International Review of UK ICT Research 2006
This page is under construction. Please revisit - and reload - to
see latest version.
Overview of Posters and Demos
ICCS
constraint-based sentence compression - an integer programming
approach
James Clarke
Abstract: The ability to
compress sentences while preserving their grammaticality and most of
their meaning has recently received much
attention. Our work views sentence compression as an optimisation
problem. We develop an integer programming formulation and infer
globally optimal compressions in the face of linguistically
motivated constraints. We show that such a formulation allows for
relatively simple and knowledge-lean compression models that do not
require parallel corpora or large-scale resources. The proposed
approach yields results comparable and in some cases superior to
state-of-the-art.
Further
information
modelling human parsing, syntactic priming, collaborative actions:
interdisciplinary research with psychology and linguistics
Frank Keller
Abstract: This poster describes
three areas of cognitive modeling that involve
researchers in Informatics, Psychology, and Linguistics. These include
information-theoretic models that account for prediction in human
parsing, corpus-based studies of syntactic priming and parallelism
(the tendency to repeat syntactic structures), and work on
collaborative tasks involving humans and robots. The use of
eye-tracking, an experimental approach that provides detailed measures
of human cognitive processing, underlies all three areas.
Further information
statistical machine translation
Miles Osborne
Abstract: The Edinburgh
Statistical Machine Translation Group consists of a
dozen faculty, postdocs, PhD students and visitors. We build
frameworks for translating between any language to any other language,
with a particular emphasis on Arabic to English, Chinese to English
and all major European languages to each other. Our current
research
areas include Factored Translation Models, Trillion word Language
Modelling, large scale discriminative approaches to translation and
methods for dealing with low-density languages. We also enter in
all
of the major international translation competitions, frequently
outperforming companies and other universities.
Further information
machine learning of dialogue management policies
Oliver Lemon
Abstract: We are
investigating machine learning methods for dialogue management
policies. In the TALK project (an EU FP6 project), we have
developed a
novel combination of reinforcement learning and supervised learning,
which
allows us to learn an entire dialogue policy from a fixed corpus of
human-machine dialogues.
We have also developed user simulations for use in automatic evaluation
and
optimization of policies.
Experiments with human users have demonstrated the advantages of the
learned policy over a state-of-the-art hand-coded policy.
In a forthcoming EPSRC project,
``End-to-end Integrated Statistical Processing for
Context-Aware Dialogue Systems'',
we will extend this work by developing tractable and effective
techniques for the integrated treatment of uncertainty in context-aware
dialogue systems, for
example using Partially Observable MDPs.
Further
information
ECB
Edinburgh Centre for Bioinformatics
Yulia Matskevich
Abstract: The research in the
group of Computational Systems Biology is focused on kinetic and static
modeling of biological processes by linking diverse data and modesl
through multiple iterations, from static ab initio models to highly
constrained kinetic models that cross multiple scales. Modelling will
be supported by the Systems Biology Software Infrastructure, a new
integrated platform, facilitating the modelling process from databases
to knowledge discovery, which is currently under development in our
group.
Current research themes of the group are:
• Edinburgh Pathway Editor
EPE is a visual editor designed for annotation, visualization
and presentation of wide variety of biological networks, including
metabolic, genetic and signal transduction pathways. It based on a
metadata driven architecture, which makes it very flexible in drawing,
storing, presenting and exporting information related to the network of
interest.
EPE was created as an Eclipse stand-alone application, with Eclipse
open framework architecture. This enables the development of extensions
to enhance the existing capabilities. Specific plug-ins, to perform
scientific computing and other tasks can be easily incorporated.
• Human metabolic network reconstruction and analysis
A better understanding of human metabolism and its relationship
with human disease is an important task in human systems biology
studies. This project aims to present metabobolic network reconstructed
from the genome annotation information. A preliminary network was first
reconstructed by integrating the information from different databases
such as EMP, KEGG, Brenda and Uniprot and the information from
literature, resulting in a network with about 3000 metabolic reactions.
We have reorganized the reactions into about 50 pathways according to
their functional relationships. The disease related metabolic enzymes
were marked for further analysis of their effect on human disease.
• Mathematical modelling and large-scale
computational simulation of complex biological systems
The purpose of this project is to develop modular open source
software to assist researchers in the building and modelling of
circuits. Dynamical system theory including bifurcation analysis and
global optimisation is employed to model the evolution of extremely
complex biochemical pathways of living organisms, using high
performance large-scale parallel computational techniques with aids of
supercomputers.
