Two postdoctoral positions in the Department of Mathematical
Sciences, Durham University
The Statistics Group are looking to recruit two postdoctoral research
assistants, as part of the RCUK-funded
MUCM project (Managing
Uncertainty in Complex Models). This is an opportunity to join one of
the world's leading research groups in
model-based inference for
complex systems, and to be part of a major initiative across some
of the UK's leading universities.
The MUCM project, funded by Research Councils UK, is a
collaboration between five universities: Sheffield, Aston,
Southampton, Durham, and the London School of Economics. It is
concerned with quantifying and reducing uncertainty in the predictions
of complex models across a wide range of application areas, including
basic science, environmental science, engineering, technology,
biosciences, and economics. The project is multi-disciplinary, and
the unifying theme is a bayesian statistical approach to inference.
At Durham we are interested both in developing methodology and in
applications. Within MUCM our theoretical interest centres on use of
multi-scale models (eg, slow and fast simulators of the same system)
and linking imperfect models to their underlying reality. Our main
applications are climate prediction and hydrocarbon reservoir
management; in both cases we work with the leading UK experts.
The Statistics Group has 11 academic staff and
15 postgraduate students. Areas of particular research interest
include Bayesian statistics, with special focus on large-scale
applications, and foundations of statistics. Members of the group
have been involved in model-based inference for complex systems (a
class of Bayesian large-scale applications) for over twenty years.
Staff currently working in this area include Peter
Craig, Michael
Goldstein, John
Little, Jonathan
Rougier, and Allan
Seheult.
There are two postdoctoral positions available at Durham. Both
appointments are on the RA1 scale (starting at pt 5: £23,500), and will
be for three years in the first instance, with the possibility of an
extension of up to one year. The primary supervisor in both cases is
Michael Goldstein; informal enquiries may be directed to him (Michael.Goldstein@durham.ac.uk;
tel +44 (0)191 334 3065), or to Jonathan Rougier (J.C.Rougier@durham.ac.uk;
tel +44 (0)191 334 3111).
The closing date for applications is Tuesday 28 February, 2006.
See below for the application procedure.
RA2: Multiscale models
Summary
When addressing a potentially very complex and demanding model, there
are usually choices that can be made in terms of the resolution of the
simulator or of the emulator. Models at coarser resolution run faster,
and emulators over a smaller region of the input space require fewer
model runs to reproduce the simulator output accurately. This part of
the project will explore various ways to exploit these choices to
obtain efficient analyses with high-dimensional models. For instance,
more efficient emulator development may be obtained by making many
runs at low resolution and only a few high-resolution runs, provided
we can model the relationship between the simulators at different
resolutions (see below, RA3).
We will also explore problems of optimisation and calibration using
a recursively defined sequence of simulators of different solver
resolutions, defined on nested parameter spaces. By using linked
emulators of these simulators we can initially explore the gross
properties of the problem, for example by exploiting local stochastic
dependencies in simulator outputs such as spatio-temporal fields. We
will also assess the minimal solver resolution necessary to achieve an
acceptable level of performance for the chosen task, or make an
informed choice about what solver resolution provides the best value
for money.
Personal specification
Essential:
- Candidate should have a PhD in Statistics or Mathematics, or
equivalent qualification (or should expect to submit a thesis for PhD
in one of these areas before 30 September 2006)
- Good communication skills
- Willingness to work in a multi-disciplinary team with a
focus on developing and applying model uncertainty technology in a
range of applications
- Familiarity with at least one programming language (e.g. Matlab,
R,C, Mathematica)
Desirable:
- Strong computing skills
- Good knowledge of multivariate statistical methods
- Knowledge of Bayesian statistics and/or Gaussian processes
- Experience of complex simulation models for real-world processes
RA3: Linking models to reality
Summary
Early work on calibration has taken a simplistic approach to the
nature of model error and the notion of 'true' parameter values,
assuming a simple discrepancy function between the simulator output at
the true input and the system itself, where the true value is unknown
and independent of the discrepancy. There is a need for more realistic
representations. We will generalise this framework to a more realistic
appraisal of simulator quality, using the Bayesian Graphical Modelling
(BGM) approach. This provides a formal graphical structure within
which we can develop a full description of the uncertainties in the
problem, by adding vertices that:
- Represent measurable system quantities that relate to simulator
input, allowing us to relax the notion of a unique correct input
value;
- Provide a structured representation of the approximations in the
solver;
- Account for irreducible uncertainty from inadequacies in the
underlying model.
For problems such as climate change, judgements about system
behaviour depend on analysis from several simulators. The BGM approach
provides a natural framework for combining analysis from different
computer simulators, based on merging the corresponding graphical
models to produce an overall uncertainty statement about system
behaviour. In this way we will resolve the role of ensembles of
models, which is an issue that goes beyond climate simulation. The BGM
approach also provides powerful tools for dealing with systematic
effects deriving from the spatio-temporal structure present in many
simulators.
Personal specification
Essential:
- Candidate should have a PhD in Statistics or Mathematics, or
equivalent qualification (or should expect to submit a thesis for PhD
in one of these areas before 31 September 2006)
- Good communication skills
- Willingness to work in a multi-disciplinary team with a focus on
developing and applying model uncertainty technology in a range of
applications
- Familiarity with at least one programming language (e.g. Matlab,
R,C, Mathematica)
Desirable:
- Strong background in Bayesian statistical methods
- Knowledge of Gaussian processes
- Experience of complex simulation models for real-world processes
The main terms and conditions of employment are as follows (the person
appointed will receive a full written statement of the terms and
conditions of employment).
- The salary will be on the RA1A salary scale £21,156 to £28,829 per annum.
- The post will be subject to 3 years probation.
- The post is full-time and fixed term for 3 years, available from 1 June 2006.
- The post is pensionable.
Applicants for this or any of the other postdoctoral research posts in
MUCM (at Durham, Sheffield, Aston, LSE or Southampton) should use the
standard University of Sheffield application form, obtainable from the
addresses below, or electronically
from here. You may apply for more than one of these posts (or all
of them) with a single application. Please quote the Sheffield
application reference number PR2325.
Please send:
- A letter of application, stating in particular which of the postdoctoral posts you wish to apply for;
- A curriculum vitae;
- A completed Summary Information Form and Equal Opportunities Monitoring Form.
| 1. By post:
|
The Staff Recruitment Service
Department of Human Resources - Personnel Services
The University of Sheffield
Firth Court
Western Bank
SHEFFIELD S10 2TN
|
| 2. In person:
|
The Staff Recruitment Service
Department of Human Resources - Personnel Services
The University of Sheffield
10 - 12 Brunswick Street
SHEFFIELD S10 2FN
(Reception is open Monday-Friday 9am - 5pm)
|
Page maintained by Jonathan Rougier.
Last modified: Tue Jan 31 13:13:04 GMT 2006