UNIVERSITY OF GLASGOW
STATISTICS SEMINAR PROGRAMME
Wednesday, 13th October, 3pm
Bayesian partition models
David DENISON (Imperial College)
Wednesday, 20th October, 3 pm
Nonparametric analysis of point process data
Peter HALL (Australian National University, Canberra)
Wednesday, 27th October, 3 pm
Preservation of local dependence properties in spatial
patterns under clustering and superposition
Marie-Colette van LIESHOUT (CWI, Amsterdam)
Wednesday, 10th November, 3 pm
Statistical aspects of BSE and nvCJD: implications for
the public health
Sheila M. GORE (MRC Biostatistics Unit, Cambridge)
Wednesday, 8th December, 3 pm
Use of Gaussian process priors in engineering applications
Roderick MURRAY-SMITH (Department of Computer Science,
University of Glasgow)
Seminars take place in Room 1f(203), Mathematics Building,
University of Glasgow
For further information please contact the seminar organiser:
Ilya Molchanov
University of Glasgow : e-mail: [log in to unmask]
Department of Statistics : Ph.: + 44 141 330 5141
Glasgow G12 8QW : Fax: + 44 141 330 4814
Scotland, U.K. : http://www.stats.gla.ac.uk/~ilya/
ABSTRACTS
BAYESIAN PARTITION MODELS
In this talk we propose a new Bayesian approach to data modelling
motivated by the difficulties encountered with some tree-based
methods. The Bayesian partition model constructs arbitrarily complex
response surfaces over the design space by splitting it into an
unknown number of disjoint regions. Within each region the data is
assumed to be exchangeable and to come from some simple distribution.
Using conjugate priors the marginal likelihoods of the models can be
obtained analytically for any proposed partitioning of the space,
hugely simplifying the sampling algorithm required to simulate from
the posterior of interest. By example we shall show how the partition
model can be used for a wide variety of problems including regression,
classification and disease mapping.
NONPARAMETRIC ANALYSIS OF POINT PROCESS DATA
Motivated by multivariate data on epicentres of seismic events, we suggest
nonparametric methods for analysis of point-process data. Our methods are
based partly on nonparametric intensity estimation, and involve techniques
for dimension reduction and for mapping the trajectory of temporal
evolution of high-intensity clusters. They include ways of improving
statistical performance by data sharpening, i.e. data pre-processing
before substitution into a conventional nonparametric estimator. This leads
to new methods for reducing bias of a wide variety of nonparametric curve
estimators, for example in nonparametric regression.
PRESERVATION OF LOCAL DEPENDENCE PROPERTIES IN SPATIAL
PATTERNS UNDER CLUSTERING AND SUPERPOSITION
Fundamental point pattern operations such as
* thinning
* clustering
* superposition
allow the construction of new, more complex models from
simpler ones, and as such are very useful in the modelling of
spatial patterns.
The simplest model for point configurations is a stationary
Poisson process. It is well-known that if such a process is
independently thinned, the result is another Poisson process,
possibly inhomogeneous if the retention probability depends on
the location. Furthermore, superposition of two independent
Poisson processes is also a Poisson process, and independent
clustering applied to Poisson parents yields a Neyman--Scott
process.
In this talk, we shall consider Markov point processes, thus
allowing for local dependence between the points. Particular
attention will be paid to the effect of the operations listed
above on the local dependence structure.
STATISTICAL ASPECTS OF BSE AND NVCJD: IMPLICATIONS FOR THE PUBLIC
HEALTH
After a brief historical account which includes statistical
aspects of quality control at abattoirs, three current
statistical issues of public health importance are addressed:
1. maternal transmission, such as from BSE-dam to calf;
2. whether age-related consumption of bovine meat products is
sufficient to explain the younger age of nvCJD cases;
3. design considerations in estimating prevalence of
pre-clinical nvCJD.
USE OF GAUSSIAN PROCESS PRIORS IN ENGINEERING APPLICATIONS
Gaussian process priors are becoming interesting alternatives to
widely-used nonlinear regression approaches in engineering
applications. I will describe how they are being used, and illustrate
their advantages with a number of static and dynamic system examples
of nonlinear systems modelling. I will discuss how the choice of
covariance functions, and efficient implementation are very important
for practical applications. I will also look at ways in which
engineers can bring in prior knowledge from currently used paradigms,
and summarise the posterior distribution in a format they can relate
to. To some extent this requires the inference approach to be 'hidden'
from the user. I will relate this approach to the multiple-model
approach and discuss potential synergy.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|