Archive for train

impressions, soleil couchant (#2)

Posted in pictures, Travel with tags , , , , , , , on July 12, 2014 by xi'an


impressions, soleil couchant

Posted in pictures, Travel with tags , , , , , , on July 9, 2014 by xi'an


MCqMC 2014 [closup]

Posted in pictures, Running, Statistics, Travel, University life, Wines with tags , , , , , , , on April 16, 2014 by xi'an

Leuven6As mentioned earlier, this was my very first MCqMC conference and I really enjoyed it, even though (or because) there were many topics that did not fall within my areas of interest. (By comparison, WSC is a serie of conferences too remote from those areas for my taste, as I realised in Berlin where we hardly attended any talk and hardly anyone attended my session!) Here I appreciated the exposure to different mathematical visions on Monte Carlo, without being swamped by applications as at WSC… Obviously, our own Bayesian computational community was much less represented than at, say, MCMSki! Nonetheless, I learned a lot during this conference for instance from Peter Glynn‘s fantastic talk, and I came back home with new problems and useful references [as well as a two-hour delay in the train ride from Brussels]. I also obviously enjoyed the college-town atmosphere of Leuven, the many historical landmarks  and the easily-found running routes out of the town. I am thus quite eager to attend the next MCqMC 2016 meeting (in Stanford, an added bonus!) and even vaguely toying with the idea of organising MCqMC 2018 in Monaco (depending on the return for ISBA 2016 and ISBA 2018). In any case, thanks to the scientific committee for the invitation to give a plenary lecture in Leuven and to the local committee for a perfect organisation of the meeting.

art brut

Posted in pictures, Travel with tags , , , , , on December 24, 2013 by xi'an


Path storage in the particle filter

Posted in Books, Kids, Statistics, Travel, University life with tags , , , , , , on July 24, 2013 by xi'an

IMG_0324In the train to Annecy, I read the recently arXived paper by my former PhD student Pierre Jacob (now at NUS), along with Lawrence Murray (Perth), and Sylvain Rubenthaler (Nice), where they obtain precise degeneracy rates of the regular particle filter applied to hidden Markov models with a compact observation space, precise enough to consider storing the entire paths at a linear occupancy rate. Interestingly, the distance to the most common ancestor is of order N log N, if N is the number of particles. And the number of nodes is O(N log N) as well. This means indeed that the whole paths can be stored, which offers a lot of potential in terms of Rao-Blackwellisation and parallelisation. I was first bemused by a silly misunderstanding about the purpose of the algorithm: it is directed at inference at the current time index, not over the whole past and not over the parameters of the model for, else how could we consider the algorithm has converged when it degenerates to a single path at some finite horizon the past? Pierre has also written a further comment of the paper on Statistfaction.

thesis in Marseille

Posted in Books, Statistics, Travel, University life with tags , , , , , on July 4, 2013 by xi'an


Today, I went to Marseille for a PhD thesis defence: I biked to the RER train station (yay!) and the early (7am) flight was smooth, with clear views of nuclear plants along the way… I had previously and critically refereed the thesis, called “Essays on on the econometrics of inequality and poverty measurements” ; despite its strongly applied economics title it indeed was primarily an econometric work about mixtures and quantile regression. The thesis author and PhD incumbent Abdoul Aziz Ndoye being from Senegal, he had prepared a buffet after the defence with Senegalese (yummy) delicacies that I definitely enjoyed after such an early (Oat Squares, thanks to A&C.!) breakfast. (Actually, Aziz had presented a poster in Kyoto so some of you may have met him already!) The afternoon train ride to Montpelier was smooth as well, with nice views of Provençal villages along the way. (Too bad the train line does not stick more to the coastline, though.)

While the part on mixtures was rather traditional (still using Chib’s approach to evaluate marginal likelihoods and decide about the number of components in the mixture, while “resolving” the label switching problem by using assymmetric priors based on the sample quantiles [ok, “priors”!]), I got more interested in the quantile regression part. Maybe because quantile regression is mostly new to me, I have some difficulties in getting the motivation for (regular) quantile regression: I would see an estimation of the whole conditional cdf as linear in the regressor as a more natural goal than picking one or several probability levels to estimate the corresponding quantile. Also, the thesis follows an alternative approach called RIF where the density of the observables y is first estimated by a mixture of (log-)normals and then a quantile regression is operated on


reintroducing the explanatory variables after estimating a joint density on the y’s, which puzzles me as well. (Note that this part of the thesis, written jointly with Michel Lubrano, got a Best Presentation Prize at the Scottish Economics meeting in 2012 and got published in the associated journal.) Overall, this is an innovative and interesting piece of work, even though it cannot be completely envisionned as a Bayesian resolution.

trip to Besançon (and the stars)

Posted in Mountains, pictures, Statistics, Travel, University life with tags , , , , , , , , , on November 26, 2012 by xi'an

Today, I made a quick TGV trip to Besançon, in French Jura, to give a seminar to astronomers and physicists, in connection with the Gaia project I had mentioned earlier. I gave my talk straight out of the train and then we started discussing MCMC and ABC for the astronomy problems my guests face. To my surprise, I discovered that they do run some local form of ABC, using their own statistics and distances to validate simulation from the (uniform) prior on their parameter space. The discussion went far enough to take a peek under the hood, namely to look at some Fortran programs they are running (and make suggestions for acceleration and adaptation). It is quite interesting to see that ABC is actually a natural approach when people face complex likelihoods and that, while they construct appropriate tools, they feel somehow uncertain about the validation of those methods and are unaware of very similar tools in other fields. In addition to this great day of exchange, I had several hours of freedom in the train (and a plug) to work on the bayess package for Bayesian Essentials (not dead yet!). Here are my slides, pot-pourri of earlier talks. (Including the one on cosmology model choice in Vancouver.)