Julian Besag passed away last Friday to little notice, save for an update to his Wikipedia entry and kind memories on some statistical blogs. I don't suppose the wider media will care, but do expect that the Bayesian statistical community will take notice and remember his contributions as the word spreads.
Besag authored a number of important papers on image processing / spatial statistics, including "On the Statistical Analysis of Dirty Pictures," which has one of the all-time great paper titles -- it's about a method for reconstructing images in the presence of noise, or static, but statisticians can't resist a chance to employ double entendre.
For me personally, however, his most important work is "Bayesian Computation and Stochastic Systems", coauthored with Peter Green, David Higdon, and Kerrie Mengersen. Higdon arrived as a visiting professor at Duke University the same fall I arrived as a PhD candidate; this was one of the papers he used to teach us about Markov Chain Monte Carlo (MCMC)**, and this was the paper that really helped me to grasp MCMC methods. Thank you, Besag, for teaching Higdon so that he could teach me.
I had a chance to meet Besag in Oaxaca during the 1995 meeting of the International Society for Bayesian Analysis (ISBA), at which I was presenting a paper written with Valen Johnson on "General Strategies for Assessing Convergence of MCMC Algorithms Using Coupled Sample Paths", and we discussed this and Propp & Wilson's then-recent work on exact sampling using coupled chains (well, Val and Julian discussed, and I mostly listened) during the bus ride to Monte Alban.
** in addition to the usual Hastings, Geman and Geman, and Gelfland and Smith papers
No comments:
Post a Comment