Kenyon Ng

Papers

Ng, K., Turlach, B. A. and Murray, K. (2019). A flexible sequential Monte Carlo algorithm for parametric constrained regression, Computational Statistics & Data Analysis. DOI, arXiv

“A flexible sequential Monte Carlo algorithm for shape-constrained regression” (poster, talk), at
- BioInfoSummer 2018, Australian Mathematical Science Institute, Perth, Dec 2018;
- Faculty Research Conference, The University of Western Australia, Perth, Oct 2018;
- Royal Statistical Society 2018, Cardiff, Sep 2018;
- May Meeting, Statistical Society of Australia (WA Branch), Perth, May 2018.

Here are the projects that I was/am still involved in:

Longitudinal modelling with shape-constrained Bayesian penalised splines
- Current, with B.A. Turlach and K. Murray;
- I am building Bayesian semiparametric regression models for modelling dataset where the covariate is time (e.g. growth of human heights). Normally we don’t take measurements just from one but thousands of subjects, and thus the measurements coming from the same subject are not independent (i.e. linear regression does not work!). The resulting models are very challenging to fit due to their high-dimensional nature, and I am developing new methods to fit these models fast and efficiently.
Flexible optimisation routines for shape-constrained regression
- Completed, with B.A. Turlach and K. Murray;
- I am exploring novel algorithms to solve constrained optimisation problems. The users only need to provide an indicator function that tells whether or not a solution satisfies the constraint.
Curve fitting using monotone rational functions
- Completed, with B.A. Turlach and K. Murray;
- It is often known by some external theory that the relationship between two variables is monotone (e.g. the growth of a tree over time, we don’t expect a tree to shrink when it gets older).I was developing methods for fitting rational function models, with a particular emphasis on enforcing monotonicity constraints on the regression curves.