When: Thursday 19th July 2018, 13:00-14:00.
Included in meeting:
The fortnightly BRAG meeting will be held 19/07/18 at 1pm in room Y801. This week we are joined by Nick Tierney who is visiting from Monash University. We will have presentations by Ethan and Nick.
Ethan’s Talk: A BRAG Guide to HPC
Abstract : This presentation will host a quick tutorial on how you can handle large data sets quickly using QUTs high performance computing infrastructure. Talk will discuss practical and efficient means to handling big data, how to significantly reduce the time of running experiments and some tips and tricks about how to navigate the high performance cluster. Interactive R examples will be demonstrated, so feel free to bring your laptop and play along whilst learning how to use HPC.
Ethan has shared the link of the BRAG guide to HPC in the slack channel. You can also find the guide in: https://github.com/ethangoan/hpc_guide. A big thanks to Ethan from all of us!
Nick’s Talk: Now you see it? Now you don’t? The role of graphics as MCMC diagnostics
Authors: Nicholas Tierney (1), Miles McBain (2), Dianne Cook (1)
Affiliations: (1) Monash University; (2) QUT
Abstract: Visualisations are often used to help diagnose whether the MCMC chain of a Bayesian model is behaving appropriately. For example, visualisations such as traceplots and density plots are typically used to help determine if the chain has converged to a sensible target density. However, there is currently no empirical evidence describing whether people can use visualisations to reliably identify features such as convergence of an MCMC chain. To explore this absence in the literature, we designed a study applying visual inference techniques (Buja et al, 2009) to answer the questions:
Can visualisations be used to identify features such as convergence of an MCMC chain?
Are some visualisations from better at this than others?
Is experience with Bayesian statistics related to performance?
In this talk, we discuss the basis for the common MCMC diagnostics visualisations, present our findings on their effectiveness, and share future directions of this research.
References: Buja, A., Cook, D., Hofmann, H., Lawrence, M., Lee, E. K., Swayne, D. F., & Wickham, H. (2009). Statistical inference for exploratory data analysis and model diagnostics. Philosophical Transactions of the Royal Society of London A: Mathematical, Physical and Engineering Sciences, 367(1906), 4361-4383.
We look forward to seeing you all there!
Trish and Farzana