Hong Kong University of Science and Technology
COMP 4211: Machine Learning
Fall 2021
Project
Due: 26 November 2021, Friday, 11:59pm
1 Preamble
Unlike the programming assignments and problem set, this individual project is intended to be
more open-ended like many other course projects or final year projects. As such, much room is
left for you to explore. Consequently, there will only be grading guidelines but not a detailed
marking scheme.
Since the project is worth 25% of the final course grade, its workload is expected to be about
the total workload of the two programming assignments. This comparison is by no means exact
but serves to give you some ideas about the expected workload.
You have a choice between two types of project:
• A software project that involves hands-on experience of solving a machine learning task
• A presentation project that involves learning and presenting an extended topic beyond the
topics covered in class.
These two types of project are described in detail in two separate sections below.
2 Academic Integrity
Please refer to the regulations for student conduct and academic integrity on this webpage:
https://acadreg.ust.hk/generalreg.
1
3 Software Project
3.1 Objective
The objective of this project is to gain and practise the hands-on skills needed for solving more
realistic machine learning tasks through pursuing a proposed study using one of the datasets
provided.
3.2 Datasets
You are asked to choose a dataset from the following list and propose a machine learning project
to work on using the chosen dataset:
• AllSides: Ratings of Bias in Electronic Media
https://www.kaggle.com/supratimhaldar/allsides-ratings-of-bias-in-electronic-media
• COVID-19 Dataset
https://github.com/GoogleCloudPlatform/covid-19-open-data
• Food Recognition - Burger, Pizza & Coke
https://www.kaggle.com/manishkc06/food-classification-burger-pizza-coke
• Online Credit Card Transactions
https://www.kaggle.com/adityakadiwal/credit-card-fraudulent-transactions
• Panama Electricity Load Forecasting
https://www.kaggle.com/pateljay731/panama-electricity-load-forecasting
• Twitter Sentiment Analysis
https://www.kaggle.com/jp797498e/twitter-entity-sentiment-analysis
For inspiration, you may take a look at the Kaggle website (https://www.kaggle.com) and
other online resources. Sometimes a data set originally used for one task may be used in a very
different way for another task that has not been studied by others before.
Note that like many real-world datasets, the datasets above may contain missing values for
some instances. Excluding those instances may not be the best treatment. Instead, you are
recommended to explore the use of imputation methods for estimating and filling in the missing
values before use.
3.3 Machine Learning Models and Computing Facilities
The machine learning tasks based on the datasets above will more likely involve supervised and
unsupervised learning techniques than reinforcement learning techniques.
In case you plan to use some more advanced machine learning methods not covered in the
course for your project, please make sure that you also include the related methods covered in
the course as baselines for comparison. Among other things, including the baselines will help
2
to justify using more advanced methods if they can indeed outperform the simpler baselines
significantly.
Depending on the computational demand of your project, you are recommended to use either
the GPU servers provided by the Department of Computer Science and Engineering (https://
cssystem.cse.ust.hk/Facilities/ug_cluster/gpu.html) or Google Colab (https://colab.
research.google.com).
3.4 Assessment Components and Submission
There are two assessment components:
• Project report
• Source code
Note that this project should not be used for earning credits in a different course to avoid
double-dipping.
3.4.1 Project Report
The report should cover at least the following aspects of the project:
• Project title
• Student with full name, student ID, and HKUST email address
• Description of the dataset and any preprocessing
• Description of the machine learning task(s) performed on the dataset
• Machine learning methods used for solving the task(s)
• Experiments and results
3.4.2 Source Code
All the source code that you have written for this project should be submitted for grading. In
case your code is modified from another source, you are expected to acknowledge it clearly in
your report. Failure to do so is considered plagiarism.
Data files should not be submitted to keep the submission file size small.
3.4.3 Submission
Assignment submission should only be done electronically in the Canvas course site.
Your submission should contain two files: report (report.pdf) and compressed source code
(code.zip). When multiple versions with the same filename are submitted, only the latest
version according to the timestamp will be used for grading. Files not adhering to the naming
convention above will be ignored.
3
3.5 Grading Guidelines
This project will be counted towards 25% of your final course grade. The breakdown is as
follows:
• Description of the dataset and any preprocessing [15 points]
• Description of the machine learning task(s) performed on the dataset [10 points]
• Description of the hardware and software computing environment, machine learning meth-
ods, and parameter settings [10 points]
• Good programming practices in source code [10 points]
• Description of the experiments [30 points]
• Visualization and discussion of the results obtained [25 points]
Grading will be based on rubrics with five levels of achievement (excellent, good, satisfactory,
unsatisfactory, poor) for each of the items above.
An important general criterion is clarity, to the extent that others can replicate your experiments
based on the information provided in the report.
Please note again that this project should not be used for another course.
Late submission will be accepted but with penalty. The late penalty is deduction of one point
(out of a maximum of 100 points) for every minute late after 11:59pm. Being late for a fraction of
a minute is considered a full minute. For example, two points will be deducted if the submission
time is 00:00:34.
At most one NQA coupon may be used to entitle you to submit this project late for one day
without grade penalty. You must download the file named NQA.pdf from Canvas and submit it
by the original deadline to show that you want one coupon to be used.
