Queen Mary, University of London

     Mathematical Sciences / Computer Science / Electronic Engineering

                      Interdisciplinary Studentships

Funding is available for joint projects in the School of Mathematical
Sciences, in collaboration with the Department of Computer Science (2
awards) or the Department of Electronic Engineering (2 awards). All areas
of mathematical sciences, including statistics, are eligible. The
following two project proposals have been developed and would be
strengthened by naming a suitable applicant.


1) Project Title:  Design and analysis of experiments in Computer Science
                involving groups of people

Principal supervisor: Professor R.A. Bailey (Statistics)

Co-supervisor: Dr P. Healey (Computer Science)

Pat Healey and Graham White have been involved in several projects to
see how people interact with each other. Communication tools, such as chat
tools, whiteboards and conferencing systems, provide an unprecedented
degree of experimental control over the mechanisms of interaction
available to users, and over the history and topology of interactions
across communities as a whole. This provides a powerful way of testing
hypotheses about the basic organisation of human interaction and tracking
the evolution of communicative practices and 'dialects'.

For example, we can create artificial sub-communities who play a
collaborative game together and then compare the kinds of communicative
convention they develop. This design has already been used to demonstrate
the emergence of group specific sub-languages.

We can also manipulate the kinds of participation that are possible
(addressee, overhearer, bystander) and the kinds of interaction mechanism
(synchronous vs. asynchronous, edit own vs. edit partner's and own
contributions). We can also study trust and deception: student projects
have investigated a game which involves the use of deception in groups of
three people who can communicate privately in pairs.

In all of these experimental situations, a group of people has to work
together, and the group usually has a strong influence on individual
behaviour. Typically the experimental unit (in the sense of what an
experimental condition can be applied to) is a group of people for a
certain time period. In the experiments done so far, each group of people
has remained together throughout. Sometimes the response is measured on
the whole group for that time period (eg length of time to complete  the
game); other times there is a response on each person in that time period
(eg number of lies told). In both cases the analysis should use the data
at the group level.  It is known that statistics become  problematic when
the pair, or group, is the unit of analysis, but this is often not
realised when Computer Scientists report their results.

Data from a pilot study of the second type showed that there was
considerably more variability between groups than there was between people
within groups. This suggests that there might be some advantage in
designing the experiment so that people do not stay in the same groups
throughout. This is a novel experimental structure so there are
statistical issues to explore in both design and analysis, and these
should be resolved before we can consider whether such a new design would
be suitable in practice.

Rosemary Bailey has had to address this same statistical issue for an
industrial client, so there is potential for widespread applicability of
the results.


2) Project title: Design of Experiments for Network Simulations

Principal supervisor: Professor S.G. Gilmour (Statistics)

Co-supervisor: Dr J.A. Schormans (Electronic Engineering)

Network simulation is widely used in the design and development of
wide-area internet protocols, the study of network design and dimensioning
techniques (e.g. as based on queueing theory), the development and testing
of monitoring and measurement techniques, and systems and methods of
optimal data storage of measured data. As networks increase in size and
complexity, and as attention focuses more on whole-system performance
evaluation, the selection of sets of simulation input parameters which
will provide efficient and relevant data becomes more important, as do
methods of accelerating these simulation experiments.

Since the simulations are stochastic in nature, this defines a problem in
the design of experiments. The application of the statistical principles
of design of experiments to large scale simulations has been an area of
increasing interest in the statistical literature in recent years.
However, none of this has concentrated on communication networks. These
have a number of features which raise interesting statistical research
questions:

  - The most interesting features of these communications networks are
highly variable, whereas in many other simulation experiments the
variability is much smaller than in physical experiments. Furthermore,
many of these features have been found to exhibit power law type
behaviour, which may result in extremely large variances.

  - The relevant simulation outputs are the extremes of distributions,
rather than means or variances (which are the focus of virtually all
statistical work in design of experiments).

  - Each simulation run requires very large amounts of computer time, so a
focus on minimising the number of runs is more important than in many
other applications. Again the computer resources required is greatly
increased in the presence of behaviour dominated by power law
distributions.

Types of experimental design that will be relevant include supersaturated
designs for initial screening of large numbers of factors and response
surface designs for more detailed modelling of particular features of
interest. These will have to be adapted for measuring the types of
features of interest in network simulations.


These Queen Mary studentships are open to any suitably qualified
applicants. Applicants should have a first degree in a mathematical
subject, involving some statistics. Prior knowledge of computer science or
electronic engineering is not necessary. The studentships provide for
living expenses at the standard research council rates and tuition fees at
the home student rate. Non-EU applicants will have to demonstrate that
they can cover the difference between home and overseas tuition fees from
other sources.

To discuss these informally, please contact Steven Gilmour at the address
below, or by email. To express an interest in applying, please send a
curriculum vitae, or a completed application form to Steven Gilmour,
preferably by email. Application forms are available online at
http://www.qmul.ac.uk/postgraduate/apply/index.html .

--

Professor Steven G Gilmour

School of Mathematical Sciences
Queen Mary, University of London
Mile End Road
London E1 4NS
United Kingdom

Tel: +44 (0)20 7882 7833
Fax: +44 (0)20 8981 9587 (department fax, not private)

Web page: http://www.maths.qmul.ac.uk/~sgg