NTU HUB and Statistics

I focus on estimation problems in imbalanced data, such as those encountered in the analysis of rare disease incidence in medical research. Two key challenges arise when analyzing imbalanced data: (1) information from dominant classes masks that from smaller classes, hindering model inference and evaluation; and (2) standard statistical estimation methods often exhibit instability.

During my stay at NTU, I aim to pursue two main research directions: (a) Developing divergence-based methods for robust estimation in imbalanced data, expanding on my research on gamma-divergence, (b) Bridging the gap between divergence analysis and survival time/psychological research.

I propose exploring how divergence analysis can be applied to survival time analysis (modeling positive-valued variables with censoring), drawing parallels to its application in imbalanced data with binary (or finite multi-category) responses. This connection arises from the common challenge of estimating models with limited or skewed information. By taking advantage of the fact between divergence-based method and M-estimation, it is expected to construct more flexible survival time models than the conventional approach (e.g., proportional hazards models). Furthermore, this framework can be adapted to item response models in psychological research, promoting robust estimation, which is currently not popular in the field.