And BRAG is back again! The first fortnightly BRAG meeting will be held this Thursday 09/02 at 1 pm via Zoom/GP-Y701. This week we will have presentations by @Charlotte and @Ryan.
NOTE: This BRAG will be held in Y701!
Zoom Link: https://qut.zoom.us/j/85315873893?pwd=bWlWci9lM2Z3RmFCdjc5bmppWG1qUT09
Password: brag@QUT (if prompted)
Charlotte’s Talk
Title: Possums, plants, and penguins going places
Abstract: Organisms share a common process that underlies many current ecological paradigms – dispersal. Understanding and predicting dispersal has the potential to inform diverse applications, including invasive species management and conservation planning, however we are challenged by gaps in our understanding of this important process, which filters into our model representations. Join me on a tenuous thread linking my past and present work together, from predicting possum density and dispersal with mark-recapture and individual-based modelling, to modelling the trajectories and impacts of 6 invasive weeds in the significant landscapes of the central South Island, New Zealand, and finally, to Antarctica, where it turns out there’s more to it than just penguins. I will discuss aspects of dispersal that were considered in each application, and some challenges and questions that have arisen while trying to model possums, plants, and (not really) penguins going places.
Ryan’s Talk
Title: Misspecification-robust Sequential Neural Likelihood
Abstract: Simulation-based inference (SBI) techniques are now an essential tool for the parameter estimation of mechanistic and simulatable models with intractable likelihoods. Statistical approaches to SBI such as approximate Bayesian computation and Bayesian synthetic likelihood have been well studied in the well specified and misspecified settings. However, most implementations are inefficient in that many model simulations are wasted. Neural approaches such as sequential neural likelihood (SNL) have been developed that exploit all model simulations to build a surrogate of the likelihood function. However, SNL approaches have been shown to perform poorly under model misspecification. In this talk, I’ll give details on a new method for SNL that is robust to model misspecification and can identify areas where the model is deficient.
Bayesian Research & Applications Group 