COMP90049 Introduction to Machine Learning, Sample Exam
The University of Melbourne
Department of Computing and Information Systems
COMP90049
Introduction to Machine Learning
June 2023
Identical examination papers: None
Exam duration: 120 minutes
Reading time: Fifteen minutes
Length: This paper has 6 pages including this cover page.
Authorised materials: Lecture slides, workshop materials, prescribed reading, your own project re-
ports.
Calculators: Permitted
Instructions to students: The total marks for this paper is 120, corresponding to the number of min-
utes available. The mark will be scaled to compute your final exam grade.
This paper has three parts, A-C. You should attempt all the questions.
You should enter your answers in a Word document or PDF, which can include typed and/or hand-
written answers. You should answer each question on a separate page, i.e., start a new page for each of
Questions 1–9 – parts within questions do not need new pages. Write the question number clearly at the
top of each page. You have unlimited attempts to submit your answer-file, but only your last submission
is used for marking.
You must not use materials other than those authorised above. You are not permitted to communi-
cate with others for the duration of the exam, other than to ask questions of the teaching staff via the
exam chat support (BigBlueButton). Your computer, phone and/or tablet should only be used to access
the authorised materials, enter or photograph your answers, and upload these files. The work you submit
must be based on your own knowledge and skills, without assistance from any person or unau-
thorized materials. Content produced by generative AI (including, but not limited to, ChatGPT) is not
your own work, and submitting such content will be treated as a case of academic misconduct, in line
with the University’s policy.
There is an embargo on discussing the exam contents for 48 hours after the end of the exam.
You must not discuss the exam with anyone during this time (this includes both classmates and non-
page 1 of 6 Continued overleaf . . .
COMP90049 Introduction to Machine Learning, Sample Exam
classmates.)
page 2 of 6 Continued overleaf . . .
COMP90049 Introduction to Machine Learning, Sample Exam
COMP90049 Introduction to Machine Learning
Final Exam
Semester 1, 2023
Total marks: 120
Students must attempt all questions
Section A: Short answer Questions [40 marks]
Answer each of the questions in this section as briefly as possible. Expect to answer each question in 1-3
lines, with longer responses expected for the questions with higher marks.
Question 1: [40 marks]
(a) Name three differences between exact optimization and Gradient descent. [6 marks]
(b) Align the concepts under (a) to their most typical type of supervision under (b). [3 marks]
(a) (b)
clustering supervised
classification semi-supervised
regression unsupervised
(c) [6 marks] On a given test data, a classifier detects 4 TP, 3 TN, 6 FP, and 0 FN. What are precision,
recall and F-score (assume β=1) of the classifier?
(d) [3 marks] Consider the following set of evaluation metrics
Accuracy =
TP + TN
TP + TN + FP + FN
Precision =
TP
TP + FP
Recall =
TP
TP + FN
Error Rate = 1−Accuracy
1. What types of machine learning algorithms can be evaluated with these measures? [1 mark]
2. Explain why. [2 marks]
(f) Consider the following two tasks: (1) predicting whether a job applicant is successful based on
the characteristics of their CV; (2) Predicting the expected salary of a job applicant based on the
characteristics of their CV. (i) For each task, (i) name the corresponding machine learning concept.
(ii) Justify your choice. [3 marks]
page 3 of 6 Continued overleaf . . .
COMP90049 Introduction to Machine Learning, Sample Exam
Section B: Method & Calculation Questions [55 marks]
In this section you are asked to demonstrate your conceptual understanding of methods that we have
studied in this subject, and your ability to perform numeric and mathematical calculations.
Question 2: K-Nearest Neighbors [8 marks]
With respect to the following data set of 6 instances with 3 attributes and two classes F and T, plus a
single test instance labelled ”?”:
instance # ele fed aus CLASS
1 1 1 1 F
2 1 0 0 F
3 1 1 0 T
4 1 1 0 T
5 1 1 1 T
6 1 1 1 T
7 0 0 0 ?
Explain why a model with K = 1 will make a different prediction compared to a model with K = 3 on
the given test instance. You do not need to show your work for this question, but should provide an
explanation which refers to the data.
Question 4: K- Means [10 marks]
Consider the following data set of 6 instances with 3 attributes and two classes F and T, plus a single
test instance labelled ”?”:
instance # ele fed aus CLASS
1 1 1 1 F
2 1 0 0 F
3 1 1 0 T
4 1 1 0 T
5 1 1 1 T
6 1 1 1 T
7 0 0 0 ?
Exclude the class labels from the dataset, and cluster all 7 instances using the method of “k-means”.
Apply the Manhattan Distance as a similarity measure; use the second (1,0,0) and third (1,1,0) instances
as seeds. Show your mathematical working.
Question 5: Data Sampling and Evaluation [3 marks]
Consider the following data set of instances.
# X1 X2 y
1 7 0 1
2 9 1 1
3 1 5 0
4 3 4 1
1. Is this data set linearly separable? Graphically demonstrate your answer. [1 mark]
page 4 of 6 Continued overleaf . . .
COMP90049 Introduction to Machine Learning, Sample Exam
2. Assume that instances 1-2 are the training set and instances 3-4 the test set. Further assume that
all parameters initialized to 0.3. Compute the negative conditional log-likelihood of the training
data set. [2 marks]
Question 6: Decision Trees [7 marks]
In the following dataset every row represents a patient with three descriptive features, i.e., fever, dry
cough, and headache , and Class indicates the label of each instance. Assume we are interested in
building a decision tree to determine whether a patient has flu or cold.
