Kenyon Ng - Research

Research Interests

My research focuses on developing machine learning and statistical methods that are both easy to use and trustworthy.

Some of my current research directions include:

Tabular foundation models (and transformers)

Prediction on tabular data has long relied on methods such as random forests. New transformer-based foundation models for tabular data, such as TabPFN, now offer strong off-the-shelf performance without task-specific tuning. I study how to make these models more trustworthy, especially by improving the calibration of their uncertainty estimates and by investigating where they can be used effectively.

Generalised Bayes

Most statistical analyses begin with assumptions about the data, and sometimes those assumptions are simply wrong. When that happens, the conclusions can be misleading. My research explores ways to make inference more robust, using methods such as loss-based inference (blog) and predictive Bayes inference (review).

Publications (methodology)

* denotes equal contributions

Tan, Y.S., Ng, K., Deng, R., Loganathan, S., Zhang, Q., Chakraborty, B., 2026. PFN-TS: Thompson Sampling for Contextual Bandits via Prior-Data Fitted Networks. under review. arxiv, code

Fortini, S.*, Ng, K.*, Petrone, S., Rousseau, J., Wei, S., 2026. A principled framework for uncertainty decomposition in TabPFN. under review. arxiv, code

Ng, K., Fong, E., Frazier, D.T., Knoblauch, J., Wei, S., 2026. TabMGP: Martingale Posterior with TabPFN. International Conference on Machine Learning (ICML). arxiv, code

Ng, K., Yu, W., Bondell, H.D., 2025. Expectation-propagation for Bayesian empirical likelihood inference. under review. arxiv, code

Ng, K., van der Heide, C., Hodgkinson, L., Wei, S., 2025. Temperature optimization for Bayesian deep learning. Conference on Uncertainty in Artificial Intelligence (UAI). arxiv, code

Ng, K., Wei, S., 2025. Pathwise gradient variance reduction with control variates in variational inference. Australasian Joint Conference on Artificial Intelligence (AJCAI). arxiv

Ng, K., Turlach, B.A., Murray, K., 2019. A flexible sequential Monte Carlo algorithm for parametric constrained regression. Computational Statistics & Data Analysis. arxiv

Publications (applied)

Ng, K., Diepeveen, D., Farre, I., Reeves, K., Biddulph, B., 2022. Grain yield response to sowing time, how many different response curves and maturity groups are there? Developing maturity type grain yield response curves to sowing time in Western Australia. 20th Agronomy Australia Conference. paper