[Statlist] Talk tomorrow on ``Adaptive interpolation method for mutual information in inference on dense and sparse graph'' Datum: 27. November 2019 um 15:49:05 MEZ

[Maurer Letizia]

Date: Thursday, November 28, 2019
Time: 15:00 hrs
Room: HG G 19.1, ETH Zurich

Speaker: Chun Lam Chan, EPFL

Title: Adaptive interpolation method for mutual information in inference on dense and sparse graphs

Abstract: Mutual information is often the central quantity when we study the fundamental limit in a Bayesian inference problem. The asymptotic mutual information is predicted to admit a single-letter variational expression derived from the replica or cavity method from statistical physics. Recently, adaptive interpolation method has been developed as a simple and versatile scheme to prove this prediction for various Bayesian inference problems defined on dense factor graphs. In this talk, I will present some new development of the adaptive interpolation; in particular, how to adopt the methods for models in community detection. I will discuss stochastic block models in the dense graph regime and censored block models in the sparse graph regime.


Afonso BandeiraProfessor of Mathematics, ETH Zurich
https://people.math.ethz.ch/~abandeira/

HG G 23.1, Ramistrasse 101, 8092 Zurich, Switzerland
Administration: Annette Ryter annette.ryter using ifor.math.ethz.ch<mailto:annette.ryter using ifor.math.ethz.ch>

	[[alternative HTML version deleted]]