The USP Lecture Series invites distinguished speakers to address a current topic from their field of specialisation, with the aim of inspiring the audience to learn more and make connections across different disciplines.

 

Imagine an online shopping platform that shows you pictures of celebrities you resemble and transplants make-up from the celebrities’ faces on to your own, allowing you to make a more informed decision by “trying-out” items you wish to purchase. Advances in computer vision – which aims to develop computers that can gain high-level understanding from digital media – are poised to revolutionise e-commerce, along with a host of diverse areas ranging from medical diagnosis to molecular biology to military operations. While these advances promise new enabling technologies, convenience, and precision, they also confront us with ethical issues such as invasion of privacy.

Dr Terence Sim works at the forefront of some of these advances. Dr Sim, who obtained his education from the hallowed halls of MIT, Stanford, and Carnegie-Mellon, specialises in facial image analysis. An Associate Professor at the NUS School of Computing, Dr Sim is a familiar face among undergraduates and graduates alike as he both teaches a foundational module and advises PhD students.

 

Terence Sim 01

Dr Terence Sim delivering the lecture.

 

Unsurprisingly, some of his students and colleagues were among the audience members who attended Dr Sim’s lecture, titled “Getting Computers to See”. This cozy setting allowed Dr Sim to better engage the audience with questions and personal anecdotes. He shared how his interest in computer vision was informed by his low clarity of vision, adding “I do what I lack.”

Over the course of his lecture, the audience – which included several members from a non-computing background – explored the various sub-fields of computer vision, application areas, key challenges, and recent breakthroughs made by Dr Sim’s team. One highlight of the lecture was Dr Sim’s demonstration of a prototype of the online shopping platform that his team had developed.

Another major breakthrough Dr Sim’s team achieved was the development of an algorithm that uses facial motion over appearance as the basis for face detection. This allows them to distinguish identical twins easily – as twins might share similar facial features but smile or react very differently.

Despite these breakthroughs, Dr Sim highlighted that teaching computers to see “is a much harder problem than meets the eye”. At one point in the lecture, Dr Sim showed the audience a photograph of a young woman and asked them to raise their hands if they knew who it was. Two hands went up. He then revealed more of the same photograph to show that there was a young man sitting beside her. More hands went up. The audience had answered correctly – the photograph showed a young Hillary and William Clinton. After this exercise, Dr Sim explained that while humans can easily perform face recognition via association (identifying Bill Clinton and then Hillary Clinton through an informed guess), it is difficult to train computers to do so.

 

Terence Sim 02

Dr Terence Sim sharing more about how computers can be taught to analyse CT scans.

 

In another memorable moment from the lecture, Dr Sim showed the audience a mathematical model of a human face that he had used in his research. Students from the School of Computing were taken by surprise as they realised that Dr Sim had just applied a fundamental concept from their introductory module on linear algebra in his high-level research. USP student Vignesh Shankar (Computer Science + USP, Class of 2021) shared, "Dr Sim made me realise that linear algebra is important and how applying AI onto real problems is immensely useful and fulfilling."

When asked about his motivations in delivering an introductory lecture on computer vision, Dr Sim remarked that it is “good for people to be aware of what technology can and cannot do.” The double-edged nature of facial recognition technologies means that “people need to know when the technology is worth the convenience and when it leads to privacy issues.” Dr Sim thus emphasised that conversations on adopting such technologies must not exclude discussions on their ethical implications.

Dr Sim’s lecture marked the launch of the University Scholars Programme’s (USP) flagship lecture series. Associate Professor Low Boon Chuan, USP’s Deputy Director for Academic Matters, explained that USP Lecture Series hopes to “bring interdisciplinary talks to the general public” on contemporary topics of importance, particularly those for which the “future might not be so obvious.” He also mentioned that this lecture series would recur every other semester, promising that there will be “many interesting topics coming soon.”