Van Dantzig Seminar

nationwide series of lectures in statistics

Home      David van Dantzig      About the seminar      Upcoming seminars      Previous seminars      Slides      Contact    

Next Van Dantzig Seminar: Friday, 1 December 2023

Speakers: Sofia Olhede, Björn Sprungsk

Location: TU Delft, Snijderszaal (EWI, LB 01.010)

15.00 - 16.00 Sofia Olhede
16.00 - 16.15 Break
16.15 - 17.15 Björn Sprungk
17.15 Drinks

Sofia Olhede - On Graph Limits as Models for Interaction Data

Network data has become a staple in many different applications, ranging from ecology, to neuroscience and systems biology. Its inference will of course depend on the application where we collect the network data, but I will discuss some general principles based on probabilistic symmetries such as permutation invariance. Just like other probabilistic invariances, the distributional invariance to permuting indices of a matrix of interactions implies a representation theorem (the Aldous-Hoover theorem). This representation is in terms of a graph limit function, or graphon. I will discuss the representation, how to make inferences based on this representation, what to do if distributional permutation invariance does not hold, and what to do if we have additional information such as time stamp of interactions, multiple interactions or additional covariate data.

Björn Sprungk - Noise-level robust sampling and Bayesian inference on the sphere

We consider two topics in this talk: the first is related to sampling from concentrated posterior distributions arising in Bayesian inference with informative data. Although a desirable situation from an inference perspective concentrated posteriors pose a computational challenge for many sampling algorithms. We present results regarding Markov chain Monte Carlo methods based on the Laplace approximation which show a statistical efficiency independent of the concentration of the posterior under suitable assumptions.

The second topic again considers the construction of Markov chain Monte Carlo algorithms but this time for dimension-independent sampling from posterior measures defined on a high-dimensional sphere as occuring in Bayesian density estimation. Both topics are related by the concept of pushforward Markov kernels.

Van Dantzig Seminar

The Van Dantzig seminar is a nationwide series of lectures in statistics, which features renowned international and local speakers, from the full width of the statistical sciences. The name honours David van Dantzig (1900-1959), who was the first modern statistician in the Netherlands, and professor in the "Theory of Collective Phenomena" (i.e. statistics) in Amsterdam. The seminar will convene 4 to 6 times a year at varying locations, and is supported financially by among others the STAR cluster and the Section Mathematical Statistics of the VVS-OR.

Supported by