Archive for the Books Category

the calculating stars [book review]

Posted in Books, Travel with tags , , , , , , , , , , , , , , on October 16, 2021 by xi'an

This fist sounded like an interesting attempt at alternate history, when a massive meteor strike obliterating the Washington DC region in 1952 forced the World to change shift towards space exploration and the eventual evacuation of Earth. The story is told from a computer (or computress) viewpoint, who is a wunderkid, a mathematician, a physicist, a war (WASP) pilot, and more, with a strong will and an independent mind, hoping to become a female astronaut. If the setting reminds you of Hidden figures, a (great) movie about the true story of NASA black female mathematicians, it is no surprise and I wonder how much inspiration the author got from these historical facts, if not from the 2016 book itself. Despite receiving many awards, like the Hugo, Nebula, and Locus awards!, The Calculating Stars is somewhat of a disappointment to me, because of the highly single-minded perspective,  where everything (related to solving the forecast extinction) seems to happen with a small group of people, because of the confusion between a mathematician and someone who can do complex arithmetics by head, to the near-perfection of the central character, who can also hotwire a car, because of the anachronisms, incl. the prescience that the asteroid crash was going to cause a deadly rise of temperatures when the dinosaur extinction was not yet linked with a similar event, because of a rosy depiction of the World uniting towards racing against the Great Extinction, and, cherry on the pie, because French sentences found throughout the book mostly make no sense as literal translations of English sentences!

“Elle va le faire mais Dieu sait ce qu’elle va parler.” [She’s going to do it but God knows what she’s going to say.]

“Il est l’ordre naturel je pense (…) Il n’y a rien de naturel.” [It’s the natural order of things I think (…) Nothing is natural.]

“Ce ne fut pas une explosion ou nous aurions senti.” [It wasn’t a blast or else we would have felt.]

GANs as density estimators

Posted in Books, Statistics with tags , , , , , , , on October 15, 2021 by xi'an

I recently read an arXival entitled Conditional Sampling With Monotone GAN by Kovakchi et al., who construct  a mapping T that transforms or pushes forward a reference measure þ() like a multivariate Normal distribution to a target conditional distribution ð(dθ|x).  Which makes the proposal a type of normalising flow, except it does not require a Jacobian derivation… The mapping T is monotonous and block triangular in order to be invertible. It is learned from data by minimising a functional divergence between Tþ(dθ) and ð(dθ|x), for instance GAN least square or GAN Wasserstein penalties and representing T as a neural network.  Where monotonicity is imposed by a Lagrangian. The authors “note that global minimizers of [their GAN criterion] can also be used for conditional density estimation” but I fail to understand the distinction in that once T is constructed, the estimated conditional density is automatically available. However my main source of puzzlement is at the worth of this construction, since it does not provide an exact generative process for the conditional distribution, while requiring many generations from the joint distribution. Rather than a comparison with MCMC, which is not applicable in untractable generative models, a comparison with less expensive ABC solutions would have been appropriate, I think. And the paper is missing any quantification on the quality or asymptotics of the density estimate provided by this involved approximation, as most of the recent literature on normalising flows and friends. (A point acknowledged by the authors in the supplementary material section.)

“In this regard, the MGANs approach introduced in the article belongs to the category of sampling techniques such as MCMC, whose goal is to generate independent samples from the law of y|x, as opposed to assuming some structural form of the probability measure directly.”

I am unsure I understand the above remark as MCMC methods are intrinsically linked with the exact probability distribution, exploiting either some conditional representations as in Gibbs or at the very least the ability to compute the joint density…

 

ensemble Metropolis-Hastings

Posted in Books, Kids, Statistics with tags , , , , , on October 14, 2021 by xi'an

A question on X validated about ensemble MCMC samplers had me try twice to justify the Metropolis-Hasting ratio the authors used. To recap, ensemble sampling moves a cloud of points (just like our bouncy particle sampler) one point X at a time by using another point Z as a pivot or origin and moving randomly X along the line [XZ]. In the paper,  the distribution of the rescaling is symmetric in the sense that f(z)=f(1/z). I indeed started by perceiving the basic step of the sampler as a Metropolis-within-Gibbs step along a random direction. But it did not work as the direction depends on the current X. I then wondered at a possible importance sampling interpretation compensating for the change of scale, but it was leading to the wrong power anyway. Before hitting the fact that this was actually a change of radius in the space with origin Z, leaving the angular coordinates invariant. Which explained for the power (n-1) in the Metropolis ratio, in agreement with a switch to polar coordinates.

COVID by numbers [not a book review]

Posted in Books, Kids, Statistics with tags , , , , , , , , , on October 13, 2021 by xi'an

David Spiegelhalter and Anthony Masters have made a book out of their COVID related columns in The Observer. Here are ten key figures extracted from that book:

  1. The UK was hit by more than 1,000 separate outbreaks (…) [with] far more imports of Sars-CoV-2 from France, Italy and Spain than from China
  2. Reported Covid deaths depend on the day of the week (due to delayed reporting, and a weekend effect, but smoothing is very rarely applied)
  3. In the first year of Covid, over-90s had 35,000 times the risk of dying of Covid-19 as young children (with no relevance of the figure per se since an extra death of a young child would have moved it from 35,000 to 32,000, since there were thankfully so few deaths of young children)
  4. 2020 saw the highest number of deaths since 1918 in England and Wales (even when correcting for population increase or population ageing)
  5. The UK has led the World in testing Covid treatments (like dexamethasone and hydroxychloroquine, thanks to the centralised NHS, making me wonder why France with another centralised and public health structure was not able to do the same)
  6. People who have died with Covid have on average lost about 10 years of life (contrary to the authors’ intial hunch, and mine as well, to oppose to the less relevant loss of life expectancy across the entire population)
  7. Most people died “of” Covid rather than “with” it, but most have also had other medical conditions (with 91% of pre-COVID conditions)
  8. Alcohol consumption stayed the same during lockdown (which came as a surprise, given the general feeling for the opposite, and still as a worrying indicator of alcoholism)
  9. Most people with Sars-CoV-2 don’t infect anyone (which would need more details, as the figure should be weighted by the base probability to infect someone)
  10. The pandemic has been a net lifesaver for young people (with 300 fewer deaths for 15-29 year old, but it also has had a potentially negative impact on their life expectancy).

congrats, Dr. Clarté!

Posted in Books, pictures, Statistics, Travel, University life with tags , , , , , , , , , , on October 9, 2021 by xi'an

Grégoire Clarté, whom I co-supervised with Robin Ryder, successfully defended his PhD thesis last Wednesday! On sign language classification, ABC-Gibbs and collective non-linear MCMC. Congrats to the now Dr.Clarté for this achievement and all the best for his coming Nordic adventure, as he is starting a postdoc at the University of Helsinki, with Aki Vehtari and others. It was quite fun to work with Grégoire along these years. And discussing on an unlimited number of unrelated topics, incl. fantasy books, teas, cooking and the role of conferences and travel in academic life! The defence itself proved a challenge as four members of the jury, incl. myself, were “present remotely” and frequently interrupted him for gaps in the Teams transmission, which nonetheless broadcasted perfectly the honks of the permanent traffic jam in Porte Dauphine… (And alas could not share a celebratory cup with him!)

%d bloggers like this: