the code base
iii. Resolution of queries/ questions raised by peers during the session.
Component 2: 50%
Students are required to work in groups on laboratory problems. Each group is expected to
submit a total of two reports containing the discussion on the solution to five-five problems each.
One such report will be due just before the mid-semester examination and the other just before
the end-semester examination.

1
Report/ Term Paper Template : IEEE Transactions on Systems Man and Cybernetics: Systems in
LaTeX format. https://drive.google.com/file/d/1sdEyE-9DxDNFbuR_n3xJ0u0Og6aw9zKU/
view?usp=sharing

It is expected that you will write the report as if you are writing a tutorial to explain the problems
and the corresponding solutions.

Component 3: 15%
A project/ lab viva voce will be conducted immediately before/ after the mid/ end-semester
examination.

Moral Code of Conduct

Zero tolerance to plagiarism. If the report submitted by any group is found to have copied
extracts from any of the other submitted reports the group members will receive no credit for the
entire report. In a report if some material is found to have been reproduced from some source
without proper citation or reference, the report will attract zero credits.

2
Week 0 :
Lab Assignment - 0

Learning Objective: To be able to model a given problem in terms of state space search problem
and solve the same using BFS/ DFS

Reference:
[1] Artificial Intelligence: a Modern Approach, Russell and Norvig (Fourth edition)
Chapter 1, 2, 3

Problem Statement:
A. In the rabbit leap problem, three east-bound rabbits stand in a line blocked by three west-
bound rabbits. They are crossing a stream with stones placed in the east west direction in
a line. There is one empty stone between them. The rabbits can only move forward one or
two steps. They can jump over one rabbit if the need arises, but not more than that. Are
they smart enough to cross each other without having to step into the water?
B. The missionaries and cannibals problem is usually stated as follows. Three missionaries
and three cannibals are on one side of a river, along with a boat that can hold one or two
people. Find a way to get everyone to the other side without ever leaving a group of
missionaries in one place outnumbered by the cannibals in that place. This problem is
famous in AI because it was the subject of the first paper that approached problem-
formulation from an analytical viewpoint.

For the above two problems,

1. Model the problem as a state space search problem. How large is the search space?
2. Solve the problem using BFS. The optimal solution is the one with the fewest number of
steps. Is the solution that you have acquired an optimal one? The program should print
out the solution by listing a sequence of steps needed to reach the goal state from the
initial state.
3. Solve the problem using DFS. The program should print out the solution by listing a
sequence of steps needed to reach the goal state from the initial state
4. Compare solutions found from BFS and DFS. Comment on solutions. Also compare the
time and space complexities of both.

3
Week 1 :
Lab Assignment - 1

Learning Objective: To design a graph search agent and understand the use of a hash table,
queue in state space search.

Reference:
[1] Artificial Intelligence: a Modern Approach, Russell and Norvig (Fourth edition)
Chapter 2 and 3
[2] A first course in Artificial Intelligence, Deepak Khemani
Chapter 2

Problem Statement:

A. Write a pseudocode for a graph search agent. Represent the agent in the form of a flow
chart. Clearly mention all the implementation details with reasons.
B. Write a collection of functions imitating the environment for Puzzle-8.
C. Describe what is Iterative Deepening Search.
D. Considering the cost associated with every move to be the same (uniform cost), write a
function which can backtrack and produce the path taken to reach the goal state from the
source/ initial state.
E. Generate Puzzle-8 instances with the goal state at depth “d”.
F. Prepare a table indicating the memory and time requirements to solve Puzzle-8 instances
(depth “d”) using your graph search agent.

4
Week 2 :
Lab Assignment - 2

Learning Objective: To understand the use of Heuristic function for reducing the size of the
search space. Explore non-classical search algorithms for large
problems.