Further information
ESI
The e-science institute
Anna Kenway
Abstract: The e-Science
Institute is the UK's interdisciplinary centre for
e-Researchers to meet, work and exchange ideas. Hosted by
Edinburgh
University, it has already been operating for 5 years and has now been
extended to July 2011. The eSI poster records its past activity
and
describes it current development into a more thematic and research
oriented mode.
Further information
IPAB
Structure inference for bayesian multisensory perception and
tracking
Timothy Hospedales
Abstract: We investigate
a solution to the problem of multi-sensor perception and
tracking by formulating it in the framework of Bayesian model
selection.
Humans robustly associate multi-sensory data as appropriate, but
previous theoretical work has focused largely on purely integrative
cases, leaving segregation unaccounted for and unexploited by machine
perception systems. We illustrate a unifying, Bayesian solution to
multi-sensor perception and tracking which accounts for both
integration
and segregation by explicit probabilistic reasoning about data
association in a temporal context. Unsupervised learning of such
a model
with EM is illustrated for a real world audio-visual application.
Further information
IPAB overview
Bob Fisher
Abstract: The Institute
of Perception, Action and Behaviour (formed 1998) is focused on
activities related to the issue of how to link computational
perception,
representation, transformation and generation processes to external
worlds. The external world may be the "real" world or another
computational environment that has its own character. Examples of where
this issue arises
occur throughout the research in IPAB:
video sequence analysis, 3D shape capture and analysis, interacting
agents in computer games or video, flexible and tolerant robot control,
learning-based robot control, biomimetic robotics, particularly for
various insects. This poster shows some example images from our
research results.
Further
information
ipab connections with industry
Matt Howard
Abstract: The Institute
for Perception Action and Behaviour has a variety of links
to industry. These include collaborative research projects as well as
spin-off technology transfer companies. This poster illustrates current
links
with the research efforts of well-known companies such as Honda and
Microsoft as well as new companies started by former IPAB members. The
diversity of these links reflects the broad range of research conducted
in the institute that is highly valued by our commercial counterparts.
Further information - Honda
Further information -
Edinburgh Robotics
Further information
- Microsoft
Further information - Dimensional Imaging
realistic nonparametric 3D surface completion
Toby Collins
Abstract: Real 3D scene
acquisition usually requires the ability to completely scan
the scene objects, but this is normally impossible due to access
restrictions or physical limits. The research presented here shows how
one can use the scanned portion of an object to hypothesise unscanned
portions, whether holes or the complete back surface. The key idea is
to
grow the known surface outward across a hypothesised underlying
surface, using matching sample neighbourhoods from theobserved front
surface.
The poster shows a reconstruction of the Leaning Tower at Pisa, where
the floors, ridges and cupolas are realistically reconstructed, even
though
no model of the building was used. The approach can be extended to
multiple scales and to incorporate colour.
Further
information
ICSA
enhancing the performance predictability of grid applications with
patterns and process algebras
Murray Cole
Abstract: The Enhance project
aims to simplify the efficient programming
of Grid systems by exploiting results from two underlying research
programmes. Skeleton based programming recognises that many real
parallel applications draw from a range of well known solution
paradigms and seeks to make it easy for an application
developer to tailor such a paradigm to a specific
problem without re-inventing the wheel. Meanwhile, stochastic process
algebras such as PEPA are used to model the behaviour of concurrent
systems in which some aspects of behaviour are not precisely
predictable. By modelling our skeletons with PEPA,
and thereby being able to include aspects of uncertainty which are
inherent to Grid computing, we are able to underpin run-time
systems which
make better scheduling and rescheduling decisions than less
sophisticated approaches.
Further information
industrial collaboration with icsa
Nigel Topham
Abstract: Most of the
research undertaken in the Institute for
Computing Systems Architecture has a direct relevance
to the embedded computing industry. There is a history
of close interaction between researchers in ICSA and
the embedded computing industry in the UK and Europe
in particular. ICSA is currently involved in: two major
collaborative European projects; two smaller projects
involving Engineering Doctorate students at ARM (Cambidge)
and Critical Blue (Edinburgh); and a close collaboration
with ARC International in the area of new high-performance
low-power embedded microprocessor architectures.
This poster describes how ICSA research results have fed
into startup activities, and explains how the long-term
research collaboration with ARC International has
cross-fertilized both the product developments at ARC and
the research activities in ICSA.
