[ibayesclub.beta] Bayes club - this Thursday 3pm-5pm

[Szabo, B.T.]

Dear All,


This is a kind reminder that the next meeting of the (International) Bayes club will take place on the 3rd of December (this Thursday) from 3 pm until 5 pm. We will have two speakers: Sergios Agapiou (U Cyprus) and Pierre Jacob (Harvard), please find below the titles and the abstracts of the talks. The seminar will be on zoom:


Topic: The (International) Bayes Club
Time: Dec 3, 2020 03:00 PM Amsterdam

https://vu-live.zoom.us/j/93507519282

 Meeting ID: 935 0751 9282
Passcode: d7eHf@

-----------------------------------------------------


3pm Speaker: Sergios Agapiou (U Cyprus)
Title: Gauss versus Laplace rates of contraction under Besov regularity

We will discuss recent results on rates of contraction with a family of priors with tails between Laplace and Gaussian, termed p-exponential priors. We will focus on the white noise model and will discuss upper bounds on the rate of contraction under Besov regularity of the truth, in L_2-loss. We will use alpha-regular priors and will see that Laplace priors achieve the same and often better rates than Gaussian ones. In particular, we will see that for spatially inhomogeneous unknown functions, that is functions which are smooth in some areas but rough in other areas, Gaussian priors appear to be suboptimal. On the other hand Laplace priors achieve better rates, which can be minimax when the prior is appropriately calibrated.

This is joint work with Masoumeh Dashti and Tapio Helin
https://arxiv.org/abs/1811.12244 (to appear in Bernoulli).


4pm Speaker: Pierre Jacob (Harvard)
Title: A Gibbs sampler for a class of random convex polytopes

Abstract: We present a Gibbs sampler for the Dempster--Shafer (DS) approach to statistical inference in the setting of for Categorical distributions, with arbitrary numbers of categories and observations. The DS framework extends the Bayesian approach, allows in particular the use of partial prior information, and yields three-valued uncertainty assessments (p,q,r) representing probabilities "for", "agains", and "don't know" about formal assertions of interest.  However DS gives rise to computational challenges even in settings as classical and seemingly simple as Categorical distributions for count data.  The proposed algorithm targets the distribution of a class of random convex polytopes which encapsulate the DS inference. The validity of the sampler relies on an equivalence between the iterative constraints of some vertex configuration and the non-negativity of cycles in a fully connected directed graph. Experiments illustrate the output of the algorithm in the setting of 2x2 contingency tables.

This is joint work with Ruobin Gong, Paul T. Edlefsen & Arthur P. Dempster and a technical report is available at https://arxiv.org/abs/1910.11953

Best wishes,
Botond
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://listserver.vu.nl/pipermail/ibayesclub.beta/attachments/20201201/31eeaad4/attachment.html>