Reference:
[1] Artificial Intelligence: a Modern Approach, Russell and
Norvig (Fourth edition)
Chapter 3 and 4
[2] A first course in Artificial Intelligence, Deepak Khemani
Chapter 3, 4

Problem Statement:
A. Read about the game of marble solitaire. Figure shows the initial board configuration.
The goal is to reach the board configuration where only one marble is left at the centre.
To solve marble solitaire, (1) Implement priority queue based search considering path
cost, (2) suggest two different heuristic functions with justification, (3) Implement best
first search algorithm, (4) Implement A*, (5) Compare the results of various search
algorithms.
B. Write a program to randomly generate k-SAT problems. The program must accept
values for k, m the number of clauses in the formula, and n the number of variables.
Each clause of length k must contain distinct variables or their negation. Instances
generated by this algorithm belong to fixed clause length models of SAT and are known
as uniform random k-SAT problems.
C. Write programs to solve a set of uniform random 3-SAT problems for different
combinations of m and n, and compare their performance. Try the Hill-Climbing, Beam-
Search with beam widths 3 and 4, Variable-Neighborhood-Descent with 3 neighborhood
functions. Use two different heuristic functions and compare them with respect to
penetrance.

5
Week 3 :
Lab Assignment 3

Learning Objective:
Non-deterministic Search | Simulated Annealing
For problems with large search spaces, randomized search becomes a meaningful option given
partial/ full-information about the domain.

Reference:
[1] A first course in Artificial Intelligence, Deepak Khemani (Chapter 4)

Problem:
Travelling Salesman Problem (TSP) is a hard problem, and is simple to state. Given a graph in
which the nodes are locations of cities, and edges are labelled with the cost of travelling between
cities, find a cycle containing each city exactly once, such that the total cost of the tour is as low
as possible.

For the state of Rajasthan, find out at least twenty important tourist locations. Suppose your
relatives are about to visit you next week. Use Simulated Annealing to plan a cost effective tour
of Rajasthan. It is reasonable to assume that the cost of travelling between two locations is
proportional to the distance between them.

An interesting problem domain with TSP instances:

VLSI: http://www.math.uwaterloo.ca/tsp/vlsi/index.html#XQF131
(Attempt at least five problems from the above list and compare your results.)

6
Week 4 :

Week 1-3 : Challenge Problem

Learning Objective:
Non-deterministic Search | Simulated Annealing | Genetic Algorithm
For problems with large search spaces, randomized search becomes a meaningful option given
partial/ full-information about the domain.

Reference:
[1] Simulated annealing from basics to applications, Daniel Delahaye, Supatcha Chaimatanan,
Marcel Mongeau
[2] A first course in Artificial Intelligence, Deepak Khemani (Chapter 4)
[3] Artificial Intelligence: a Modern Approach, Russell and Norvig (Fourth edition)
Chapter 3 and 4

Problem:

Ever played jigsaw puzzles in childhood? Time

to make an agent learn to solve the puzzle.

Some rough, confusing code snippets are posted

on the codes folder of the course along with the
scrambled_lena.mat

Formulate the problem as a state space search

problem. Try to solve the same using simulated
annealing.

Have fun!

This is a bonus challenge! If you solve this, you are up for some extra credits.

7
Week 5 :
Lab Assignment 4

Learning Objective:
Game Playing Agent | Minimax | Alpha-Beta Pruning

Systematic adversarial search can lead to savings in terms of pruning of sub-trees resulting in
lesser node evaluations

References:
[1] A first course in Artificial Intelligence, Deepak Khemani (Chapter 8)
[2] Artificial Intelligence: a Modern Approach, Russell and Norvig (Fourth edition)
(Chapter 5)

Problem:

1. What is the size of the game tree for Noughts

and Crosses? Sketch the game tree.
2. Read about the game of Nim (a player left
with no move losing the game). For the
initial configuration of the game with three
piles of objects as shown in Figure, show that
regardless of the strategy of player-1, player- 2
will always win. Try to explain the reason
with the MINIMAX value backup argument
on the game tree.
3. Implement MINIMAX and alpha-beta
pruning agents. Report on number of evaluated nodes for Noughts and Crosses game tree.
4. Use recurrence to show that under perfect ordering of leaf nodes, the alpha-beta pruning
time complexity is O(bm/2), where b is the effective branching factor and m is the depth of
the tree.

8
Week 6 :
Lab Assignment 5

Learning Objective:
Understand the graphical models for inference under uncertainty, build Bayesian Network in R,
Learn the structure and CPTs from Data, naive Bayes classification with dependency between
features.