Further information
Using Machine learning to forcus iterative optimization
Mike O'Boyle
Abstract: Iterative compiler
optimization has been shown to outperform static
approaches. This, however, is at the cost of large numbers of
evaluations of the program. This paper develops a new methodology
to
reduce this number and hence speed up iterative optimization. It uses
predictive modelling from the domain of machine learning to
automatically focus search on those areas likely to give greatest
performance. This approach is independent of search algorithm, search
space or compiler infrastructure and scales gracefully with the
compiler optimization space size. Off-line, a training set of programs
is iteratively evaluated and the shape of the spaces and program
features are modelled. These models are learnt and used to focus the
iterative optimization of a new program. We evaluate two learnt
models, an independent and Markov model, and evaluate their worth on
two embedded platforms, the Texas Instrument C6713 and the AMD Au1500.
We show that such learnt models can speed up iterative search on large
spaces by an order of magnitude. This translates into an average
speedup of 1.26 on the TI C6713 and 1.27 on the AMD Au1500 in just 2
evaluations.
Further
information - machine learning
Further information -
compilers`
Research consortium in speckled computing
D.K. Arvind
Abstract: The Research
Consortium in Speckled Computing is a multidisciplinary grouping of
Computer Scientists, Electronic Engineers, Physicists and
Electrochemists, drawn from five universities, researching the next
generation of miniature (5X5X5mm) mobile computing devices called
specks: each speck combines sensing, processing and wireless networking
capabilities. The Consortium is vertically integrated, ranging from the
design, realisation and intergration of the miniature specks, to the
efficient organisation of networks of specks - called specknets - as a
fine-grained distributed computation platform. The research is funded
jointly by the Scottish Funding Council (Strategic Research Development
Grant) and the Engineering and Physical Sciences Research Council
(Basic Technology Grant) in excess of £5Million for the period
2004-10, and is led by the School of Informatics, University of
Edinburgh. Contact; DK Arvind (
dka@inf.ed.ac.uk)
Quoting
mob@inf.ed.ac.uk:
Further information
LFCS
SMOQE: A System for providing secure access to xml
Xibei Jia
Abstract: This poster
outlines the results of database research in two areas of
data integration. The first concerns SMOQE, the first system
to provide
efficient support for answering queries over virtual and possibly
recursively defined XML views. XML views have been widely used to
integrate data, speed up query answering, and above all, enforce
XML security, for which virtual XML views are necessary. SMOQE
encompasses an array of novel techniques for specifying XML
(security) views, and for rewriting, evaluating and optimizing
XML queries posed on views without materializing the views.
The second provides a picture of a uniform system for integrating,
cleaning, maintaining and securing data. The system supports a number
of functionalities, including (a) automated schema mapping via a novel
notion of schema embedding, (b) XML data publishing for exporting data
from relational databases as XML documents, (c) XML data integration
for combining data from multiple distributed and heterogeneous
data sources, (d) cleaning integrated data based on a set of new
integrity constraints designed for detecting inconsistencies,
(e) incremental maintenance of integrated data, and
(f) securing integrated data (SMOQE). This is the
first uniform system capable of doing almost everything one
needs for data integration.
Further
information
two themes in digital curation: archiving & annotation
Heiko Mueller
Abstract: This poster
presents the results of database research in two areas of
digital curation. The first concerns the preservation aspect of
curation; it demonstrates a technique for archiving all
versions of a scientific database. It allows the efficient
retrieval of any version and also permits temporal queries on the
history of components of the database. The storage overhead is small,
and depends on the amount of change rather than the frequency of
changes.
The second concerns data annotation, which is an important activity
especially in curated biological databases. As yet there is no
generic technology for annotating databases and for querying those
annotations. Mondrian is a system for annotating relational databases
through blocks and colours. The colours represent the
annotations, and
the blocks allow annotations to be attached to relationships among
data elements.
Other areas of database research that are important for digital
curation include data provenance, data publishing, data integration
and data citation.
Further
information
IANC
Chris Williams, John Quinn, Neil McIntosh
Abstract:
Premature babies in intensive care are monitored continuously, with
several different physiological measurements taken per second. These
measurements indicate the state of health but are noisy and require a
lot of experience to interpret. We model this data probabilistically as
a Factorial Switching Kalman Filter, and show that this allows us to
make inferences about the state of the baby and the operation of the
monitoring equipment.
