Duke-NUS Medical School, National University of Singapore

Singapore, Singapore

During my time at Duke-NUS, we worked on adapting European cardiac arrest prediction models for Asian healthcare settings. The challenge was taking models trained on European patients and making them work well across diverse Asian populations. We collaborated with doctors and healthcare partners across multiple countries in the region, using their multi-country data to fine-tune the models. We managed to get them performing at 76-78% accuracy while ensuring they treat all patient groups fairly - which means our partners and hospitals throughout Asia can leverage these advanced tools without the massive cost of building new models from the ground up.

On a research project with colleagues, we developed ShapKAN to address a critical challenge in interpretable AI. We worked with Kolmogorov-Arnold Networks (KANs), a type of model that's designed for interpretability - it can actually derive symbolic formulas from neural networks by using B-splines instead of traditional activation functions. The problem was that existing pruning methods rely on magnitude-based approaches, which produce inconsistent results when the data distribution changes. Our solution, ShapKAN, uses the marginal contribution of each neuron to determine importance, making the pruning process more stable and reliable. This approach improved results by 4-44% over existing methods while maintaining consistent performance even when domain shifts occur - pretty important for real-world applications where data characteristics can vary. This work is currently under review at ICLR.

NCCU Data Science Club & Dogger Instrument

Taipei, Taiwan

Ｗe identified the critical commercial indicators and fused them with detailed sales‑behavior insights to build predictive models. It helps to track product trends and inform strategic decisions through revamped internal dashboards. In parallel, I created an intuitive abnormal‑stock warning system that continuously monitors inventory levels and flags irregular sales patterns, boosting operational efficiency and responsiveness.

Chung-Hua Institution for Economic Research

Taipei, Taiwan

My primary research focused on variable selection and regression modeling for recyclate markets. I produced regular reports on domestic and international recyclate prices, evaluating how global events affected market dynamics and offering actionable insights. By pinpointing indicators that closely track national price fluctuations, I built a transparent, reliable multiple linear regression model for forecasting market trends and drafted policy recommendations aimed at stabilizing the recycling industry and safeguarding recyclers’ livelihoods.