The registration for the incoming, exciting, Bayes Comp 2025 conference (and its satellites) is now open, including information regarding accommodations for the conference. Early bird rates run till 15 March. Furthermore, the call for contributed talks is open till 17 February.
My final exam for the Monte Carlo course I taught last semester proved too much of a challenge for my fourth year students, despite being rather elementary and centred on accept-reject algorithms and importance/bridge sampling. One of the problems was a decomposition of the truncated Normal simulation method proposed by Marsaglia in 1963, found on X Validated!, which I had posted on the course Teams forum (and discussed on the ‘Og!). Obviously overlooked by the students as hardly anyone solved the central question… Quite a disappointment (and more exams to grade for the resit exam in June!)
The ICSDS 2024 meeting in Nice is quite impressive and not primarily because it is in Nice under a beautiful December sun. As other (numerous) IMS meetings I attended (since the initial one in Uppsala in 1990!), the program is of high quality and along topics that are currently moving fast or emerging. From the sessions I attended, e-values are strongly represented, although it remains unclear to me why they should constitute a major departure from p-values, as they stick to hypothesis testing, Type I error, power, and the whole paraphernalia of Neyman-Pearson formalism. If I manage to attend a BIRS workshop on the subject next Summer, I may manage to get a better e-derstanding!
The MCMC (only!) session included a presentation by Guanyang Wang that generalised different approximate MCMC schemes into a unified one. And one by Filippo Ascolani on Gibbs beating the competition! I also attended the Bayesian prediction session, where my friends Sonia Petrone and Chris Holmes have presentations on their respective Series B papers. I discussed both on the ‘Og, on 15 March 2023 and 07 November 2022, respectively. This time, I found that both talks had a Bayesian bootstrap flavour, which is not surprising when considering the non-parametric nature of the approach. And they left me wondering at it being protected from overfitting.
My only plenary session was Cynthia Dwork’s on outcome indistinguishability, which, while related to the privacy topics I was topic, remained somewhat obscure as to its purpose. Meaning I have to get through the paper to get a more holistic perspective.
Of course, Nice in Winter is a very nice place, with the waterfront available for running an uninterrupted 15km as we found out with Jérémie Houssineau (at a brisk 4’09” pace I had not planned before starting!) and the sea all for myself (for a dozen minutes before losing digits!). Unfortunately I had to skip the final day due to examinations of the Paris Dauphine MASH master. And miss Stan receiving a student award. But I am looking forward the next iterations of ICSDS. (Not including Copenhagen, Madrid and many many other places in 2025, since ICSDS seemed a most common name for conferences, some presumably predatory! The true location is Sevilla, to keep up with the Mediterranean theme of ICSDS!)
 | David on Bayesian workflow [book review… |
 | Ewan on Tidy up, Claude! |
 | xi'an on Tidy up, Claude! |
| Ewan on Tidy up, Claude! |
| Ewan on Já við viðræðum. Já við Evrópu… |