Further information
Surround modulation by long-range lateral connections in an
orientation map model of primary visual cortex development and function
Judith S. Law and James A. Bednar
Abstract: Neuronal response
properties are often smoothly topographically
organised across the cortical surface. The prototypical example is the
map of orientation preference in primary visual cortex (V1). Many
models of orientation map development have been very successful in
reproducing the features of biological maps. The majority of these
models are based on a principle of short-range excitatory and
long-range inhibitory connections between neurons, e.g. von der
Malsburg, 1973, Swindale, 1992, Obermayer et al., 1990 and the LISSOM
model, Sirosh and Miikkulainen, 1997. However, biological data
suggests that long-range connections between V1 neurons arise
primarily from putatively excitatory pyramidal cells (Gilbert &
Wiesel,1989, Hirsch & Gilbert, 1991, Weliky et al.,1995, Angelucci
et
al.,2002). Furthermore, simple models with long-range excitation and
shortrange inhibition have shown how a biologically realistic
circuitry can reproduce features of adult V1 function such as
extra-classical receptive field phenomena (Schwabe et al., 2006).
These models of adult function suggest that long-range excitatory
connections are facilitatory when input is at low contrast, yet
stronger activation of local inhibitory neurons at high contrast will
cause these connections to act supressively. Previous developmental
map models with long-range inhibitory connections are therefore unable
to account for aspects of surround modulation. However, it is not yet
clear how such circuits can arise, which parts of the system are
plastic, or in general how to reconcile these findings with otherwise
successful developmental models such as LISSOM. We present the
first
model which is consistent with this realistic connectivity, yet also
reproduces the features of successful developmental models of
topographic map formation. Future work will address how this
connectivity can lead to surround modulation both in adult V1 and
throughout development.
Further
information
neuroinformatics doctoral training centre
Mark van Rossum
Abstract: This poster
gives an overview of the structure, aims, and goals of the
NeuroInformatics Doctoral Training Centre.
neuroinformatics: computing and the brain
Mark van Rossum
Abstract: This poster presents
a selection of Phd projects done in the NeuroInformatics DTC. It
addresses: How does a rat know where it is going? How to make a
lasting neural memory? How does human memory work? and How
do we see at low contrast.
Further information
CSTR
ami and amida: meeting browsers and remote meeting support
Jean Carletta
Abstract: The AMI
Consortium develops new technologies to aid groups that hold
meetings. The AMI project concentrated on ways of using archives
of
face-to-face meetings, and AMIDA will contribute aids for people who
need to attend a meeting, but can't be together. Edinburgh's main
contributions are in project coordination, data collection, speech
processing, and language technology.
Further information
speech recognition: novel approaches
Simon King
Abstract: In this
poster, we describe some of the novel approaches to speech
recognition that we are investigating. These complement the more
mainstream work being carried out in the AMI project, described
elsewhere. The motivations of our novel approaches come from two
inadequacies of current approaches (Hidden Markov Models of phonemes):
Describing speech as a linear string of phonemes is inadequate and
causes many problems for statistical modelling; Modelling speech
directly in an acoustic observation space makes separation of classes
very difficult.
To avoid the problems of the phonemic representation, we are working
in two quite different directions. The first has strong linguistic
motivations based on properties of speech production and includes work
with articulatory measurement data and articulatory/phonetic features,
all of which are factored (multi-stream) representations. To build
statistical models of such representations, we use Dynamic Bayesian
Networks. Our second, more recent, direction has purely "engineering"
motivations: we model speech as a string of graphemes; this is
linguistically implausible (especially for English), but avoids the
need for pronunciation dictionaries (which are poor representations of
natural, spontaneous speech); accuracies using grapheme models are
already almost as good as for phoneme models. This is further evidence
that phonemes are inadequate.
Instead of classification using Gaussian mixture models of acoustic
observations (which are derived from the short term spectrum of the
speech signal), we are looking at alternative techniques, including
features based on class posterior probabilities, produced by some
classifier (usually a neural network). This approach is not new in
itself, but our novel contribution is to consider what classification
task this neural network should be performing: conventionally, this is
always phoneme classification, but we are looking at
articulatory/phonetic features and graphemes as alternatives.
The above research is complemented by more theoretical work and by
application-driven work. We are developing theory for learning the
sub-word unit inventory (rather than pre-specifying it as phonemes or
graphemes, for example) and for learning graphical model structure
and the structure of precision (i.e. inverse covariance)
matrices. Both of these topics involve automatically selecting models
of appropriate structure and complexity for the data, to optimise
classification performance. On the applications side, we are testing
our models in areas including multi-lingual speech recognition and
audio search. Both of these areas stand to benefit from using sub-word
units other than phonemes.
