Research Degrees in Computing Science and Mathematics
PhD Studentship Topics
The Department has scope for PhD studies in a number of areas, some of which are discussed below. Funding for studentships is available from time to time. The University is offering some studentships and we may be able to offer a small number of Departmental studentships in 2010. There are, in addition, some SICSA prize studentships available in Computing Science.
If you are interested in applying, please contact the person named in the particular proposal.
Select the appropriate tab below to view all topics, topics relevant to Computing Science, or topics relevant to Mathematics.
Investigating Behaviour through Modelling, Simulation, and Virtual Experimentation
There are many complex socio-economic phenomena that emerge as a result of individual behavioural choices made by large numbers of people. For example, whether a flu outbreak is quickly suppressed or spreads to become a full-scale epidemic depends on individual decisions about vaccination, hand-washing, social contact, and so on. These kinds of behavioural choices are very difficult to study in the real world (we cannot infect a city with flu to study how the inhabitants respond!) Our research involves studying human behaviour using a range of computational techniques such as mathematical modelling, agent based simulation, and virtual experiments using computer games and virtual environments such as Second Life. This is a multi-disciplinary research area and involves collaboration with researchers in Mathematics, Economics, and Psychology. Specific PhD topics can be designed that are tailored towards the student's interests and expertise. Examples of possible topics are:
Using computer simulation to compare models of human decision-making
This project would use agent-based simulation to compare a number of
different models, taken from Economics and Psychology, of how people
make decisions.
Virtual experiments for investigating human responses to epidemics
This project would involve using existing virtual world or computer game
technology to design, create and conduct experiments on human
behavioural responses to a simulated epidemic of an infectious disease.
Neural networks using three-compartment neuron models (two projects)
Neural network research tends either to use networks of very simple neurons (logistic units, or integrate-and-fire units), or alternatively very complex biologically realistic multi-compartment models. We have been developing a network of three compartment neurons (one modelling a distal dendrite, one modelling a proximal dendrite, and one modelling a spike-generating soma) in object-oriented MATLAB. As matters stand, the simulation has a number of types of synapse as well as gap junctions. There are a number of possible directions that this research might take.
- The investigation of the generation of oscillatory local field potentials from such networks. Extracellular recordings of cortical tissue show large oscillatory components, and it is not currently known how these relate to the behaviour of real neural networks. Simulation of realistic architectures of these (relatively) simple units might enable a better understanding of this. It has been suggested that these are critically important in attention. This would involve working with neuroscientists at Newcastle University.
- Using these types of network for processing spike-coded auditory signals. We have developed a spike-coding for auditory signals over a number of years (1) and we are interested in how this type of network might be used to process these signals. This would involve implementing some type of adaptation system (e.g. spike time dependent plasticity) within the network, as well as matching the time constants of the signals to those of the network. This work would be in conjunction with the new EPSRC research grant on A Multichannel adaptive integrated MEMS/CMOS Microphone.
- Learning and adaptation in this type of network. A considerable amount of work has been done on learning algorithms for simple neurons in networks (back-propagation, or Kohonen networks, or radial basis function networks, to name a small number). Little biological evidence exists for any of this type of learning. Instead, there is considerable evidence for local alterations at the connections between neurons (synapses): it is now clear that these junctions though tiny, are very complex, and may adapt pre- or post- synaptically, and have many different forms. It would be interesting to take this 3-compartment model which has a number of different types of synapse object, and experiment with learning and temporal delays in this network.
I would also be interested in supervising other types of research based on early auditory processing and/or computational neuroscience.
(1) see, for example L.S. Smith, S. Collins Determining ITDs using two microphones on a flat panel during onset intervals with a biologically inspired spike based technique IEEE Transactions of Audio, Speech and Language Processing, 15, 8, 2278-2286, (2007).
Second Life: Virtual and real world interaction
Multiple user virtual worlds such as Second Life (SL), offer an immersive environment where multiple users can interact through avatars. Users roam and interact with other avatars through behaviour and an instant message style communication. Some avatars elect to use voice too. To roam, avatars can walk, swim, fly, use moving objects such as cars, bikes, and even teleport at will.
