València summer school

In another continuation of the summer of Bayesian conferences in Europe, the Universidat de Valencià is organising a summer school on Bayesian statistics, from 16 July till 20 July, 2018. Which thus comes right after our summer school on computational statistics at Warwick. With a basic course on Bayesian learning (2 days). And a more advanced course on Bayesian modeling with BayesX. And a final day workshop.

journée algorithmes stochastiques à Dauphine vendredi

A final reminder (?) that we hold a special day of five talks around stochastic algorithms at Dauphine this Friday, from 10:00 till 17:30. Attendance is free, coffee and tea are free (while they last!), come and join us!

a Ca’Foscari [first Italian-French statistics seminar]

Apart from subjecting my [surprisingly large!] audience to three hours of ABC tutorial today, and after running Ponte della la Libertà to Mestre and back in a deep fog, I attended the second part of the 1st Italian-French statistics seminar at Ca’Foscari, Venetiarum Universitas, with talks by Stéfano Tonellato and Roberto Casarin. Stéfano discussed a most interesting if puzzling notion of clustering via Dirichlet process mixtures. Which indeed puzzles me for its dependence on the Dirichlet measure and on the potential for an unlimited number of clusters as the sample size increases. The method offers similarities with an approach from our 2000 JASA paper on running inference on mixtures without proper label switching, in that looking at pairs of allocated observations to clusters is revealing about the [true or pseudo-true] number of clusters. With divergence in using eigenvalues of Laplacians on similarity matrices. But because of the potential for the number of components to diverge I wonder at the robustness of the approach via non-parametric [Bayesian] modelling. Maybe my difficulty stands with the very notion of cluster, which I find poorly defined and mostly in the eyes of the beholder! And Roberto presented a recent work on SURE and VAR models, with a great graphical representation of the estimated connections between factors in a sparse graphical model.

Journée algorithmes stochastiques

On December 1, 2017, we will hold a day workshop on stochastic algorithms at Université Paris-Dauphine, with the following speakers

• Rémi Bardenet – CNRS Lille / CRISTAL [10:00]
• Nicolas Chopin – ENSAE / CREST [11:00]
• Aymeric Dieuleveut – ENS / DI & INRIA [14:00]
• Aude Genevay – Dauphine / CEREMADE & INRIA [15:00]
• Pierre Monmarché – UPMC / LJLL [16:30]

 Details and abstracts of the talks are available on the workshop webpage. Attendance is free, but registration is requested towards planning the morning and afternoon coffee breaks. Looking forward seeing ‘Og’s readers there, at least those in the vicinity!

And while I am targetting Parisians, crypto-Bayesians, and nearly-Parisians, there is another day workshop on Bayesian and PAC-Bayesian methods on November 16, at Université Pierre et Marie Curie (campus Jussieu), with invited speakers

• Arnak Dalalyan
• Alain Durmus
• Peter Grunwald
• Stéphanie van der Pas
• James Ridgway
• Elodie Vernet

and a similar request for (free) registration.

conference carbon footprint

As a local organiser of the recent BNP 11 conference in Paris, and hence involved in setting and cleaning coffee breaks and [now famous] wine&cheese poster sessions, I was rather shocked by the amount of waste generated by those events, albeit aware of the importance of the social exchanges they induced… And thus got to wonder how the impact of those conference events could be reduced. One solution is the drastic one, namely to provide exactly nothing at all during the breaks between talks and expect anyone hungry or thirsty enough to bring one own’s food or drink. Another one, as suggested by my daughter at the dinner table, is to provide Ecocups, namely reusable plastic glasses that can given to all participants at the beginning of the conference. Or sold (or rented) to those who have not brought their own mug or bottle. (Of course, this may be a poor idea in that manufacturing and shipping a hard-plastic glass that most likely will be discarded after a few days may be more damaging than producing the equivalent number of “disposable” thin plastic glasses. And in the end all this agitation is peanuts compared with the impact of flying participants to the conference. For which I have no handy solution… As biking to the conference location is a privilege very few can enjoy.) Still, and even though this puts another stone in the already rocky organisers’ garden, I wish we could adopt more positive policies at the meetings we organise and sponsor.

oxwasp@amazon.de

The reason for my short visit to Berlin last week was an OxWaSP (Oxford and Warwick Statistics Program) workshop hosted by Amazon Berlin with talks between statistics and machine learning, plus posters from our second year students. While the workshop was quite intense, I enjoyed very much the atmosphere and the variety of talks there. (Just sorry that I left too early to enjoy the social programme at a local brewery, Brauhaus Lemke, and the natural history museum. But still managed nice runs east and west!) One thing I found most interesting (if obvious in retrospect) was the different focus of academic and production talks, where the later do not aim at a full generality or at a guaranteed improvement over the existing, provided the new methodology provides a gain in efficiency over the existing.

This connected nicely with me reading several Nature articles on quantum computing during that trip,  where researchers from Google predict commercial products appearing in the coming five years, even though the technology is far from perfect and the outcome qubit error prone. Among the examples they provided, quantum simulation (not meaning what I consider to be simulation!), quantum optimisation (as a way to overcome multimodality), and quantum sampling (targeting given probability distributions). I find the inclusion of the latest puzzling in that simulation (in that sense) shows very little tolerance for errors, especially systematic bias. It may be that specific quantum architectures can be designed for specific probability distributions, just like some are already conceived for optimisation. (It may even be the case that quantum solutions are (just next to) available for intractable constants as in Ising or Potts models!)

India snapshop [jatp]