[ibayesclub.beta] Bayes Club, 19th of March, from 2 pm (social program afterward the talks)

[Szabo, B.T.]

Dear  Colleagues,

The next (International) Bayes Club (https://www.math.vu.nl/thebayesclub/) meeting will take place on the 19th of March (Friday) from 2 pm. We have two speakers: Joris Bierkens (Delft) and Ankur Moitra (MIT). This is a special occasion as we celebrate our 10th year anniversary. Therefore we will book a virtual meeting place at gather.town from 4pm so we can chat and toast virtually (please prepare with beers/wine/...). Would be great if everyone could arrange to stay for a while after the talks as well. Please find below the titles and the abstracts. The seminar will be on zoom:

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

Topic: International Bayes club
Time: Mar 19, 2021 02:00 PM Amsterdam

Join Zoom Meeting
https://vu-live.zoom.us/j/91629816385?pwd=T1RKLzRxTWw4N2hOcXYrb2ppMUh0UT09

Meeting ID: 916 2981 6385
Passcode: 904232


--------------------------------------
2pm lecture by Joris Bierkens (TU Delft)

Title: The Boomerang Sampler and Bayesian inference for diffusions

Abstract: The Bayesian estimation of (parameters of) a diffusion process is
challenging due to the lack of a closed form expression for the
transition densities. A natural approach is employing a Gibbs sampler in
which the diffusion process has to be simulated for a fixed choice of
parameters as a latent variable.

This simulation of a specified diffusion is challenging itself due to
its infinite dimensional and nonlinear nature. We discuss a recent
approach to this problem involving the Boomerang Sampler, which is an
example of a piecewise deterministic Monte Carlo sampler. The Boomerang
Sampler has important differences as well as similarities with other
piecewise deterministic samplers such as the Zig-Zag process and the
Bouncy Particle Sampler.

This talk is based upon joint work with Sebastiano Grazzi, Kengo
Kamatani, Gareth Roberts, Moritz Schauer and Frank van der Meulen.

----------------
3pm lecture by Ankur Moitra (MIT)

Title: An Invitation to Semi-Random Models for Bayesian Statisticians

Abstract: The stochastic block model is one of the oldest and most ubiquitous models for studying clustering and community detection. The basic inference problem is to use the observed graph to estimated the planted community structure that was used to generate it. Exciting recent work has precisely characterized when it is possible to achieve non-trivial accuracy, and what the optimal error behaves like. These works all revolve around belief propagation which attempts to sample from the posterior distribution using a simple iterative algorithm.

Here we revisit these thresholds from the perspective of semirandom models where we allow an adversary to make `helpful’ changes that strengthen ties within each community and break ties between them. We show a surprising result that these `helpful’ changes can shift the information-theoretic threshold, making the community detection problem strictly harder. Conceptually, any algorithm meeting the exact information-theoretic threshold in the average-case model must exploit the precise structure of the noise. These results point to an interesting new direction: Can semirandom (and related) models help explain why some algorithms are preferred to others in practice, in spite of gaps in their average-case performance?

This is based on joint work with Amelia Perry and Alex Wein.

Best wishes,
Botond (on behalf of the organising team)


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://listserver.vu.nl/pipermail/ibayesclub.beta/attachments/20210303/088d868b/attachment.html>