Please note that updated Statistical Laboratory seminar lists are shown
on our web page at
http://www.statslab.cam.ac.uk/Dept/seminars.html
SEMINARS
Room 27, Statistical Laboratory
University of Cambridge
Friday, 7 May 1999 (TWO seminars)
------------------
2.05 Anthony Ledford (Surrey)
DEPENDENCE WITHIN EXTREMES
In this talk I will discuss the classical approach for modelling
dependence between the extreme observations of a multivariate process
and then will examine an alternative framework that provides a
richer class of dependence models. Associated techniques for examining
dependence within time series extremes will be presented. Statistical
methods based on these techniques will be illustrated by examining some
environmental and financial data sets.
3.30 Guenther Walther (Stanford)
ON MIXTURES AND PENALIZED MAXIMUM LIKELIHOOD
There exist a number of nonparametric approaches to assess the evidence
that a mixture of several distributions is present, in the case where
only nonparametric assumptions on the class of compoenent distributions is
appropriate: looking for bimodality or `bumps', for example.
I will present some cogent reasons why it is preferable to use
a criterion that is not derived from shape properties of the density
itself, and will introduce instead a special semiparametric model.
This approach will employ a novel use of penalized maximum
likelihood methods and will provide some new insights into what
penalized maximum likelihood does for the statistician.
-----------------------------------------------------------------------
Friday, 14 May
--------------
2.05 David Denison (Imperial College)
BAYESIAN TREE-BASED CLASSIFICATION METHODS
In this talk I shall describe an algorithm for finding "good" classification
trees. It uses Bayesian methods as a tool to perform a
stochastic search of the space of tree structures. Generation from
the full posterior of this space is not possible due to the hierachical
nature of the basis functions that define a tree. We shall discuss this
problem as well as other related models which attempt to overcome the
difficulties of tree-based methods,
such as Bayesian MARS and Bayesian Partition Models.
-----------------------------------------------------------------------
Thursday, 20 May
----------------
2.05 Stephen Walker (Imperial College)
CONSTRUCTING STATIONARY TIME SERIES MODELS
AND MARKOV INCREMENT PROCESSES
I will show how to construct first-order
stationary time series models based on exponential
family and the development to Markov increment processes;
generalising independent increment (Levy) processes
such as the beta and gamma.
*********************************
ALL INTERESTED ARE WELCOME
*********************************
Organizer: Susan Pitts
UNIVERSITY OF CAMBRIDGE
DEPARTMENT OF PURE MATHEMATICS AND MATHEMATICAL STATISTICS
STATISTICAL LABORATORY
16 MILL LANE, CAMBRIDGE CB2 1SB
Tel: (01223) 337958
Fax: (01223) 337956
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|