Page 84 - Undergraduate Research Opportunities Program Proceedings 2017-18

School of Engineering
Department of Computer Science & Engineering

Supervisor: WONG Raymond Chi Wing / CSE
Student: WANG Wenlong / COMP Course: UROP1100, Spring
UROP2100, Summer

Every time the students look into the courses they could take for one semester, they would always spend a lot of time in
deciding which courses to take. The course recommendation platform could suggest the most suitable courses or
pathway the student should take, according to the students’ own situation. When deciding to suggest which courses to
different students, the student would better have a chance to know what grade he would most probably get according
to his previous academic (or non-academic) performance. So here comes the grades prediction system. There could be a
lot of prediction algorithms, but here I used two most commonly used method, model-based and user-based
collaborative filtering.

Supervisor: WONG Raymond Chi Wing / CSE
Student: ZHAO Xinran / COMP Course: UROP1100, Summer

The title of my UROP project is “Research on Mining Course Structure”. During the UROP, our group is separated into 3
parts. Two of us will mine the past students’ course data and demographic data to predict the grade of a course for new
students who have not taken yet with Model-based approach and Rule-based approach. Another one was assigned to
build the user interface. I was assigned with the job to store the major requirement information, build course
dependency graph and find a way to recommend study paths with highest grade. After getting the major requirement
sheets and the course data from my.ust.hk, I was finally able to store this information in python class. However, I was
not able to find the exact highest-grade path within polynomial time. My UROP study will continue as there are still
possible improvements for the requirement and course data capturing and better algorithms could be designed after I
am better equipped with algorithm knowledge. These possible improvements and thoughts would be stated more
detailedly later in this report.

Knowledge Discovery over Database
Supervisor: WONG Raymond Chi Wing / CSE
Student: CHIU Yik Hei / RMBI Course: UROP1000, Summer

Nowadays, there are many publicly available satellites image data. In this project, we aim to extract useful features from
this data. Especially those useful for self-driving car. Such as, region representing road within a given image and sign on
the road that hint where the vehicle can turn. I have tried different algorithms for different features. Including J-net for
road segmentation and Harris Corner Detection for ground sign detection. The former generates acceptable result while
the later failed. CapsNet for ground sign detection is still under implementation at the time of writing this report.

Supervisor: WONG Raymond Chi Wing / CSE
Student: LUENAM Phoomraphee / CPEG Course: UROP1100, Spring

A k-Regret Minimizing Set is a useful concept in which a smaller subset of a database is selected while maintaining
desirable traits of the original database. Specifically, a k-RMS selects r records from the database such that the regret
th
ratio between the k best score in the database and the best score in the selected records is minimized for any possible
utility function. In this paper, we show that the problem of approximating k-RMS within is NP-Hard when k and the
dimensionality of the database are unconstrained. Furthermore, we introduce a discrete version k-RMS which, while
seemingly more tractable, can also be shown to be NP-Hard to approximate. However, we will show that there is a

74