DEPARTMENT OF COMPUTING SCIENCE AND MATHEMATICS
UNIVERSITY . COMPUTING SCIENCE . SEMINARS

SEMINARS - Spring 2005

[Talk Schedule] [Abstracts] [Previous Seminars]

The Department of Computing Science and Mathematics presents the following seminars. Unless otherwise stated, seminars will take place in Room 4B94 of the Cottrell Building, University of Stirling from 15.00 to 16.00 on Friday afternoons during semester time. For instructions on how to get to the University, please look at the following routes.

If you would like to give a seminar to the department in future or if you need more information, please contact the seminar organisers, either Julie Cowie (Phone 01786 467446, Email jco@cs.stir.ac.uk) or David Cairns (Phone 01786 467445, Email dec@cs.stir.ac.uk).

Talk Schedule [Top] [Abstracts]

Abstracts [Top] [Schedule]

25th February [Schedule]

Towards Paradigm Unification Modelling
Manuela Bujorianu
Department of Computing Science, University of Stirling

Systems are becoming increasingly complex. Real Life demand consideration of multiple aspects like uncertainty, discrete/continuous interaction, distribution, communication, etc. In our recent work, we have developed a unifying model for systems paradigms: starting with a stochastic hybrid system model we added extra features like distribution and communication. Applications are of a great complexity including Air Traffic Control, financial markets, biology, etc.

4th March [Schedule]

Adaptive Structured Parallel Programming for Grid Computing
Horacio Gonzalez-Velez
Institute for Computing Systems Architecture, University of Edinburgh

Abstract
Grids pose a challenge to known distributed systems techniques as a result of their dynamism. One of the most prominent research areas concerns the availability of proved programming paradigms with special emphasis on the performance side. Thus, adaptable performance improvement techniques have been the subject of intense scrutiny.

Algorithmic skeletons abstract commonly used patterns of parallel computation, communication, and interaction. They present a top-down transformational approach where programs are formed from skeleton nesting. By demonstrating a predictable communication and computation structure, they provide a preponderant foundation for performance modelling and estimation.

Scant research has been conducted on using this predicting information to enhance performance in heterogeneous environments. We propose the use of these predicting properties to adaptively enhance the performance of skeletons, in particular of a processor farm, within a computational grid.

11th March [Schedule]

Research Funding - how working with a business partner can help
Kate Wooding, SURE (Stirling University Research & Enterprise)

This seminar will focus on the role of SURE (Stirling University Research & Enterprise) and how the Unit is working with academic colleagues to partner their research interests with external organisations and businesses. As well as outlining the activities of SURE in general, some time will be spent looking at sources of funding that can be accessed by working with external partners. In addition there will be some case studies of successful projects from different departments that have benefitted from SURE's help to develop academic expertise into commercially viable projects.

18th March [Schedule]

Exploiting symmetrical system structure: mathematics for more manageable
model checking
Alastair Donaldson, Department of Computer Science, University of Glasgow

Concurrent systems are hard to design and implement correctly. Consequently, a lot of effort has gone into the use of (non-concurrent) software tools to aid in the development and debugging of concurrent systems. One class of such software tools are model checkers. A model checker can check a property of a finite-state concurrent system by exhaustive search of the system's state-space. Model checking can be used to detect, for example, whether messages will be passed correctly, or whether buffer overflow may occur in a network protocol. Model checkers are useful for finding subtle errors which go unnoticed by human designers. The application of model checking is limited, since the state-space associated with a complex concurrent system may be prohibitively large. However, various mathematical techniques can be used to reduce the state-space which must be searched to check a system property.

One such technique, symmetry-based reduction, involves exploiting the replicated structure of a distributed, concurrent system. In this talk I will give an overview of symmetry reduction techniques for model checking. I will discuss the two main problems which a symmetry reduction scheme must handle: detection of valid symmetries from a system description; and efficient exploitation of these symmetries during model checking. In particular, I will describe the application of computational group theory to the solution of these two problems.

23rd March [Schedule]

Mobile Interaction Design in the Developing World: Avoiding the pot holes
Dr Gary Marsden
Department of Computer Science, University of Cape Town, South Africa

Much has been written about the potential of ICT to uplift the developing world and close the digital divide. The creation and deployment of these solutions, however, falls far short of the rhetoric. This talk looks at how interaction design and mobile computers can be used to create effective ICT solutions in the developing world. In particular it looks at new interface development methodologies and critiques HCI as it is practiced in the developed world.

