Meeting in Mathematical Statistics

MMS 2018 - December 16 - 21
-----------------------------------------------

Home

Participants

Program

Practical Information

 

!!!! Registration is open - Please register before November 30, 2018 !!!!!

Home

Welcome to the 2018 edition of the Meeting in Mathematical Statistics. It is our pleasure to announce that this year the workshop will take place in the south of France, at a NEW PLACE:

AZUREVA, Village Vacances - Frejus, France

 

Tutorials


Title -
Outlier-robust estimation of a high-dimensional mean vector

Abstract: In these lectures, we will review some recent results on the estimation of the mean of a Gaussian distribution when the sample contains outliers. Different mathematical frameworks for modeling outliers will be presented. Then, we will show that in a p-dimensional problem with s outliers out of n observations, the minimax quadratic risk is at least of order (p/n) + (s/n)^2. This turns out to be the optimal rate, since it is achieved by the Tukey's median. We will then present some computationally tractable estimators, which have provably better rates of convergence than the standard estimators such as the (coordinatewise or geometric) median.

 


Title -
Spectral method, Lloyd Algorithm, EM, and Variational Bayes for Clustering

Abstract:  In this talk, I will review some recent attempts to understand spectral method, Lloyd algorithm, EM, and variational Bayes for clustering. Our goal is to have a more or less complete answer to the question whether and when each of the four algorithms attains the optimal statistical accuracy for Gaussian mixtures. This is a joint work with Matthias Löffler, Yu Lu, Tal Sarig, Yihong Wu, and Anderson Zhang.




Organizing Committee





Sponsors

   CREST - Statistique    Ecodec     ANR