An exciting aspect of SL is the ability to program in-world objects through a scripting language (Linden* Scripting Language or LSL) that is rather C-like in nature. The emphasis is on a state machine approach to provide behaviour in response to stimuli. There is a limited ability through HTTP and XML-RPC to communicate with external servers and devices. This affects the coordination between in-world objects and real-world devices.
This proposed PhD programme will investigate how environments such as SL can be extended to provide a more powerful link between in-world and real-world experiences. The goal is to use portable devices to interact with in-world objects to allow users that are not in-world to maintain a presence and activity in-world. The desire is to provide a more natural interaction rather than simply use an inadequate small screen as a conventional window.
* Linden Laboratories are the creators of Second Life.
Rigorous Decision Support
The aim of this PhD research would be to design a generic and rigorous methodology for creating a clinical DSS (Decision Support System). This would be evaluated in the field of chronic heart disease. The focus would be on developing new techniques for designing technical aspects of decision support, and would be complementary to existing guidelines, models, methods, formats and tools. There is a substantial evidence showing that clinical DSS has the potential to improve practitioner performance, to reduce medical errors, and to improve patient care. The plan is to concentrate on two key aspects of clinical DSS design that need improvement: abstractness and analysis. See the more detailed description (PDF) of this topic for additional background and work plan.
An Individual Interactions Approach to Systems Biology
At Stirling we already have some success in using process algebra to model disease spread. We are keen to take the techniques developed for modelling epidemiology and apply them in the area of Systems Biology.
Systems Biology has become very popular as an application area for theoretical computer science in recent years. Most process algebra approaches are based on stochastic modelling of kinetic reactions in signal transduction pathways. Our approach will focus on the lifespan of certain cell elements. For example, when EGF (Epidermal Growth Factor) binds to a receptor at the cell surface the receptor is internalised and one of two things happens. Either the receptor is destroyed (reprocessed in some way), or is recycled to the cell surface again, without the EGF attached. In both cases as long as the EGF is attached it keeps signalling. Is is possible to investigate signalling levels by treating the process of EGF reception and subsequent death or recycling of the receptor as infection and death or recovery, i.e. as a disease?
A student in this area would be supported by those on the complementary EPSRC grant System Dynamics from Individual Interactions: A process algebra approach to epidemiology, and also by members of the applied formal methods, and mathematical biology research groups. The application area described above is just one example of many possible projects. The technique can be applied to a wide range of biological and computer systems: immunology, epidemiology, peer-to-peer networking, malware analysis, etc. The particular application can be modified depending on the experience and interests of the successful student.
Projects in Computational Neuroscience
Projects are available in the mathematical modelling and computer simulation of aspects of biological nervous systems. Topics range from studying cellular and subcellular details of single neurons, to the dynamics of information processing in neural microcircuits, to developmental aspects of neurons and the networks they form. Tools of the trade include the NEURON simulation package, serial and parallel simulations (72 node cluster available), MATLAB and standard programming languages such as Java. Specific topics around which projects can be defined include (but are not limited to):
- Signal integration in pyramidal cells
- Roles of inhibitory interneurons in neural microcircuits
- Signal transmission properties of chemical synapses
- Development of the electrical properties of neurons
- Development of neuronal morphology
Search Algorithms for Peer-to-Peer Overlay Networks
Peer-to-Peer (P2P) overlay network commonly does not require any central server component for data storage and lookup. This is in sharp contrast to client-server based architectures. Central servers do not scale well in terms of numbers of clients, render the system vulnerable due to the single point of failure, and are expensive to maintain. Due to these disadvantages of client-server based systems, P2P systems have recently greatly gained in popularity. P2P overlay system work on top of the IP network, typically the Internet. P2P overlays feature wide area routing, efficient search of data items, redundant storage, massive scalability, and fault tolerance.