8th April [Schedule]

The role of retinal non-linearities in visual perception
Mattias Hennig Department of Computing Science, University of Stirling

In this talk, I will review the work I did in the last four years on non-linear processing in the vertebrate retina. I will first introduce a model of the retina which can account for the experimentally observed non-linearities in retinal ganglion cells. Using simulation results, I will then show how these non-linearities, in conjunction with small, involuntary eye movements, could improve stimulus detection and discrimination in various situations. I will further show that consequences of non-linear processing are directly visible to (almost) everyone in a certain group of visual illusions. Finally, I will give a demonstration of a neuromorphic VLSI implementation of some simple retinal and cortical circuits (called the 'Physiologist's Friend Chip', see http://www.ini.unizh.ch/~tobi/friend/chip/index.html), which is a very nice tool for teaching basics of visual neuroscience and neural computation.

15th April [Schedule]

Reasoning about the ERK signal transduction pathway using BioSigNet-RR
Dr Carron Shankland

Department of Computing Science, University of Stirling Abstract
Understanding the nature of ERK activation and how different drugs affect that activation is key to developing therapies for cancer. We show how the answer set based tool BioSigNet-RR can be used to model signal pathways (in particular the ERK signal transduction pathway) and to carry out qualitative analysis of the system. Three kinds of analysis are possible: prediction, explanation, and planning. Such analysis allows us to make early evaluation of the behaviour of the pathway, e.g. in the presence of new drugs. Explanation in particular can be useful to generate plausible initial parameters for the system. All of these can be used to inform experimentation.

This is joint work with Nam Tran and Chitta Baral (Arizona State University) and Walter Kolch (Beatson Institute for Cancer Research, University of Glasgow).

 

22nd April [Schedule]

Representing and Analysing Web Services
Prof. Kenneth J. Turner
Department of Computing Science, University of Stirling

Abstract
Despite the popularity of web services, creating them manually is an intricate task. Composite web services are defined using the industry-standard language Bpel (Business Process Execution Logic). The need for graphical and automated support then becomes even more apparent. It will be explained how Cress (Chisel Representation Employing Systematic Specification) has been extended to meet this challenge. The approach will be compared to some well-known industrial tools. Cress supports straightforward graphical descriptions of composite web services. Sample Cress descriptions will be presented of these. It will be seen how they are automatically implemented using the underlying web languages. The same graphical descriptions are also translated into Lotos, permitting rigorous analysis and automated validation. A feature concept will be introduced for web services, and feature interaction discussed in this context.

29th April [Schedule]

This is a two part talk given by Gregor Kiddie and Dr Bruce Graham from the Department of Computing Science & Mathematics at the University of Stirling.

Modelling the variation in dendritic outgrowth between different neuronal types
Gregor A. Kiddie, Arjen van Ooyen and Bruce P. Graham

The growth of a neuronal dendritic tree depends on the neuron’s internal state and the environment within which it is situated. Different types of neuron develop dendritic trees with specific characteristics, such as the average number of terminal branches and the average length of terminal and intermediate segments. A key aspect of the growth process is the construction of the microtubule cytoskeleton within the dendritic tree. Neurite elongation requires assembly of microtubules from free tubulin at the growth cone. The stability of microtubule bundles is an important factor in determining how likely it is for a growth cone to split to form new daughter branches. Microtubule assembly rates and bundle stability are controlled by microtubule-associated proteins, principally MAP2 in dendrites. Extending our previous work (Hely et al, J. Theor. Biol. 210:375-384, 2001) we have developed a mathematical model of neurite outgrowth in which elongation and branching rates are determined by the phosphorylation state of MAP2 at the tips of each terminal branch. Tubulin and MAP2 are produced in the cell body and transported along the neurite by a combination of diffusion and active transport. Microtubule (dis)assembly at neurite tips is a function of tubulin concentration. The rate of assembly depends on the amount of unphosphorylated MAP2 bound to the microtubules and linking them together. Phosphorylation of MAP2 destroys its linking capability and destabilises the microtubule bundles. Each terminal has a probability of branching that depends on the phosphorylation of MAP2 which, in turn, is a function of calcium concentration. Initial results from this model show that changes in the (de)phosphorylation rates of MAP2 affect the topology of the final dendritic tree. Higher phosphorylation promotes branching and results in trees with many short terminal branches and relatively long intermediate segments. Reducing phosphorylation promotes elongation and inhibits branching.