4
4 Presentation Project
4.1 Objective
Since this is an introductory machine learning course, the topics covered are mostly elementary
though they are fundamental to understanding more advanced topics. To strike a balance
between learning elementary and more advanced/recent machine learning topics, you are asked
to self-learn an extended topic which is related in some ways to one or more of the basic topics
covered in the course. The objective of this project is learning to learn, or called meta-learning
in machine learning. Through the course, you do not just learn the topics covered but also the
learning skills to learn new topics yourself. This is arguably a more important reason for taking
the course.
4.2 Extended Topics
You are asked to choose an extended topic from the following list of 20 topics which are grouped
under different categories for your convenience:
• Automated machine learning (AutoML)
– Neural architecture search
– Automated data augmentation
• Generative models
– Variational autoencoders
– Generative adversarial networks
• Conventional supervised learning models
– One-class support vector machines
– Gradient boosting trees
• Unsupervised learning
– t-distributed stochastic neighbor embedding
– Nonlinear dimensionality reduction
– Spectral clustering
– Self-supervised learning
– Contrastive learning
• Reinforcement learning
– Policy gradient algorithms
• Models for graph data
– Graph convolutional networks
• Sequence processing
– Recurrent neural networks with attention mechanism
– Transformer networks
• Miscellaneous
– Model compression
– Semi-supervised learning
– Knowledge distillation
– Anomaly detection
– Adversarial attacks and defenses
Each of these extended topics is related in some ways to the topics covered. The list is already
comprehensive enough. No other topics outside this list will be allowed.
5
Although some of these topics are covered in books, the easiest way for you to learn would be
to look for relevant materials online. If you happen to know a topic, I suggest that you do
not choose it but a topic that is new to you. After all, the whole purpose is for you to learn
something new to expand your machine learning knowledge.
4.3 Assessment Components and Submission
There are two assessment components:
• Presentation slides
• Video presentation
Note that this project should not be used for earning credits in a different course to avoid
double-dipping.
4.3.1 Video Presentation and Slides
You are asked to prepare an oral presentation of your chosen topic in the form of a video that
is about 30 minutes long. Your face should be visible throughout the whole presentation. As a
general guideline, you are advised not to have more than one slide per minute, i.e., no more than
30 slides in total, excluding the slides for the title and section headings. The target audience
of your presentation should be students immediately after taking COMP 4211, i.e., students
who have learned the basic machine learning topics. In other words, in your presentation, you
may refer to concepts and techniques covered in the course without having to review them
again.
Here are some guidelines which may help you prepare your slides:
• Motivate the topic using intuitive examples
• Discuss how the chosen topic is related to some basic topics covered in the course
• Choose a representative model or algorithm for the topic to present
• Discuss the strengths and weaknesses of the model or algorithm
• Briefly mention or list some variants without going into details
• List the references that you have read to help you prepare the slides
Please note that this presentation flow is by no means the only possibility. It is just a suggestion
for your reference. Note that presenting a broad topic at too high a level is generally discouraged.
Rather, it is better to spend more time on a specific model or algorithm although it would be
useful to mention about variants of your choice later in the presentation.
If some of your slides are modified from another source, you are expected to acknowledge it
clearly at the end of your presentation. Failure to do so is considered plagiarism.
When your video is ready, upload it to YouTube as an ‘unlisted’ (not ‘private’ or ‘public’) video
and include its hyperlink clearly in the first slide. The video should be ready by the time you
submit the slides and no change should be made to it after the deadline.
6
4.3.2 Submission
Assignment submission should only be done electronically in the Canvas course site.
Your submission should be named slides.pdf or slides.pptx. When multiple versions with
the same filename are submitted, only the latest version according to the timestamp will be used
for grading. Files not adhering to the naming convention above will be ignored.
4.4 Grading Guidelines
This project will be counted towards 25% of your final course grade. The breakdown is as
follows:
• Good time management in presentation [10 points]
• Clarity of content in slides [20 points]
• Clarity of oral presentation [20 points]
• Effective use of examples and visual aids [20 points]
• Appropriate content for illustrating the topic [30 points]
Grading will be based on rubrics with five levels of achievement (excellent, good, satisfactory,
unsatisfactory, poor) for each of the items above.
Please note again that this project should not be used for another course.
Late submission will be accepted but with penalty. The late penalty is deduction of one point
(out of a maximum of 100 points) for every minute late after 11:59pm. Being late for a fraction of
a minute is considered a full minute. For example, two points will be deducted if the submission
time is 00:00:34.
At most one NQA coupon may be used to entitle you to submit this project late for one day
without grade penalty. You must download the file named NQA.pdf from Canvas and submit it
by the original deadline to show that you want one coupon to be used.
7























































































































































































































学霸联盟