Statistics Across Campuses

Statistical Inference for Implicit Models using Bayesian Synthetic Likelihood Date: 9 September 2021, Thursday Time: 12pm AEST Contact the organizer: Andriy Olenko a.olenko@latrobe.edu.au Speaker: Prof Christopher Drovandi (QUT) Abstract: Implicit models are defined as those that can be simulated but the associated likelihood function is intractable.  Such models are prevalent in many fields such as biology, ecology, cosmology and epidemiology.  Given the unavailability of the likelihood function, statistical inference for implicit models is challenging as we must rely only on the ability to generate mock datasets from the model of interest, and compare it with the observed data in some way.  This talk will explain a useful method called Bayesian synthetic likelihood for conducting such statistical inference.  I will discuss how BSL can be extended to reduce the number of model simulations required and to make it more robust to model misspecification.  I will a...

    SIMPLE: Statistical Inference on Membership Profiles in Large Networks Date :  Thursday, 2 September 2021 Time:   10 am  - 11 am Speaker:  Professor Xiao Han Abstract: Network data is prevalent in many contemporary big data applications in which a  common interest is to unveil important latent links between different pairs of nodes. Yet a simple fundamental question of how to precisely quantify the statistical uncertainty associated with the identification of latent links still remains largely unexplored. In this paper, we propose the method of statistical inference on membership profiles in large networks (SIMPLE) in the setting of degree-corrected mixed membership model, where the null hypothesis assumes that the pair of nodes share the same profile of community memberships. In the simpler case of no degree heterogeneity, the model reduces to the mixed membership model for which an alternative more robust test is also proposed. Both tests are of ...

      Screening methods for linear errors-in-variables models in high dimensions Date :  Friday, 27 August 2021 Time:   3 -4 pm Speaker:  Dr   Linh Nghiem (ANU) Abstract: Microarray studies, in order to identify genes associated with an outcome of interest, usually produce noisy measurements for a large number of gene expression features from a small number of subjects. A common approach to analyzing such high-dimensional data is to use linear errors-in-variables models; however, current methods for fitting such models are computationally expensive due to non-convex optimization or sampling from high dimensional distributions. We develop two efficient screening procedures to reduce the number of variables for final model building; both of them have strong theoretical support and can be computed efficiently even with a large number of features. Through simulation studies and an analysis of microarray data, we demonstrate that the two new screening proce...

   When Does Fast Bootstrap Work? Date :  Thursday, 19 August 2021 Time:   10 am  - 11 am Speaker:  Dr Nan Zou (Macquarie University) Abstract: In classic statistical inference, the bootstrap stands out as a simple, powerful, and data-driven technique. However, when coping with massive data sets, which are increasingly prevalent these days, the bootstrap can be computationally infeasible. To speed up the bootstrap for massive data sets, the “bag of little bootstraps” has been invented in 2014. Despite its considerable popularity, little is known about the bag of little bootstraps’s theoretical properties, including reliability. Indeed, our preliminary results have already raised questions on the applicability of the bag of little bootstraps under a simple but important setting. This talk will first introduce the bag of little bootstraps procedure and then investigate its theoretical applicability. Specifically, for this applicability, this talk will present...

     Wind energy forecasting in an operational context: traversing research and industry Date :  Friday, 20 August 2021 Time:   3 -4 pm Speaker:  Dr  Rachael Quill (Weatherzone) Abstract: Forecasting in an operational context requires the combination of scientific rigor with computational efficiency and ongoing validation with on-the-fly development. Predicting wind energy generation in real-time necessitates modular and dynamic forecasting systems which are capable of ingesting various and variable data streams to produce consistent, reliable and accurate power predictions. From data cleaning to atmospheric processes, in this seminar we discuss case studies, from different perspectives across academia and industry, which highlight some of the challenges faced in formulating such a comprehensive operational forecast system. Short Bio:  Dr Rachael Quill is an early career statistician with experience in modelling, analysing and predicting wind fiel...

  Spectral Subsampling MCMC for Stationary Multivariate Time Series Date: 19 August 2021, Thursday Time: 12pm AEST Contact the organizer: Andriy Olenko a.olenko@latrobe.edu.au Speaker: Dr Matias Quiroz ,  University of Technology Sydney Abstract: Spectral subsampling MCMC was recently proposed to speed up Markov chain Monte Carlo (MCMC) for long stationary univariate time series by subsampling periodogram observations in the frequency domain. This talk presents an extension of the approach to stationary multivariate time series. We also propose a multivariate generalisation of the autoregressive tempered fractionally differentiated moving average model (ARTFIMA). The new model is shown to provide a better fit compared to multivariate autoregressive moving average models for three real world examples. We demonstrate that spectral subsampling may provide up to two orders of magnitude faster estimation, while retaining MCMC sampling efficiency and accuracy, compared t...

    Dealing with multicollinearity in Geographically Weighted Regression Date :  Thursday, 12 August 2021 Time:   10 am  - 11 am Speaker:  Dr Patricia Menéndez  ( Monash University) Abstract: Geographically weighted regression (GWR) is a popular technique to deal with spatially varying relationships between a response variable and a set of predictors. However, GWR estimates might be affected by multicollinearity issues related to locally poor designs. In this study, we propose two regularization methods to deal with those issues. The first one is based on a generalized ridge regression, which can also be seen as an empirical Bayes method. We show that it can be implemented using ordinary GWR software with an appropriate choice of the weights. The second one augments the local sample using an enrichment strategy. The methods will be illustrated with simulations and with an example of housing prices in the city of Bilbao (Spain). Zoom Link...

    Representativeness and Generalisability of Inference for Statistical Models for Samples of Networks Date :  Friday, 13 August 2021 Time:   3 -4 pm Speaker:  Dr   Pavel Krivitsky (UNSW) Abstract: Joint modelling of large samples of networks collected from similar settings—classrooms, households, etc.—has a long history, with a variety of methods available to pool information in model estimation and inference. In the exponential-family random graph modelling framework, these methods range from post-hoc two-stage meta-analyses to sophisticated multilevel approaches. However, relatively little attention has been devoted to the generalisability of this inference, especially when the sample of networks is effectively a convenience sample, and when the population of networks is heterogeneous in size and composition. We consider two samples of within-household contact networks in Flanders, Belgium, which used very similar survey instruments but very differ...

  Tracy-Widom law for the extreme eigenvalues of large signal-plus-noise matrices Date :  Thursday, 5 August 2021 Time:   11 am  - 12 noon Speaker:  Zhixiang Zhang  ( Nanyang Technological University ) Abstract: We study the asymptotic distribution for extreme eigenvalues of large signal-plus-noise type of matrices. Assume that the data matrix is S=R+X where the signal matrix R is allowed to be full rank and the noise matrix X contains i.i.d. standardized entries. Under some regularity conditions of the signal matrix R that assure the square root behavior of spectral density near the edge, we prove that the extreme eigenvalues of signal-plus-noise matrices have Tracy-Widom distribution under a tail condition of entries of X. Moreover, the tail condition is proved to be necessary and sufficient to assure the Tracy-Widom laws. Applications of our results on signal detection and data privacy will be discussed. Zoom Link:  Please contact Yanrong Yang ...