Patient # Fever Dry cough Headache CLASS
1 yes no mild Flu
2 yes yes severe Flu
3 no yes moderate Flu
4 no no moderate Cold
5 yes no severe Cold
6 no no severe Cold
1. Determine the attribute that a decision tree would select first based on the information gain criteria.
(Note: you need to provide the results of each step to get full marks. Show your work for computing
information gain for all three attributes. [6 marks]
You may need to use the following results:
log2(1/2) = −1, log2(1/4) = −2, log2(3/4) = −0.41, log2(1/3) = −1.58, log2(2/3) = −0.58, log2(1) =
0)
2. Calculate the Total error of the best decision stump you built in the previous step. [1 mark]
Question 7: Evaluation [7 marks]
Given the following learning curve for Naive Bayes, where N’ is the number of samples used in the training
set, answer the following questions:
1. How can you detect whether a model is overfitting or underfitting the data using the learning curve?
[2 marks]
page 5 of 6 Continued overleaf . . .
COMP90049 Introduction to Machine Learning, Sample Exam
2. Does the Naive Bayes model in the above plot have high bias or high variance? Why? [2 marks]
3. Briefly describe one strategy to overcome underfitting. (1-2 sentences) [3 marks]
Question 8: Multi-layer Perceptron [16 marks]
Consider the following labelled data set of 4 instances, 3 features (X1 ... X3) and label Y. Instances 1
and 2 are training instances, and instances 3 and 4 are test instances.
N.B: Show your mathematical working for all calculations.
ID X1 X2 X3 Y
1 0.1 0.9 -0.9 B
2 0.1 0.08 -0.5 A
3 6.4 0.9 9.8 A
4 0.3 0.9 4.5 C
The following formulas might be useful for answering the questions:
ˆ Rectified linear unit (RelU) function: z = max(
∑
i ai, 0), i.e., returning either 0 or the summed
inputs, whichever is larger.
ˆ Softmax: softmax(ai) =
exp(ai)∑
k exp(ak)
, where k ranges over all elements in vector a and i indexes one
specific element.
Please answer the following questions.
1. Describe the given machine learning task, making sure to specify the concept, features and labels.
Justify your definitions. [2 marks]
2. Construct a multi-layer perceptron which predicts a probability distribution over possible outputs,
which consists of an input layer, one hidden layer of width 2, and an output layer. Define all neces-
sary parameters including output functions and loss. Draw your multi-layer perceptron. [3 marks]
3. Initialize all MLP parameters according to the formula θlayerin,out = layer+ in× out. (For example, in
weight layer 2 the weight connecting incoming node 1 to outgoing node 2 is θ21,2 = 2 + 1 × 2 = 4.
Assume a constant bias of 1.0. (i) Perform only the forward pass of a single training epoch. For
the hidden layers, assume the ”Rectified linear unit” (RelU) activation function. For the remaining
functions, use your choices from question 2. (ii) What is the accuracy of your model for the training
instances? [7 marks]
4. Compute the loss of your model, given your results in question 3, and choice of loss function in
question 2. [4 marks]
Question 9: Feature Engineering [4 marks]
Many machine learning algorithms benefit from feature normalization as a pre-processing step. During
this step, each feature is normalized to zero mean and unit variance.
ˆ Give the formula for the normalized feature x˜j as a function of the original feature xj and the mean
µj and standard deviation σj of that feature. [2 marks]
ˆ Provide one concrete example machine learning problem (data, features, concepts, ...) where you
expect normalisation to be particularly useful. [2 marks]
page 6 of 6 Continued overleaf . . .
COMP90049 Introduction to Machine Learning, Sample Exam
Section C: Design and Application Questions [25 marks]
In this section you are asked to demonstrate that you have gained a high-level understanding of the
methods and algorithms covered in this subject, and can apply that understanding. Expect your an-
swer to each question to be from one third of a page to one full page in length. These questions will
require significantly more thought than those in Sections A–B, and should be attempted only after having
completed the earlier sections.
Question 10: Insurance Policy [25 marks]
You are a manager of a life insurance company and want to provide optimal insurance quotes to your
potential customers. The quotes fall into one of three categories ‘high’, ‘medium’ or ‘low’ premium. Your
company is so popular that you cannot sort through all applications manually. Instead, you want to
pre-sort applications into meaningful groups. Each application comes with features such as
ˆ Name of applicant
ˆ Age of applicant
ˆ Favorite color of applicant
ˆ Longest period spent in hospital
ˆ Marital status of applicant
ˆ Gender of applicant
Please answer the following questions with respect to the machine learning problem introduced above.
1. Describe the machine learning concept and features underlying this task. [3 marks]
2. Assume you have access to the following ML methods: (a) Decision trees; (b) neural networks; (c)
k-means. For each algorithm, state whether it is appropriate in this situation as well as a reason
for your decision [6 marks]
3. Now assume a slightly different situation where you (a) have access to a set of 50 admission decisions
from previous years. Describe how this new information will change (a) your machine learning
approach. [8 marks]
4. Further questions e.g., on evaluation or feature selection or bias ... [8 marks]
— End of Exam —
page 7 of 6 End of Exam