by MALINDRIE DHARMARATNE
The Australian and New Zealand Statistical Conference 2021 (ANZSC 2021) was held online from the 5th to 9th July, where this conference brought together a broad range of researchers across a variety of statistical disciplines. The conference’s theme this year was “Modelling Data for a Brighter Future” and the conference program included presentations by both international and national keynote speakers, including experts in the statistical arena, mini tutorials on trending research areas of statistics, as well as oral and poster presentations.
I had the opportunity to participate and present at ANZSC 2021, where I received a scholarship covering my registration fee from the SSA Qld. One of the main highlights of the conference was the panel of keynote speakers, and how they each represented a variety of statistical disciplines. The keynote presentation by Distinguished Professor Kerrie Mengersen, (Not) Aggregating Data, was particularly of interest as she discussed how different statistical models can be used to bring together cancer data from different registries around Australia, and how they can be combined with GIS/location details and digital earth technology.
I also had the opportunity to give an oral presentation during the Biostatistics session and my talk was titled, “scShapes: A statistical framework for identifying distribution shapes in single-cell RNA-sequencing data,” where I presented the work from my PhD research. I particularly enjoyed presenting my work to such a diverse audience, and for receiving valuable feedback. However, the biggest highlight of the conference for me was the opportunity to be part of the panel at the Women in STEM session, since I felt extremely privileged to talk about my PhD journey in my talk “Pursuing a PhD: Journey so far,” and to be on the same panel as Professor Melanie Bahlo; who’s at the forefront of statistical bioinformatics both nationally and internationally.
I would like to thank the SSA Qld for providing me with financial support to attend ANZSC 2021, as I was extremely pleased with my conference experience since I learned new statistical methods, and applications in the field which I believe will be very useful in my career. Although this year the conference was online, the organisers ensured that participants had the full conference experience through platforms like Zoom and Slack, as we had the opportunity to ask questions and network with other presenters and participants. Attending ANZSC 2021 was a fantastic experience because students and experts working in a variety of statistical disciplines came together to share their research and provide feedback to each other.
--------------------------------------------------------------------------------------------
by MOHOMED ABRAJ MOHOMED HASEEM MOHOMED AMSAR
The ANZSC2021 conference began with an inspiring keynote presentation by Professor Frauke Kreuter, as her presentation on combining data from different sources for social media was useful, and quite new to me. Afterwards, I enjoyed the wonderful talk about randomized trials by Hon. Dr Andrew Leigh, MP. However, I need to highlight the talk by Distinguished Professor Kerrie Mengersen about The Australian Cancer Atlas, since this talk helped me to understand geological based cancer data in Australia, and I hope to use this data in my research. Noel Cressie, one of my favourite authors, presented an interesting talk on the comparison of global geophysical models, and all the talks were useful for my research career, but specifically for my area of research in copula modelling, spatial modelling, and spatio-temporal modelling. The sessions for environmental statistics, modelling, statistical theory, and methods were useful, because they provided numerous insights to my PhD project.
I presented my conference talk on Day 3 titled, “Copula modelling for spatial data: a new approach to model multivariate spatial dependency,” although I was quite nervous before the talk began since this was my first experience at a virtual conference. However, I presented well and answered all the questions from the audience, and Noel Cressie’s question was especially helpful in improving my methods. Also, other members of the audience had positive feedback on Slack, which encouraged me to progress well with my research.
Firstly, I would like to express my gratitude to my principal supervisor Associate Professor Helen Thompson, who suggested I attend this conference and helped me in preparing the conference talk and abstract. Secondly, I sincerely thank my Associate supervisor Professor You-Gan Wang, who also reviewed my abstract and gave me valuable suggestions. Finally, I must thank QUT, ACEMS, and SSA Qld for their financial assistance.
------------------------------------------------------------------------------------------
by OWEN FORBES
Extending Bayesian model averaging methodology for application across multiple unsupervised clustering methods
A variety of methods have been developed to combine inference across multiple sets of results for unsupervised clustering, within the ensemble and consensus clustering literature. The approach of reporting results selected from the “best” model out of several candidate clustering models ignores the uncertainty that arises from choosing the model, and results in inference that is sensitive to the chosen model and parameters, especially with small sample size data. Bayesian model averaging (BMA) is a popular approach for combining results across multiple models that offers some attractive benefits in this setting, including intuitive probabilistic interpretation of an overall cluster structure integrated across multiple sets of clustering results, with quantification of model-based uncertainty.
Previous application of BMA for clustering has been developed in the context of finite mixture models, using the Bayesian Information Criterion (BIC) to approximate model evidence for weighted averaging of results across selected models. In this work we proposed an extension to BMA methodology to enable weighted model averaging across results from multiple clustering algorithms, by using a combination of clustering internal validation criteria in place of the BIC to weight results from each model. We presented results exploring the utility of this approach with a case study applying BMA across results, from several popular unsupervised clustering algorithms, to identify robust subgroups of individuals based on electroencephalography (EEG) data. We also used simulated clustering datasets to explore the utility of this technique to identify robust integrated clusters.
Personal Highlights
Overall, I was amazed at how smoothly and seamlessly this online conference ran, and massive kudos to the organisers for making everything run without hiccups across Slack and Zoom! Some talk highlights for me included: Professor Renate Meyer's plenary about Bayesian Time Series Tools for Gravitational Wave Astronomy; Dr Karen Lamb's talk about Obesogenic Environments and the Health Effects of Residing in 20-minute Neighbourhoods; and Dr Edgar Santos-Fernandez's talk about Spatio-Temporal Models for River Networks. I was also helped by Dr Nicole White's mini tutorial on implementing mixture models for unsupervised clustering, since this tutorial really helped bed down my familiarity and confidence with these methods, and I am grateful for Nicole's expertise and clear explanations. There were so many interesting talk titles and abstracts, that it was hard to choose which parallel session to attend live, and I’m looking forward to watching recordings of some more great talks over the next couple of weeks!
Thank you very much to the SSA Qld branch for funding my attendance, as I’m grateful for the opportunity, and I had an awesome time at ANZSC2021.
@StatSocAus