Further information
SysBio
eSI
understanding human development through gene expression
Jano van Hemert
Abstract: The
Developmental Gene Expression Map project ams to design the
infrastructure for a pan-European collaborative network on the
study of gene expression in early human development. Where the
project includes the ethical and biological side of the
infrastructure, in this poster, we focus on the ICT oriented
research components that would contribute to a better
understanding of human development. These include collaborative
experiment planning, spatial-temporal gene expression databases,
3D reconstruction and visualisation, data integration, data
mining, integrative biology, computational modelling, and systems
biology.
Further information
EPCC
OGSA-DAI (poster joint with ESI)
Neil Chue Hong/Konstantinos Karasavvas/Malcolm Atkinson
Abstract:
OGSA-DAI demonstrates the University
of
Edinburgh's ability to combine research on data access and integration
using
grid and web service technology and in-house software engineering
expertise to
create outputs which benefit international research. OGSA-DAI enables
diverse
heterogeneous data sources to be accessed through uniform interfaces
and
provides a flexible framework for managing additional processing
functionality
on the data exposed, reducing overall data transfer. Recent research
has
focused on designing a pipelining model which enables data integration
activities to be orchestrated, and data transferred between them in an
efficient manner.
Further
information
leadership in europe
Neil Chue Hong/Kostas Kavoussanakis
Abstract:
As part of its mission, EPCC is committed to
transferring
skills and knowledge to UK
and European industry.
EPCC coordinates the 21-partner NextGRID
project, which seeks to ensure that Europe
is
a world leader in the next generation of Grid technology. The
three-year
project envisions the development of an architecture for Next
Generation Grids
which will enable their widespread use by research, industry and the
ordinary
citizen.
http://www.nextgrid.org/
With Grid middleware reaching maturity, industrial
uptake of
Grid solutions is paramount for European economy. The 74-partner BEinGRID
project includes top European Grid, IT and business experts, supporting
business experiments as they pilot Grid solutions in diverse market
sectors.
Instrumental in the design and management of the project, EPCC also
leads Data
Management support to the business experiments.
http://www.beingrid.eu/
outreach to other disciplines
Robert Baxter/George Beckett/Mark Parsons
Abstract:
EPCC has an outstanding reputation for providing
computing
solutions to disciplines across sciences. We showcase two of our
current
projects.
EPCC is working with eight other academic and
commercial
technology providers in ITI Techmedia’s Condition-Based
Monitoring
(CBM) Programme, investigating the application of CBM technologies to
commercial farming. EPCC leads the biological modelling work, applying
expertise in software, data management and data analysis to develop key
intellectual property for the ITI CBM platform.
http://www.ititechmedia.com/defaultpage131abcde0.aspx?pageID=806
Distributed Grid Storage (DiGS) is a
grid
application that combines disparate storage resources to form a unified
'data
grid', capable of meeting the data management challenges of both QCD
Physics
and a wider scientific community. It combines disparate mass storage
technologies (e.g. RAID units or SAN systems) to provide a unified,
multi-Terabyte 'data grid' facility.
http://www.gridpp.ac.uk/qcdgrid/
cisa
the helpful environment
Austin Tate
Abstract: The Planning
and Activity Management Group within the Artificial Intelligence
Applications Institute (AIAI) in the School of Informatics at the
University of Edinburgh is exploring representations and
reasoning mechanisms for inter-agent activity support. The agents
may be people or computer systems working in a coordinated
fashion. The group explores and develops generic approaches by
engaging in specific applied studies. Applications include crisis
action planning, command and control, space systems,
manufacturing, logistics, construction, procedural assistance,
help desks, emergency response, etc.
Our long term aim is the creation and use of task-centric virtual
organisations involving people, government and non-governmental
organisations, automated systems, grid and web services working
alongside intelligent robotic, vehicle, building and
environmental systems to respond to very dynamic events on scales
from local to global.
The group is involved in collaborative research projects,
programmes, standards and other activities internationally.
Further information
how safe is your pin?
Graham Steel
Abstract: Cash machines
(ATMs) and other critical parts of the electronic payment
infrastructure contain tamper-proof hardware security modules (HSMs),
which protect highly sensitive data such as the keys used to obtain
personal identification numbers (PINs). These HSMs have a restricted
API that is designed to prevent malicious intruders from gaining access
to the data. However, several attacks have been found on these APIs, as
the result of painstaking manual analysis by experts such as Mike Bond
and Jolyon Clulow. At the University of Edinburgh, a project is
underway to formalise and mechanise the analysis of these APIs. We aim
to develop techniques that help API designers to specify their systems
precisely and check them for flaws. This poster introduces the
challenges of the ATM network scenario, and describes our methods for
analysing security APIs, using theorem provers, protocol analysis
tools, and the PRISM probabilistic model checker.
Further information
.
.
.
.