Page 82 - Undergraduate Research Opportunities Program Proceedings 2017-18

School of Engineering
Department of Computer Science & Engineering

In-memorycachecandrasticallyimprovetheperformanceofdataintensive jobs. In a shared environment, the huge
performance advantage and limited size of memory require a well-designed cache sharing scheme that provides both
isolation guarantee and high utilization. This scenario deviates from classical resource allocation as the cached files allow
non-exclusive access from multiple users, which poses a unique challenge to achieving strategy proofness. Several al
location policies have been proposed, but they either fail to prevent harmful manipulation or introduce high
computational overheads. In this project, it is shown that cache sharing process can be identified as a potential game.
We measure its performance by analyzing the worst-case efficiency, and make comparisons with other cache sharing
schemes. It outperforms existing policies and reaches near-optimal global efficiency in simulations.

Supervisor: WANG Wei / CSE
Student: YANG Baichen / SENG Course: UROP1100, Summer

When dealing with data-intensive jobs, using in-memory cache to store hot files can drastically speed up reading and
writing procedure. Meanwhile, under a cloud environment, multiple users can share the in-memory cache non-
exclusively, which increase files’ utilization. However, while the huge performance benefits are given, memory size is
comparatively small and limited. So, this situation encourages strategic users to game the system and get more reward
at a lower expense, which is famous as "free-riding" problem. In this project, we treated the caching process of multiple
users as a potential game. And we used a Game Theoretical View to design a pluggable cache management system,
GTCS, atop Alluxio, a popular memory-centric file system. We measured the performance of the system and give an
evaluation for it.

Research on Mining Course Structure

Supervisor: WONG Raymond Chi Wing / CSE
Student: CHANG Yung Wei / COGBM Course: UROP1100, Fall
UROP2100, Spring

The recent introduction and development of Massive Online Open Courses (MOOCs) has brought many new
opportunities to the world of online learning. One of these opportunities is customized learning pathways that offer
students with personalized curriculums that are tailor-made to suit their own needs. In order to achieve this, it is critical
for us to analyze data gathered from existing MOOCs and extract student performance indicators as well as their
correlation to pathway design. This project is a continuation of work performed by Mr. CHEUNG Tsz Him, Ms. LU Xiuyuan,
and the author of this report, CHANG Yung Wei, and is carried out under the supervision of Professor Raymond WONG
of the Hong Kong University of Science and Technology.

Supervisor: WONG Raymond Chi Wing / CSE
Student: CHEN Qixu / COMP Course: UROP1100, Summer

The title of our UROP is “Research on Mining Course Structure”. Our final objective is to build a system that can predict
the grade of a student based on his/her profile and make recommendation according to the prediction results. My role
in this project is implementing the rule-base prediction part. The rule-base prediction serves as a subsystem of the
whole course recommendation project. It applies the approach of using association rules to predict the letter grade an
individual student will get when taking a course. During this UROP, I have learned a lot about python programming,
specifically the library for association rule “Orange3” and the library for data handling “Pandas”. This report will give a
general idea about how the association rules are generated and formatted, and how the predictor is built.

72