Brag Meeting Thursday 23rd February 2023

The fortnightly BRAG meeting will be held this Thursday 23/02 at 1 pm via Zoom/GP-Y801. This week we will have presentations by Wala and Ethan

Zoom Link:

Password: brag@QUT (if prompted) 

Wala’s Talk

Title: Modelling Heterogeneity in Regression for Clustered Spatial Dependent Data

Abstract:  In this presentation, I will introduce an approach to Bayesian Geographically Weighted Regression (BGWR) analysis. The aim of this approach is to tackle the challenges posed by non-constant variation and non-normal error distributions in GWR modelling. This method combines a Gaussian mixture model and Dirichlet process to identify clustering in covariate effects and can handle both continuous and categorical variables. Additionally, the optimization of the algorithm for large spatial scales and multiple regions results in substantial improvement in Markov Chain Monte Carlo estimation. To demonstrate the effectiveness of this novel method, I will present a case study on children’s development domains in Queensland, using real data from the Children’s Health Queensland and the Australia Early Development Census, as well as simulated data.

Ethan’s Talk

Title: Piecewise Deterministic Markov Processes for Bayesian Neural Networks

Abstract: Practical application of MCMC methods for neural networks requires sub-sampling of the likelihood. PDMP methods offer a way to enable this whilst still targeting the posterior, though require sampling event times from a challenging Inhomogeneous Poisson Process. This talk will discuss ways we can approximately sample these event times to allow for inference in neural networks, and some results that highlight the potential and limitations of applying these methods to neural networks.


Jamie & Scott
BRAG Co-Chairs

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s