Due to a tight June schedule (3rd conference in a week!), I only stayed one day at the SIOD 2013 conference in Saigon. (SIOD means Statistics and interaction with other disciplines.) The conference was housed by Ton Duc Thang University, on a very modern campus, and it sounded like the university had drafted a lot of his undergrads to catter to the SIOD participants: similar to the Bayesian conference in India a few months ago, those students would stand at the ready to guide us around the campus and to relay any problem to the organisers. This was very helpful and enjoyable, a plus being that most female students wore the traditional pink costume adopted by the university, but it also made me a wee bit uncomfortable as I do not know how much say those students had in this draft… In particular, most of the students I talked with were from other fields than Statistics. (And definitely not complaining, but being on the opposite very friendly the whole time!) A funny side story is that I got a wake-up call from the conference organisers in the morning as I had missed a welcome ceremony with the president due to oversleeping (itself due to an excess of iced coffee rather than minimal jetlag!). Among the few talks I attended, some French school statistics due to the presence of a large contingent from Toulouse, a talk about zero inflated normal distributions which sounded like missing-at-random normal observations (hence easy to process), and a talk about the point of using Bayes factors in hypothesis testing which essentially if independently provided a second version of my course from the previous day.
Yesterday, I also had a short discussion with Paul Minh who presented a talk on a general regenerative device for MCMC algorithms, using a bound on the target density rather than on the Markov transition in order to achieve easier regeneration. While a neat idea, this method requires the construction of a lower bound that can easily simulated. Furthermore, if the regeneration probability is low, the mixing speed may remain similar to the original MCMC sampler, as the method ressorts to a standard MCMC step on the remaining part of the target density.