[Forum SIS] Avviso seminario Alessio Farcomeni
Luca Greco
luca.greco a unisannio.it
Gio 4 Ott 2018 11:36:44 CEST
Buongiorno,
chiedo di dare diffusione del seguente avviso di seminario.
We are glad to announce the following Statistics Seminar:
Tuesday, October 16th, 4:15 pm
Department DEMM - University of Sannio - Benevento
Sala Consiglio, Piazza Arechi II, Palazzo De Simone, Benevento
Prof. Alessio Farcomeni from Sapienza University of Rome
Trimming and wild trimming procedures for robust clustering with mild
and gross outliers
Abstract
Trimming principles play a central role in robust statistics. In this
seminar we recap the main uses of impartial trimming for robust
cluster analysis, with special attention to three recent developments.
In the first part we introduce a trimmed mixture of contaminated
Gaussian densities, which is useful for increasing efficiency by
incorporating mild outliers into the estimation set. A penalized
likelihood procedure with a theoretically optimal penalty parameter
allows us to select the optimal trimming proportion, also in the more
classical case of Gaussian mixtures (tclust). We then remove the
necessity of selecting a trimming parameter in advance by using an
iteratively reweighted version of the procedure, initialized from a
high trimming level and then refined as possible. The resulting
procedure is highly efficient despite a breakdown point corresponding
to the initial trimming level. Finally, we remove the necessity of
specifying both the trimming parameter and the number of clusters
through an iterative clustering procedure rooted in the Forward Search
approach, which replaces the original K-population (robust) estimation
problem with K distinct one-population steps. The procedure is
grounded on novel theoretical results concerning wild trimming, that
is, trimming more than 50% of the observations. The three methods are
illustrated on an original data example concerning the identification
of source from illicit drug shipments seized in Italy and Spain.
Based on joint works with Andrea Cerioli, Francesco Dotto, Luis-Angel
Garcìa-Escudero, Agustìn Mayo-Iscar, Antonio Punzo, Marco Riani.
--
Luca Greco
Ricercatore SECS/S-01 - Statistica
Università degli Studi del Sannio
tel: 0824 305245
url: https://sites.google.com/site/lucagrecohomepage/
----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.
Maggiori informazioni sulla lista
Sis