Reference:
1. https://www.bnlearn.com/
2. http://gauss.inf.um.es/umur/xjurponencias/talleres/J3.pdf

Problem Statement:
A table containing grades earned by students in respective courses is made available to you in
(codes folder) 2020_bn_nb_data.txt.
1. Consider grades earned in each of the courses as random variables and learn the
dependencies between courses.
2. Using the data, learn the CPTs for each course node.
3. What grade will a student get in PH100 if he earns DD in EC100, CC in IT101 and CD in
MA101.
4. The last column in the data file indicates whether a student qualifies for an internship
program or not. From the given data, take 70 percent data for training and build a naive
Bayes classifier (considering that the grades earned in different courses are independent
of each other) which takes in the student’s performance and returns the qualification
status with a probability. Test your classifier on the remaining 30 percent data. Repeat
this experiment for 20 random selection of training and testing data. Report results about
the accuracy of your classifier.
5. Repeat 4, considering that the grades earned in different courses may be dependent.

9
Week 7 :
Lab Assignment 6

Learning Objective:
To implement Expectation Maximization routine for learning parameters of a Hidden Markov
Model, to be able to use the EM framework for deriving algorithms for problems with hidden or
partial information.

Reference:
1. A Revealing Introduction to Hidden Markov Models, Mark Stamp, 2018
2. What is the expectation maximization algorithm? Chuong B Do and Serafim Batzoglou,
Nature Biotechnology, Vol 26, Num 8, August 2008

Problem Statement:
A. Read through the reference carefully. Implement routines for learning the parameters of
HMM given in section 7. In section 8, “A not-so-simple example”, an interesting exercise
is carried out. Perform a similar experiment on “War and Peace” by Leo Tolstoy.
B. Ten bent (biased) coins are placed in a box with unknown bias values. A coin is
randomly picked from the box and tossed 100 times. A file containing results of five
hundred such instances is presented in tabular form with 1 indicating head and 0
indicating tail. Find out the unknown bias values. (2020_ten_bent_coins.csv) To help
you, a sample code for two bent coin problem along with data is made available in the
work folder: two_bent_coins.csv and embentcoinsol.m
C. A point set with real values is given in 2020_em_clustering.csv. Considering that there
are two clusters, use EM to group together points belonging to the same cluster. Try and
argue that k-means is an EM algorithm.

10
Week 8 :
Lab Assignment 7

Learning Objective:
Construction and evaluation of identification (decision) tree classifier

Problem:
A car dataset is uploaded in the lab-work folder with description specifying all the attributes, the
values that each attribute can take and the class labels.
1. Randomly select 60 percent of labeled data (from each class) for constructing the tree
(training). Test for the rest of 40 percent data. Find out the accuracy of the classification
tree with the help of confusion matrix and F-score. Use the entropy measure for selection
of attributes.
2. Repeat the above exercise 20 times. Calculate the average accuracy of classification.
3. Repeat steps 1 and 2 with Gini index as a measure for selection of attributes.
4. Repeat the steps 1, 2 and 3 considering 70 and 80 percent data (random selection) for
training.
5. Describe the problem of overfitting in your words with an example created from the data-
set.

Note: You may use already available routines like ID3 for this exercise.

Read about random forest classifiers. I am sure you will be happy to know that a random Forest
is a collection of trees!

Lab Assignment for Self Learning

Learning Objective:
To model the low level image processing tasks in the framework of Markov Random Field and
Conditional Random Field. To understand the working of Hopfield network and use it for
solving some interesting combinatorial problems

Reference:
1. http://www.cs.utoronto.ca/~strider/Denoise/Benchmark/
2. Interesting way to denoise an image using random walk:
http://www.cs.toronto.edu/~fleet/research/Papers/BMVC_denoise.pdf
3. MRF Image Denoising:
https://web.cs.hacettepe.edu.tr/~erkut/bil717.s12/w11a-mrf.pdf
4. Single Neuron and Hopfield Network: Chapter 40, 41, 42
Information Theory, Inference and Learning Algorithms, David MacKay

