Bayesian Analysis (last issue)
Just got the email that the latest issue of Bayesian Analysis was complete. Plenty of goodies! From Kristian Lum’s and Alan Gelfand’s spatial quantile regression (discussion) paper that would have been particularly well-suited for LGM 2012, to my nearest neighbours Vincent’s and Judith’s paper on infinite exponential families, to two papers on Beta processes, incl. one by my former student Roberto Casarin, to two non-parametric papers, and even to our comparative study of regularization in regression… Enjoy!
Xi'an's Og 
June 4, 2012 at 1:28 pm
David Bolin (who was at the LGM meeting talking about something else) wrote a paper about Galerkin approximations (and Markovian approximations) to Laplace random fields. I tried to find a link, but it doesn’t appear to be on the internet (which is odd as I have an official Lund tech report version….). It’s a very very nice paper!
The idea is basically the same – after Markovian approximation you get, conditioned on n Gamma random variables (n=number of nodes), that the thing is a (Markovian) Gaussian random field.
It doesn’t do quantile regression, but (to me) that’s not the interesting bit of Lum and Gelfand’s paper.
Pet peeve: I really wish they were less vague about the behaviour of their MCMC sampler. This is a *hard* problem and it would be nice (or even necessary to “believe” the results) to see some proof that the scheme converged!
June 4, 2012 at 1:31 pm
Oh – I just saw this:
“As in Banerjee et al. (2008), we randomly select 100 locations from among the 3,229 locations of the births to be the knots.”
That’s not an exceptionally good idea! As shown in the subsequent papers by Banerjee and co-authors (and also by me in a different context), knot placement is extremely important for these types of random fields. To the extent that bad knot placement can lead to catastrophic results!
Oh well, at least they don’t say it’s O(N)….