Xian posts his memories of Julian Besag, who is perhaps most famous for publishing the Hammersley-Clifford theorem (see here for some background). I met Besag in 1989 when I spoke at the University of Washington; also I have a memory of a conference in 1992 or 1993, I think it was, when he objected strongly to my use of the chi-squared test to check the fit of a Bayesian model. (In retrospect, I don’t think I presented my ideas clearly enough; some of the material in that talk ended up in this article with Meng and Stern.) I also recall a talk I gave in Seattle around 1996, when Besag commented that the models I was using for spatial analysis were pretty crude–a fair comment, actually. I don’t actually recall any of Besag’s lectures, but I read many of his papers. As Xian said, Besag did innovative and important work on spatial statistics, work that will be long used and remembered. He was in many ways ahead of his time and was a rarity in his generation of Bayesian statisticians in being motivated by models and applications rather than by theory.
Indirectly learned some things from him.
When I first heard about MCMC I hypothesized that people worked with conditional distributions because they could not write down joint distributions.
Found an early paper by Besag where he showed how to get the (relative) joint distribution from the conditional distributions.
So I took a toy 3D MCMC example from some where and converted it to drawing from joint distribution (just MC) – and thereby made MCMC forever obsolete!
Fortunately, before I mentioned this to many others, I did the famous pump example and after a whole bunch of Mathematica programming to get the joint – I ran head first into the curse of dimensionality.
A couple years later, someone visited the Stats department and presented the same thesis – they had yet to do an example with more than 3 dimensions (or as Radford Neal added – even a hard 2D example). In their defence, they retorted “I discussed this work with Julian Besag and he was very encouraging”
When I later mentioned this to someone who knew Julian Besag – he simply said “That’s Julian”
Keith
p.s. the presenter did develop a few supplementary theorems out of this (connections between conditional and joint distributions) which likely were true