Archive for August, 2009

An incomplete but revised history of Markov Chain Monte Carlo

Posted in Statistics with tags , , , on August 30, 2009 by xi'an

Our incomplete MCMC history paper with George Casella has now been refereed with (almost inevitably) a lot of requests for further quotes and other branches as well as, paradoxically, a demand to get down to 20 pages!, and the revised (albeit mostly unchanged) version has been resubmitted (for publication in the MCMC handbook Xiao-Li Meng is editing). We managed to keep within the 20 pages, except for the 5 page bibliography, which is obviously a main part of the paper… The new version is available on arXiv.

The Millenium Trilogy (tome 1)

Posted in Books with tags , , , , , , on August 29, 2009 by xi'an

After watching so many people reading Millenium Trilogy in the Paris metro for more than a year, I decided—while in New York—to buy the first volume of Stieg Larson‘s The Girl with the dragon tattoo to check whether or not it was worth the hype. The book is definitely gripping: I started it yesterday late afternoon, read it till midnight and finished it tonight! I am nonetheless not completely impressed by the novel, nor do I understand the fundamental reason for its success (more than a million copies sold in France, where the three novels were apparently translated in 2006-2007, much earlier than in the US). The central mystery plot is a classical “huis-clos”, with a murder being committed in a closed place with no obvious murderer and no corpse to show, the solution being rather predictable (because of the flowers) and anti-climactic. Some elements of the financial plot are highly unrealistic, like the school failure Lisbeth Salander speaking Oxford English and perfect German, and breaking in a few minutes into any computer or off-shore bank account, or the major villain Wennerström keeping all the informations about his criminal activities on a single hard-drive. The side inquiry about an unsuspected serial killer is more interesting (for a while) but again not very innovative compared with the rather large current literature on serial killers and female detectives (like Sara Paresky’s V.I. Warshawski)…. This is particularly striking since I read the book in English rather than in French and I did not find much differences in the style and more globally in the setting between The Girl with the dragon tattoo and current American detectiitve storiites like Paresky’s or Cowell’s. (Of course, some Swedish specificities pop up from time to time, but it could almost take place in northern Maine!) In that regard, the criticism of the Swedish social-liberal model is much more present in the older series by Maj Sjöwall and Per Wahlöö… At last, and in connection with the previous point, while the character of the autistic and unlikely investigator Lisbeth Salander is a fairly interesting creation, her vigilante attitude of implementing her own justice does not really fit within the moral higher grounds of her associate Mikael Blomkvist. Nor with the author’s left-wing and non-violent positions. Nor with the feminist Similarly, the brand name-dropping, especially for Apple products, is a bit at odds with the author’s ideological principles. To be completely honest about this book, I must add that I will most likely read both next volumes of the Millenium Trilogy when they appear in paperback, because they still make for an enjoyable one afternoon read but, again, nothing to rank it as the [Swedish [detective [techno-killer]]] novel of the century!

Re-Read RSS Paper

Posted in Statistics, University life with tags , , , , on August 28, 2009 by xi'an

In a kind of apologetic twist, the Research Section of the Royal Statistical Society has decided to have papers discussed that were already published as ordinary papers in JRSS Series B, because of their observed impact on the field. I think it is a terrific idea and I am looking forward the resulting discussions, since, given people have had much more time to think about and to implement the proposed methodology, their discussions should be deeper and more definitive than discussions about new methodologies. (The danger being an implosion of the discussion topic into summaries of independent papers!) The first paper selected for this type of discussion is Benjamin-Hochberg’s ‘Controlling the false discovery rate: a practical and powerful approach to multiple testing’ that has had an long-term impact on the way multiple testing is handled. The discussion will occur during the Society’s 175th anniversary conference in Edinburgh on September 9th.

The next Read Paper will take place on October 14 in London and is about `Particle Markov chain Monte Carlo’ by Christophe Andrieu, Arnaud Doucet, and Roman Holenstein, which is a novel approach to the construction of Markov kernels via sequential Monte Carlo methods. The paper should soon be available on the RSS website and anyone is welcome to submit written comments on the paper by October 28; if submitted before October 14, they may even be read during the meeting. Those contributions must be no longer than 400 words (plus figures) and should be submitted to Charlotte Stovell. If things proceed as last year, there should be in addition a preordinary meeting preceding the regular meeting in order to better explore the paper being discussed, organised by the Young Statisticians section of the RSS.

The Gathering Storm (a review)

Posted in Books with tags , , , on August 28, 2009 by xi'an

Although it had been posted for about a month, I only came across a review of The Gathering Storm this afternoon. This is the first volume in the Brandon Saunderson’s planned trilogy, supposed to end the Wheel of Time monumental series… (See this post for a fairly interesting if partial analysis of Jordan’s style.) The interesting point is that the review is published before the book appears on October 27! Unfortunately, it is quite limited in the informations it gives about the plot, while being vaguely reassuring about the continuity with Robert Jordan’s style and purpose. (The cover may be one of the worsts in the series, by the way.)

New York fan

Posted in Kids with tags on August 27, 2009 by xi'an

pict7139A home-made montage celebrating New York city by Rachel…

New York skyline

Posted in Travel with tags , on August 27, 2009 by xi'an


A Vanilla Rao-Blackwellisation (comments)

Posted in Statistics with tags , , , , , on August 26, 2009 by xi'an

One of the authors of “On convergence of importance sampling and other properly weighted samples to the target distribution” by S. Malefaki and G. Iliopoulos, sent me their paper (now published in JSPI, 2008, pp. 1210-1225) to point out the connection with our Vanilla Rao-Blackwellisation paper. There is indeed a link in that those authors also exploit the sequence of accepted values in an MCMC sequence to build up geometric weights based on the distribution of those accepted rv’s. The paper also relates more strongly to the series of papers published by Jun Liu and coauthors in JASA in the early 2000’s about random importance weights, and even more to the birth-and-death jump processes introduced by Brian Ripley in his 1987 simulation book, and studied in Geyer and Møller (1994), Grenander and Miller (1994) and Phillips and Smith (1996) that led to the birth-and-death MCMC approach of Mathew Stephens in his thesis and 2000 Annals paper. As later analysed in our 2003 Series B paper, this jump process approach is theoretically valid but may lead to difficulties in the implementation stage. The first one is that each proposed value is accepted, albeit briefly and thus that, with proposals that have a null recurrent or a transient behaviour, it may take “forever” to go to infinity and back. The second one is that the perspective offered by this representation—which in the case of the standard Metropolis algorithm does not involve any modification—gives a vision of Metropolis algorithms as a rough version of an importance sampling algorithm. While this somehow is also the case for our Vanilla paper, the whole point of using a Metropolis or a Gibbs algorithm is exactly to avoid picking an importance sampling distribution in complex settings because they are almost necessarily inefficient and instead exploit some features of the target to build the proposals. (This is obviously a matter of perspective on the presentation of the analysis in the above paper, nothing’s being wrong with its mathematics.)