Page 86 - Undergraduate Research Opportunities Program Proceedings 2017-18

School of Engineering
Department of Computer Science & Engineering

C.Long, R.C.Wong and H.V.Jagadish. Based on that paper, we’ll give a problem statement for our new algorithm suitable
for lane information. And then, we’ll try to develope our own algorithm based on the algorithms that has been
mentioned above.

Supervisor: WONG Raymond Chi Wing / CSE
Student: WANG Guanzhi / COSC Course: UROP2100, Fall

This report is a group progress report. Our project mainly focuses on implementing Velocity Vector Preserving Trajectory
Simplification algorithm, which was done by a previous student attended in UROP program. You may read “Velocity
Vector Preserving Trajectory Simplification” for details of the algorithm. This report completely builds on that report and
all notations in this paper are same as those in Table 1 of “Velocity Vector Preserving Trajectory Simplification”. We
implement all three methods including the parallel version mentioned in the report. The implemented codes will be
provided below, which follows and improves “Velocity Vector Preserving Trajectory Simplification”. We also have a
detailed plan to test our algorithm, which is not done yet.

Learning Analytics for a Personalized E-Learning Platform
Supervisor: YEUNG Dit Yan/ CSE
Student: CHAN Wing Kin / COSC Course: UROP1100, Fall

To understand the progress of students’ and provide adequate interference for the success of academic study, it is
important to provide a novel way to gain insight of students’ intrinsic learning state. The extracted intrinsic learning
state can be used to conduct other prediction. In this report, the method of extracting students’ learning state with
Deep Knowledge Tracing (DKT) model is demonstrated. To show the feasibility of DKT model, our team participated in
the ASSISTments 2017 Challenge. In the challenge, collected data from an online question answering system is used to
predict whether students will enter STEM field. This report will study how the extracted result from DKT is streamlined
as an extra feature, which has boosted the predictability significantly.

Supervisor: YEUNG Dit Yan/ CSE
Student: LIN Zizheng / MATH-CS Course: UROP1100, Fall
UROP2100, Spring

This report will cover discussion, comparison and evaluation of different algorithms of knowledge tracing adopted in the
2017 ASSISTments Data Mining Competition, which aims at using data from a longitudinal study for predicting a brand
new outcome of students never studied before by the research community. Specifically, it facilitates research in
developing predictive models that predict whether the first job of a student out of college belongs to a STEM field. To
tackle this challenge, we first estimate the expected knowledge states of students with respect to different skills using
two deep knowledge tracing models including DKT and DKT+. We then combine the learned features expected
knowledge states with other features from the static attributes of the dataset to train different machine learning models.

Supervisor: YEUNG Dit Yan/ CSE
Student: LIU Ji / COMP Course: UROP1100, Fall

As online learning becomes more and more convenient and widely spread, a huge amount of students are studying on E-
learning platforms. However, there are still many drawbacks in existing online learning platforms. For example,
problematic students cannot be pinpointed and personalized assistance cannot be delivered accurately to individuals

76