Models of synaptic transmission at the calyx of Held
Bruce P. Graham 1, Matthias H. Hennig 1, Brian Billups 2, Adrian Y.C. Wong 2 and Ian D. Forsythe 2
1 Dept of Computing Science & Mathematics, University of Stirling, Stirling FK9 4LA, U.K.
2 Dept of Cell Physiology & Pharmacology, University of Leicester, Leicester LE1 9HN, U.K.

In order to understand information transmission in the auditory ‘relay’ pathway involving the calyx of Held synapse and its postsynaptic MNTB target, we have developed a family of mathematical models for synaptic transmission at the calyx based upon electrophysiological recordings of the postsynaptic response to action potential (AP) trains at various physiological frequencies. The aim is to isolate the competing influences on EPSC amplitude, such as vesicle recycling, facilitation and desensitisation. We can then use the models to predict how each factor contributes to modulation of the EPSP during an action potential train, and hence the effect on information transmission. We are investigating two classes of models: (1) a stochastic model incorporating explicit release sites with recycling and release of individual vesicles, and (2) an ‘average response’ model which describes only the average number of vesicles and average EPSC amplitude over all release sites. The more complex stochastic model has been used to investigate how postsynaptic action potential timing may vary during an AP train (Graham et al, Neurocomput. 38-40:37-42, 2001), and the effectiveness of low affinity AMPA receptor blockers in revealing the extent of desensitisation during high frequency trains (Wong et al, J. Neurosci. 23:4868-4877, 2003). This latter study makes the prediction that each active zone may contain 4 or 5 readily releasable vesicles that contribute to multivesicular release in response to the onset of an AP train. The simpler ‘average response’ model has been optimised against EPSC depression curves (Graham et al, Neurocomput. 58-60:449-454, 2004). Very tight fits to the data are obtained over a broad stimulus frequency range when there is significant postsynaptic AMPA receptor desensitisation. These data also predict presynaptic activity-dependent vesicle recycling. This model has been used to examine the information contained in the EPSC amplitudes about the presynaptic interspike intervals of Poisson-distributed AP trains (Graham, ICANN2002, LNCS 2415:45-50, 2002). It demonstrates that the combination of desensitisation and activity-dependent vesicle recycling can act to maintain high information transfer at stimulation frequencies of tens to hundreds of Hertz. Current work with the ‘average response’ model has confirmed that presynaptic mGluR receptors may significantly decrease release probability during a high frequency train.

 

6th May [Schedule]

Non-linear Predictive Modelling For Future Speech Processing Applications
Dr. Mohamed Chetouani , University of Pierre Marie-Curie, Paris, France

Speech signals are special signals since they are produced by human beings. Several studies have been carried out for the processing of these signals. The most used methods are linear in the sense that they do not take into account the non-linear phenomena involved in speech production (e.g. air flow turbulences and so on). Non-linear speech processing is an emerging topic in speech processing. The key idea is to introduce non-linear modelling for the improvement of traditional systems for speech compression, prediction, recognition, speaker and language recognition.
 
The aim of this Talk is to introduce an emerging non-linear speech processing technique for extracting features from speech signals. Conventionally, feature extraction is usually done using linear methods such as Linear Predictive Coding (LPC) or using auditory inspired methods such as the Mel Frequency Cepstral Coding (MFCC) technique. The improvement of the feature extraction process is most important as it is the first stage involved in all speech processing systems irrespective of their final application (that is, speech or speaker recognition etc.). For this purpose, we propose a new model termed Neural Predictive Coding (NPC) which is inspired by the speech production model and can be considered a non-linear extension of the well-known LPC model. This model shows promising improvements in phoneme classification tasks. A novel auditory processing inspired sub-band framework to further enhance the non-linear speech feature extraction process is also briefly introduced.

Previous Seminar Series [Top] [Abstracts] [Schedule]

Last Modified: 17th February 2005