I am a researcher at the Institute of Advanced Intelligence and Computing (IAIC), A*STAR. Before that, I received my B.Eng. with highest distinction from Nanyang Technological University. [...]
I am looking for Ph.D. opportunities starting in Fall 2027. My Ph.D. studies will be fully funded by A*STAR’s National Science Scholarship (PhD).
I work broadly on vision and learning. But my research interests? Those depend on your answer to one question:
How do you see AI?
I see AI as an engineering discipline, with the goal of building learning systems that are more scalable, robust, and efficient.
Specifically, my research approaches this through data-centric AI from three perspectives:
How do data distribution, quality, diversity, modality, and individual samples shape what models learn and how well they generalize?
How can learning methods be designed to work effectively across different domains, modalities, and real-world data conditions?
Can trained models, representations, parameters, and other learned artifacts themselves become data from which new models or knowledge can be derived?
I also see AI as an empirical science. As we move toward increasingly general intelligence, I view this process not only as building more capable models, but also as creating an artificial world of learning systems within our hands.
Just as the physical world is governed by laws that allow us to understand and predict how objects behave, I believe the artificial world we create may also exhibit its own fundamental laws of learning. My research therefore asks whether artificial learning can become an empirical science with discoverable laws of change and invariance.
Specifically, I explore this question from three perspectives:
How do models respond when we change their data, optimization, architecture, or other learning conditions, and can these changes be described by consistent principles?
Despite differences in architectures, objectives, datasets, and modalities, are there common structures, behaviors, or principles that consistently emerge during learning?
Can the regularities discovered across learning systems be distilled into compact principles that explain existing phenomena and predict the behavior of previously unseen systems and conditions?