I have keen interests in methodological development (big data analytics, experimental design, interpretable machine learning) and their real applications (banking and finance, education and biomedicine).

Interpretable Machine Learning

Unlike traditional approach to machine learning, we emphasize statistical models, inferences and interpretations. On the left is our holistic view of multiple machine learning algorithms in terms of both prediction accuracy and model explainability. The interpretable machine learning lies in the northeast direction.

Big Data Analytics

The rapid growth of sample size and feature dimensionality poses challenges for traditional statistical methods. To tackle such big data challenges, we study the effective ways for performing statistical modeling based on reduced data via subsampling, projection, partitioning and aggregation techniques.

Experimental Design

UniDOE: a stochastic optimization package for constructing uniform space-filling designs

R:UniDOE package

Financial Risk Modeling

Dual-time analytics and joint age-horizon credit risk models.

Educational Data Mining

Machine learning approach vs. traditional latent variable modeling in large-scale educational assessments as well as learning anlaytics.

Medical Image Analysis

Fully automated deep learning approach vs. conventional radiomics models.

2018 Data Science Bowl (Top 1%, Silver solution)