Wynne Hsu's Home Page


			Director Institute of Data Science, National University of Singapore. email: idsbox@ids.nus.edu.sg Provost's Chair Professor Department of Computer Science, School of Computing. email: whsu@comp.nus.edu.sg fax: (65) 6779-4580

Ongoing Research Projects

· AI in Health Grand Challenge

The AI in Health Grand Challenge is the one of the AI Grand Challenges funded by AI Singapore to promote bold ideas and apply innovative AI technologies and approaches to solve these challenges. In particular, it aims to address the following challenge: “How can Artificial Intelligence (AI) help primary care teams stop or slow disease progression and complication development in 3H – Hyperglycemia (diabetes), Hypertension (high blood pressure) and Hyperlipidemia (high cholesterol) patients by 20% in 5 years?”. We propose to tackle this challenge with 3 P's: Predictive care instead of reactive care, Personalized care instead of one-size-fit-all, emPowered patients instead of passive patients. Multiple positions are available. Applicants with machine learning, representation learning, causal inference, human-computer-interface, emotion-aware chatbots, and AI in healthcare background are preferred.

· Social Media Analytic

These projects are in collaboration with Institute of Data Science and Center of Trusted Internet and Community. The goal is to use big data analysis tools to help understand the credibility of information in social media. We aim to study the flow of both credible information and misinformation to design strategies to diminish the influence of misinformation. We have developed a demo system to check the claims on COVID-19. Research Fellow and research assistant positions are available. Applicants with NLP background are preferred.

· Collaborative Machine Learning

This is an international and multi-institutional collaboration with SMU (Singapore Management University), MIT (Massachusetts Institute of Technology) and UCB (University of California, Berkeley) under the AISG research pillar to build trustable model-centric sharing for collaborative machine learning. Research Fellow and research assistant positions are available.

· AI for Customer Service Automation

This is a work package under the NUS-CISCO corp lab. Under this work package, we will investigate data integrity issues, knowledge augmentated representation, and intelligent interactions. Positions are available for interested applicants.

Completed Research Projects

· Medical AI Projects

These projects are related to the design, development, and validation of AI in various medical applications.

· GeoVisualization of Spatio-Temporal Disease Spread

The project is funded by CIDER, School of Public Health, for the interactive exploration of disease trends and patterns. The GeoVast system automatically highlights regions with abnormally high concentration of incidences and performs hot spots prediction.

· Flagship Project on Ocular Imaging

The project is funded by ASTAR Exploit to fully automate the Singapore Eye Vessel Assessment System (SIVA). SIVA brings together various technologies from image processing and artificial intelligence (AI) to construct vascular models from retinal images. Subsequently, these models of blood vessels can be queried for a variety of measurements which have been shown to be correlated to diseases such as stroke, diabetes, hypertension etc. This project involves collaboration between Singapore Eye Research Institute (SERI) and NUS School of Computing.

· SiRIAN: Singapore Retinal Imaging and Archival Network

The SiRIAN programme, funded by ASTAR SBIC, is focused on linking retinal image features with demographic and clinical data for risk prediction. This project involves collaboration between Centre of Eye Research Australia (CERA), I2R and NUS School of Computing.

Mining from Spatio-Temporal Databases

Spatio-temporal applications are gaining momentum especially in the last few years. The availability of spatio-temporal databases introduces the possibility of mining a new class of rules that captures changes and movements. We have designed and developed new spatio-temporal rule mining algorithms that capture the trends and behavior of spatio-temporal data. Related publications can be found here. The code for mining interval-based patterns is also available for download at https://dl.dropboxusercontent.com/u/15522119/Sigmod_code.zip.

RetinaMiner: Mining Changes in Retinal Images

A retina image provides a window into what is happening inside the human body. In particular, changes in the vascular structure of retina image have been shown to accurately reflect the cardio-vascular states of the body. The project aims to extract the vascular structure from the 2-dimensional digital retinal images and tag them with customized XML tags to enable physicians to query the changes that have occurred in the retina images. An automated spatio-temporal miner will be designed to highlight the interesting changes that occur in these vascular structures. Related publications can be found here.

Knowledge Discovery in Biological and Clinical Data

This is an I2R-SoC joint research project, funded by AStar, aimed at developing new knowledge discovery technologies for biological and clinical data. A suite of ``challenge'' databases and knowledge discovery systems for selected problems in biological and clinical data analysis are constructed. Among them, the work on protein-protein interaction network reliability and motif finding, called IRAP, is available for free download here.

Cleaning Biomedical Data

Data cleaning refers to a series of processes used to improve data quality. Existing approaches in detecting and correcting defective data are highly manual, tedious and incomplete, primarily focusing on a small subset of variables within a database. In many biomedical applications, the linkages among various data repositories such as biobank, clinical data, risk factors, clinical outcomes and imaging data, provide a rich source of knowledge for identifying likely erroneous data or records. This project will adopt a holistic approach to leverage on the data linkages for the identification of data artifacts. We will utilize data mining techniques to discover the context, trend and correlation in the data. The objective is to improve the quality of data for higher accuracy in analysis and preventing percolation of errors.

RETINA

RETINA is a joint collaboration between the National Healthcare Group Polyclinics, Tan Tock Seng Hospital, and School of Computing, NUS to investigate computed-aided screening of retinal images with: 1) abnormal cup-disc ratio, 2) small red and yellow leisons, and 3) tortuous blood vessles. This project was funded by NSTB-University Research Fund. A prototype system had been deployed in 9 polyclinics under the National Healthcare Group.

Mining from Image Databases

Images are powerful means of conveying information to human. As a result, many real-life applications involve processing and analyzing a large number of images. In spite of the widespread use of images, there is no effective techniques to mine interesting patterns from images. In this project, we investigate the unique characteristics of image data and design algorithms to automatically discover interesting image patterns. This project is funded by the Academic Research Fund at National University of Singapore.

Data Mining and Intelligent Data Analysis

Data mining has been recognized as an important technology for businesses internationally. Locally, there are many companies in Singapore that are interested in this technology. Here in School of Computing, we have designed and built a number of data mining tools (both generic and industry-specific) that can be readily adapted and used by industry users. One such tool to discover association rules and used the rules for accurate prediction is the Classification based on Association Rules, CBA. The tool is available for free download here.