Research

Bayesian Modelings

Bayesian quantile models for space and survival, fast enough for real data.

Bayesian models give honest uncertainty, but they can be slow. The usual asymmetric Laplace likelihood for quantile regression also fixes the shape of the errors once the quantile level is chosen. We use a more flexible error distribution and fit it with variational inference, so a spatial model with thousands of sites runs in minutes.

The same tools carry over to survival and causal questions, where flexible priors avoid strong shape assumptions.

Get in touch

Questions about our work, a collaboration, or joining the lab? We would like to hear from you.

Email

sangkyulee (at) konkuk.ac.kr

Office

Office 704, Sanghuh Research Building
120 Neungdong-ro, Gwangjin-gu
Seoul 05029, South Korea