11
 http://www.inference.phy.cam.ac.uk/mackay/itila/

Problem Statement:
A. Many low level vision and image processing problems are posed as minimization of
energy function defined over a rectangular grid of pixels. We have seen one such
problem, image segmentation, in class. The objective of image denoising is to recover an
original image from a given noisy image, sometimes with missing pixels also. MRF
models denoising as a probabilistic inference task. Since we are conditioning the original
pixel intensities with respect to the observed noisy pixel intensities, it usually is referred
to as a conditional Markov random field. Refer to (3) above. It describes the energy
function based on data and prior (smoothness). Use quadratic potentials for both
singleton and pairwise potentials. Assume that there are no missing pixels. Cameraman is
a standard test image for benchmarking denoising algorithms. Add varying amounts of
Gaussian noise to the image for testing the MRF based denoising approach. Since the
energy function is quadratic, it is possible to find the minima by simple gradient descent.
If the image size is small (100x100) you may use any iterative method for solving the
system of linear equations that you arrive at by equating the gradient to zero. Extra
Credit Challenge: Implement and compare MRF denoising with Stochastic denoising
(reference 2).
B. For the sample code hopfield.m supplied in the lab-work folder, find out the amount of
error (in bits) tolerable for each of the stored patterns.
C. Solve a TSP (traveling salesman problem) of 10 cities with a Hopfield network. How
many weights do you need for the network?

12
Week 9 :
Lab Assignment 8

Learning Objective: To understand the working of Hopfield network and use it for solving some
interesting combinatorial problems

Problems:
1. For the sample code supplied in the lab-work folder, find out the amount of error (in bits)
tolerable for each of the stored patterns.
2. Setup the energy function for the Eight-rook problem and solve the same using Hopfield
network. Give reasons for choosing specific weights for the network.
3. Solve a TSP (traveling salesman problem) of 10 cities with a Hopfield network. How
many weights do you need for the network?

Lab Assignment for Self Learning

Learning objective:
Basics of data structure needed for state-space search tasks and use of random numbers required
for MDP and RL

Title: Matchbox Educable Naughts and Crosses Engine

Reference: http://people.csail.mit.edu/brooks/idocs/matchbox.pdf

Problem Statement:
Read the reference on MENACE by Michie and check for its implementations. Pick the one that
you like the most and go through the code carefully. Highlight the parts that you feel are crucial.
If possible, try to code the MENACE in any programming language of your liking.

13
Week 10 :
Lab Assignment 9

Learning objective:
Understanding Exploitation - Exploration in simple n-arm bandit reinforcement learning task,
epsilon-greedy algorithm

Title: n-armed bandit

Reference: Reinforcement Learning: an introduction by R Sutton and A Barto (Second Edition)
(Chapter 1-2)

Problem Statement:
(1) Consider a binary bandit with two rewards {1-success, 0-failure}. The bandit returns 1 or
0 for the action that you select, i.e. 1 or 2. The rewards are stochastic (but stationary).
Use an epsilon-greedy algorithm discussed in class and decide upon the action to take for
maximizing the expected reward. There are two binary bandits named binaryBanditA.m
and binaryBanditB.m are waiting for you.

(2) Develop a 10-armed bandit in which all ten mean-rewards start out equal and then take
independent random walks (by adding a normally distributed increment with mean zero
and standard deviation 0.01 to all mean-rewards on each time step).
{function [value] = bandit_nonstat(action)}

(3) The 10-armed bandit that you developed (bandit_nonstat) is difficult to crack with a
standard epsilon-greedy algorithm since the rewards are non-stationary. We did discuss
how to track non-stationary rewards in class. Write a modified epsilon-greedy agent and
show whether it is able to latch onto correct actions or not. (Try at least 10000 time steps
before commenting on results)

Credits: Be as descriptive as possible while coding. Provide substantial evidence in the form of
graphs, analysis etc. supporting your claim. Prepare a small report on the assignment (preferably
in tex).

