Here are some slides from a few of my recent, and not so recent talks. To whatever extent a set of slides are useful without someone talking over them, please enjoy my deeply odd children. I have a bad tendency to reuse talk titles, so it’s possible that the slides here don’t match up with your memory of a talk (should you have had the dubious pleasure of having actually seen one of these). For the most part, I’ve used the most recent version, because it’s almost always better than previous versions.
First up we have a set of talks covering a pile of my current thoughts on Bayesian modelling and computing. The first talk covers how hard it is to make computers compute the thing that you asked for. The second is about how hard it is to make priors do what you thought they’d do. The third one is a constructive system to build a type of prior distribution (which we called PC priors) that will help alleviates some of the “Esther Williams” problems. The fourth is a first stab a a program of work I’ve started with Lauren Kennedy trying to really dig down into the fundamentals, frustrations, and felicities of survey modelling with multilevel modelling and poststratification (MRP) and its various extensions.
- Sometimes all we have left are pictures and fear (MIT, 2019)
- Esther Williams in the Harold Holt Memorial Swimming Pool (Astrophysics, Monash, 2019)
- Placating pugilistic pachyderms: Proper priors prevent poor performance (University of Sydney, 2019)
- Oh shit! It’s a survey! (University of Washington, 2019)
Next up we have a triptych of talks that cover some of the work I’ve done on spatial modelling. There’s a lot more out there (in particular, I haven’t put any talks on my favourite spatial paper about log-Gaussian Cox processes), but these are fairly representative talks.
- Sometimes having a continuous formulation is useful. Sometimes it isn’t. (University of Wollongong, 2019)
- With low power comes great responsibility: Challenges in modern spatial data analysis (Chalmers University, Gothenburg, 2015)
- Practical spatial statistics: utilising the continuous Markov property (University of Reykjavik, 2014)
Finally, we have three talks about various pieces of computational work I’ve done. This definitely is not representative of the talks I’ve given on this (there’s no MCMC here and very little INLA), but the oldest one is an unpublished extension of my PhD work, which I liked even if it turned out to be mostly useless.
- Everything* I know about importance sampling. *That I can cover in 45ish minutes. (University of New South Wales, 2019)
- The numerical challenges of moving beyone “Uncertainty Quantification” and towards “Statistics” (Rutherford Appleton Laboratory, 2016)
- How important are Cholesky factorisations for performing computations with large Gaussians (Southern Workshop on Uncertainty Quantification, Dunedin, 2013)