Kang, Sung Ha
Professor of Mathematics, Georgia Institute of Technology

Research Interests and Directions

Through the study of mathematics, we can solve and understand problems arising in many real world problems. I am interested in new modeling of various applied problems motivated by, but not limited to, using variational functionals and differential equations. Various analysis of the properties provides more insight into the problem and the proposed methods, and designing practical, fast and advanced numerical methods are also of interest. These research is closely related to data science for example, reg-K-means and the serious of work on identification of differential equations.
Identifying Differential Equation from Data Top

Can we identify differental equation from a given single set of data using numerical methods ?
We explore a new direction of problem, where real physical phenomenon is captured by different digital devices, and we identify the underlying governing differential equations to help understand the phenomenon. Identifying unknown differential equations from a one given set of discrete time dependent data is a challenging problem. A small amount of noise can make the recovery unstable, Nonlinearity and varying coefficients add complexity to the problem.
We explore how far we can identify using numerical techniques.


WeakIdent can identify PDEs and ODEs. This example shows identifying ODEs: from the given data (red), green curve is generared from the identified differential equation. WeakIdent is stable againt high level of noise, and give accurate recovery.


Regularized k-means Clustering and applications Top

How to automatically cluster the given data without knowing the number of clusters ?
Extending from unsupervise segmentation model, we explore data clustering. We explore parallel algorithm using regularized k-means, and this method is able to find changing number of clusters as more data streams and the characteristic changes.


Mathematical Image Processing Top

There are many interesting imaging problem, such as image denoising, deblurring, color image analysis, multiphase segmentation and others. The following shows some of such examples using variational and PDE based model osved by various numerical methods.