14
Week 11:
Lab Assignment 10

Learning Objective: Understand the process of sequential decision making (stochastic

environment) and the connection with reinforcement learning

Title: Markov Decision Process and Dynamic Programming

Reference:
[1] Artificial Intelligence a Modern Approach, Russell and Norvig (third edition)
Chapter 16, 17, 21
[2] Reinforcement Learning, Sutton and Barto (second edition)
Chapter 3, 4

Problem Statement:
(1) Suppose that an agent is situated in the 4x3 environment as shown in Figure 1.
Beginning in the start state, it must choose an action at each time step. The interaction
with the environment terminates when the agent reaches one of the goal states, marked
+1 or -1. We assume that the environment is fully observable, so that the agent always
knows where it is. You may decide to take the following four actions in every state: Up,
Down, Left and Right. However, the environment is stochastic, that means the action
that you take may not lead you to the desired state. Each action achieves the intended
effect with probability
0.8, but the rest of the
time, the action moves
the agent at right
angles to the intended
direction with equal
probabilities.
Furthermore, if the
agent bumps into a
wall, it stays in the
same square. The
+
immediate reward for moving to any state (s) except for the terminal states S is r(s)= -
0.04. And the reward for moving to terminal states is +1 and -1 respectively. Find the
value function corresponding to the optimal policy using value iteration.

Find the value functions corresponding optimal policy for the following:
(a) r(s)=-2
(b) r(s)=0.1
(c) r(s)=0.02
(d) r(s)=1

15
 (2) [Gbike bicycle rental] You are managing two locations for Gbike. Each day, a number of
customers arrive at each location to rent bicycles. If you have a bike available, you rent it
out and earn INR 10 from Gbike. If you are out of bikes at that location, then the
business is lost. Bikes become available for renting the day after they are returned. To
help ensure that bicycles are available where they are needed, you can move them
between the two locations overnight, at a cost of INR 2 per bike moved.

Assumptions: Assume that the number of bikes requested and returned at each location
are Poisson random variables. Expected numbers of rental requests are 3 and 4 and
returns are 3 and 2 at the first and second locations respectively. No more than 20 bikes
can be parked at either of the locations. You may move a maximum of 5 bikes from one
location to the other in one night. Consider the discount rate to be 0.9.

Formulate the continuing finite MDP, where time steps are days, the state is the number
of bikes at each location at the end of the day, and the actions are the net number of bikes
moved between the two locations overnight.

Download and extract files from gbike.zip. Try to compare your formulation with the
code. Before proceeding further, ensure that you understand the policy iteration clearly.

(3) Write a program for policy iteration and resolve the Gbike bicycle rental problem with
the following changes. One of your employees at the first location rides a bus home each
night and lives near the second location. She is happy to shuttle one bike to the second
location for free. Each additional bike still costs INR 2, as do all bikes moved in the
other direction. In addition, you have limited parking space at each location. If more
than 10 bikes are kept overnight at a location (after any moving of cars), then an
additional cost of INR 4 must be incurred to use a second parking lot (independent of
how many cars are kept there).

Note: Problem (2) and (3) are directly taken from the book by Sutton and Barto. The problem if
attempted independently (without referring to the solutions available) is a good exercise in
understanding the implementation issues and finite MDP. Changing the Jack’s car rental to
Gbike bicycle rental is just to put things in context. The instructor has no intention of earning
credit for posing the problem. Credit goes to Sutton and Barto.

16
Week 13:
Lab Assignment 11

Objective: to understand the design of type-1 (Mamdani) Fuzzy expert system

Problem:

We all have seen, at least in advertisements, washing machines work. Some of us have used
them also. Don’t feel surprised if I tell you that it was one of the initial devices to have used
fuzzy logic.

In this laboratory, we will try to identify the appropriate time needed to wash the load of clothes,
given the dirtiness and the volume of the load. This is a simplified design, in practice a lot more
variables are involved including: water level, amount of detergent to be dispensed, and
temperature of water (recent development) and variables that you and I may imagine in future.

A simple rule-base for the washing machine time problem is given below for reference.

Load Dirtiness/ vd md ld nd
Load Volume

fl vlot vlot lot lit

ml vlot mt mt lit

ll lot lot lit lit

vd: very dirty, md: medium dirty, ld: lightly dirty, nd: not dirty
fl: full load, ml: medium load, ll: low load
vlot : very long time, lot: long time, mt: medium time, lit: little time

The rule table consists of 12 rules. For example, the entry in the second row and the third
column of the table specifies the rule:

If load volume is medium load

And load dirtiness is lightly dirty
Then washing time is medium time

17