The fortnightly BRAG meeting will be held this Thursday 21st March at 1 pm via Zoom/GP-Y801. This week we will have presentations by Amalan and Brodie.
Zoom Link: https://qut.zoom.us/j/83835087884?pwd=d2NFOExWbkswU0NNcTV5QXMxRWVNdz09
Meeting ID: 838 3508 7884
Password: 360608
Amalan’s Talk
Title: A subsampling approach for Generalised Linear Models in large data sets with model misspecification
Abstract: Subsampling is the analysis of a subset of the data for inference instead of the entire data set and is a computationally efficient and scalable method in the context of big data. When employing subsampling techniques, a crucial consideration is how to select an informative subset based on the queries posed by the data analysis. A recent method proposed involves calculating subsampling probabilities for individual data points. However, a drawback of this approach is that the calculated subsampling probabilities are contingent on the presumed accuracy of the model for the large data set. To address this limitation, our article introduces a model misspecified approach assuming the selected model is potentially misspecified. Here, the subsampling probabilities are evaluated based on minimising the loss function of the normalised average mean squared error of the response prediction. We apply our subsampling approach, designed to handle model misspecification, in a simulation study and three real-world scenarios, where its performance is benchmarked against existing subsampling techniques. The findings indicate that our approach surpasses other methods.
Brodie’s Talk
Title: What’s the Average of Gaussians?
Abstract: Through the concept of the Fréchet mean, we can take the “average” of any group of similar objects, so long as we can define some measure of distance on the space in which they live. Simply, the average is the minimiser of the weighted sum of distances to all of the objects being averaged. For example, weighting according to mass and using Euclidean distance, we gain the familiar concept of the centre of mass. In this talk, I will discuss how different measures of statistical distance can be used to generate Gaussian distributions that are, in a sense, the average sets of Gaussian distributions (of equal dimension). As this talk will cover, many choices of distance measure lead to familiar concepts in statistics, or at least to an intuitive understanding of their behaviour. However, the search for interpretations and uses of the Fréchet mean defined by the Fisher metric continues.
Bayesian Research & Applications Group 