Archive for University of Oxford

improving bridge samplers by GANs

Posted in Books, pictures, Statistics with tags , , , , , , , on July 20, 2021 by xi'an

Hanwen Xing from Oxford recently posted a paper on arXiv about using GANs to improve the overlap bewtween the densities in bridge sampling. Bringing out new connections with noise contrastive estimation. The idea is to optimise a transform of one of the densities h() to bring it closer to the other density k(), using for instance normalising flows. (The call to transforms for bridge is not new, dating at least to Voter in 1985, the year I was starting my PhD!) Furthermore, using an f-divergence as a measure of functional distance allows for a reasonably straightforward update of the transform. That can be reformulated as a GAN target, which is somewhat natural in that the transform aims at confusing simulation from the transform of h and from k. This is quite an interesting proposal,  even though calculating the optimal transform is time-consuming and subjet to the curse of dimensionality. I also wonder at whether or not iterating the optimisation, one density after the other, would be bring further improvement.

Conformal Bayesian Computation

Posted in Books, pictures, Statistics, University life with tags , , , , , on July 8, 2021 by xi'an

Edwin Fong and Chris Holmes (Oxford) just wrote a paper on Bayesian scalable methods from a M-open perspective. Borrowing from the conformal prediction framework of Vovk et al. (2005) to achieve frequentist coverage for prediction intervals. The method starts with the choice of a conformity measure that measures how well each observation in the sample agrees with the sample. Which is exchangeable and hence leads to a rank statistic from which a p-value can be derived. Which is the empirical cdf associated with the observed conformities. Following Vovk et al. (2005) and Wasserman (2011) Edwin and Chris note that the Bayesian predictive itself acts like a conformity measure. Predictive that can itself be approximated by MCMC and importance sampling (possibly smoothed by Pareto). The paper also expands the setting to partial exchangeable models, renamed group conformal predictions. While reluctant to engage into turning Bayesian solutions into frequentist ones, I can see some worth in deriving both in order to expose discrepancies and hence signal possible issues with models and priors.

Metropolis-Hastings via Classification [One World ABC seminar]

Posted in Statistics, University life with tags , , , , , , , , , , , , , , , on May 27, 2021 by xi'an

Today, Veronika Rockova is giving a webinar on her paper with Tetsuya Kaji Metropolis-Hastings via classification. at the One World ABC seminar, at 11.30am UK time. (Which was also presented at the Oxford Stats seminar last Feb.) Please register if not already a member of the 1W ABC mailing list.

two ISBA meetings in 2022

Posted in pictures, Statistics, Travel, University life with tags , , , , , , , , , , , , , on May 11, 2021 by xi'an

As in 2019, both the O’Bayes and BNP conferences will occur the same year, if not back-to-back as in 2019 (when they were in neighbouring Oxford and Warwick, respectively). To quote from the current Chair of the OBayes section of ISBA, my friend Gonzalo,

“the next International Workshop on Objective Bayes Methodology (O-Bayes, OBayes, O’Bayes, Ohhh Bayes,..) is scheduled for September 2022 from 7th (Wed) to 10th (Sat) and will be hosted by University of California, Santa Cruz. This will be the 14th meeting of one of the longest-running and preeminent meetings in Bayesian statistics (the 1st was in USA 1996; the last one in 2019 in UK). In this conference, we will be celebrating the 70th birthday of Luis Pericchi an extraordinary person who has been very influential in the successful development of OBayesian ideas.”

thus seeing the O’Bayes meeting taking place in North America in early Fall, followed by BNP 13 in South America a month later (and thus Spring!), quoting from Alessandra Guglielmi:

“Due to the ongoing COVID-19 pandemic, the 13th World Meeting on Bayesian Nonparametrics (BNP Workshop) is postponed until 2022. The meeting is currently scheduled to take place in Puerto Varas, Chile, October 24-28, 2022.”

El lago Llanquihue, con el volcán Osorno al fondo Yuri de Mesquita Bar / Getty Images/iStockphoto

 

on Astra and clots

Posted in Books, Kids, pictures, Statistics with tags , , , , , , , , , , , , on March 16, 2021 by xi'an

A tribune this morning in The Guardian by David Spiegelhalter on having no evidence that the Oxford/AstraZeneca vaccine causes blood clots.

“It’s a common human tendency to attribute a causal effect between different events, even when there isn’t one present: we wash the car and the next day a bird relieves itself all over the bonnet. Typical.”

David sets the 30 throboembolic events among the 5 million people vaccinated with AstraZeneca in perpective of the expected 100 deep vein thromboses a week within such a population. Which coincides with the UK’s Medicines and Healthcare Products Regulatory Agency statement that the blood clots are in par with the expected numbers in the vaccinated population. (The part of the tribune about the yellow card reports, based on 10 million vaccinated people, reiterates the remark but may prove confusing to some!) As for hoping for a rational approach to the issue,  … we would need a different type of vaccine, far from being available! As demonstrated by the decision to temporarily stop vaccinating with this vaccine, causing sure additional deaths in the coming weeks.

“Will we ever be able to resist the urge to find causal relationships between different events? One way of doing this would be promoting the scientific method and ensuring everyone understands this basic principle. Testing a hypothesis helps us see which hunches or assumptions are correct and which aren’t. In this way, randomised trials have proved the effectiveness of some Covid treatments and saved vast numbers of lives, while also showing us that some overblown claims about treatments for Covid-19, such as hydroxychloroquine and convalescent plasma, were incorrect.”