COVID-19: Latent Clinical Phenotypes on Admission in Hong Kong across 2020

In the face of the COVID-19 pandemic, efforts in 2019 and 2020 have been devising immediate practices. However, limited knowledge on COVID-19 can lead to ineffective interventions and undesired consequences. To better understand the underlying mechanism of COVID-19, a population-based retrospective study was conducted.
It covered 7606 COVID-19 positive patients in Hong Kong across 2020. This large sample size brings great statistical power and generalizability to the results. It covers a broad disease severity spectrum and is representative of Hong Kong population.
The k-prototype clustering used here demonstrates an unprecedented strength to capture 86.6% deceased cases in one of the four clusters and identify most of the clinical characteristics of a poor prognosis without knowing them in the clustering process.
Upon further verification and investigation into these clusters, new understandings based on different clinical representations of COVID-19 will subsequently benefit disease prevention and the development of customized treatment protocols.


“The rights of many tenants in Hong Kong are left unenforced. Not only are the relevant regulations and clauses inaccessible and convoluted, it is also hard for the average tenant to know whom to approach when there is a “health and safety” issue in their buildings.

We develop a technology solution that would inform tenants about their rights using computer vision technology. Tenants can simply use their mobile phones to scan items in the common areas of their buildings to learn about the regulations and clauses relevant to those items. This would enable them to anonymously report these breaches for enforcement.”

Logistics delivery trajectories analysis by machine learning

“In traditional logistics operations, activities and their durations were not properly recorded. Thus, it is difficult to understand and model the daily operations.

In this project, with given trajectories, a methodology is proposed to infer the types of activities of the delivery workers in a journey.

The method is to analyze the trajectories which can be identified from the collections of GPS data collected by the devices the deliver worker carried when they are on duty.

We would want to know whether the delivery worker is traveling or delivering the goods given a trajectory segment.”

Augmented Reality in Teaching and Learning

A common difficulty for beginner programmers is the concept of objects and references, which is more apparent when learning Python, which hides these concepts with simple semantics. This project is a mobile application that aims to use AR visualization to teach this concept, using some data types in Python. The user is able to execute supported Python statements and view an AR animation visualizing the components involved and the underlying operation. Their behavior are shown using their AR models and animations. Sandbox mode and exercise mode are available for user to explore on their own or follow exercises that showcase the properties of the data types. With the help of this application, the user can understand the concept of object and reference more clearly.

Electrospun reusable nanofibrous membrane filters for environmental applications

Water and air pollution have become major challenges for humans in the 21 century. With limited accessible resources, a sustainable solution is needed for pollution control. Electrospun nanofibrous membranes are made of tiny fibers in nano scales with high pororsity. The special features make the membranes highly efficient and durable in pollutant adsorption. Moreover, by simple washing with easily accessible agents, the membranes can be almost fully regenerated to achieve reusability. There is high commercialization and sustainability value behind this project.

AI Catering System with Jetson Nano

This project is to design an AI catering system for automating the tedious tasks by using jetson nano and robot in order to reduce the operating costs and increase efficiency of the restaurant operation. In this project, three AI robot: Tableware classification robot, Tray collection Jetbot and Trash collection Jetbot are developed. The robots are operated by Nvidia Jetson Nano computer, a computer specialized for AI development. The robots can reduce the long queuing time for ordering and food collection, messy dirty dishes and wastes lead to high workload on dish washing, unpopular food waste collection mechanisms and floor trash recycle in a restaurant. A demo restaurant is built in this project.

Impact of Different Factors on Hyporheic Exchange in a Meandering Pool-riffle Stream

The hyporheic zone (HZ) is the region of saturated sediment surrounding a stream which connects surface water and groundwater flow. The overlying water with dissolved matters infiltrates into the HZ, stays there for some time and interacts with groundwater, and exfiltrates out of the HZ, resulting in hyporheic exchanges (HE).

Most of the rivers are designed with narrow bottom section and large upper section of concrete streambed which results in bad river eco-system. In recent decades, more and more stream restoration projects involve the recovery of HE, however, effective guidance in restoring HE is still missing. Therefore, this study aims to examine the effectiveness of different key factors (e.g., stream discharge, slope, sinuosity and groundwater flow) in restoring HZ in a meandering channel with floodplain using flume experiments and numerical simulations.