Within P2P there are two different types of networks: structured and unstructured. In unstructured overlays (Gnutella, Kazaa), there is no fixed structure to the topology of the overlay, whereas in structured networks, a Distributed Hash Table (DHT) is used. Unstructured networks are very flexible and do not exhibit any significant amount of maintenance traffic. However, searches in unstructured overlays typically employ flooding algorithms or random walk algorithms. With flooding algorithms there is a good chance the data is found in the network, but at a cost of high volume of traffic. Random walk approaches may not find data which is not duplicated often in the network. However, unstructured overlays support wildcard searches. Searches in unstructured overlays are often referred to as Blind searches as the approach is not affected by what is actually looked for.
Structured overlay networks (Chord, CAN, D1HT, EpiChord, and Tapestry) use DHTs to structure the nodes and to assign data items to particular nodes. In such systems, each node is assigned a unique node ID. This is typically generated encoding their IP address with a secure hash function, such as SHA1. Likewise, data to be stored is assigned a file ID. Again this is generated applying a hash function on the file name or similar keyword. Each node stores data whose ID falls in a certain section of the overall ID space. Nodes locate content using a protocol, often also referred to as routing algorithm. Structured systems include short path lengths to insert/retrieve data and the guarantee that existing data will be found in the network. Structured searches are not blind as the routing of the search is based on the hash of the search string. However, structured overlays do not support wildcard searches where not the full name of the data item is known. An abbreviated name will result in a different hash which bears no relation to the hash of the data item.
This project will investigate wildcard searches in structured P2P overlay networks.
There is some ongoing research on blind search algorithms in structured overlays. However, current blind-search algorithms employ multiple indexing of data items creating a very large key base. Unlike previous work, the approach proposed here is to employ blind searches in the overlay by means of intelligently broadcasting the search. At the first glance this appears to create an enormous traffic overhead in the overlay as a message will need to be sent to every node. However, some overlays employ intelligent maintenance algorithms to inform nodes of a node joining or leaving the network. One such maintenance algorithm is EDRA* used by D1HT. This project will investigate the viability using such maintenance algorithms for wildcard searches in structured P2P overlays. An important research question will be finding solutions to minimise the message overhead.
Natural Computing / Computational Intelligence
A number of research topics are available, including in the following six interdisciplinary research areas:
Title 1:
Novel Computational Intelligence Methods for Immune System Modeling and
Analysis (in collaboration with the BBSRC Research Network on Immunology
Imaging and Modelling, I2M)
Title 2:
Novel Computational Intelligence Techniques for Real-world Problem Solving
e.g. in the medical, defense or business (such as financial and
telecommunications) industries - depending on the topic, this research
could be in collaboration with the Harvard Medical School, MIT Media Lab,
Boston, USA, and Sitekit Labs Ltd., Scotland.
Title 3:
Multi-modal (audio-visual) processing methods to improve the next generation
of telecommunications services (including intelligent avatars, remote health
monitoring and interactive dialogue systems), and development of new speech
processing (i.e. enhancement, analysis, synthesis and recognition) methods
including for foreign languages (Arabic, Urdu, etc.) - in collaboration with
the European Science Foundation (ESF) funded European Research Network
(COST-2102).
Title 4:
Neurobiologically inspired Cognitive Modeling and Control for Complex
real-world industrial & medical applications (such as, robotic control,
autonomous vehicle control, insulin regulation of blood sugar & diabetes
etc.) - in collaboration with the Department of Computational Neuroscience,
Sheffield University.
Title 5:
Common Sense Computing (including intelligent agents, natural language
processing, statistical machine learning, semantic data mining and multi-modal
HCI methods) for developing Next Generation Intelligent Web Applications
in e.g. e-health. e-health, e-tourism etc. (in likely collaboration with
MIT Media Lab, Harvard Medical School, Boston, USA, and Sitekit Labs Ltd.,
Scotland)
Title 6:
Non-linear Computational Intelligence based Signal Processing algorithms for
challenging real world applications such as high-resolution detection and
localization of multiple (non-stationary moving) targets and broadband signal
separation (in collaboration with the Centre of Excellence in Signal Image
Processing, Strathclyde University)
For further information on any of the above PhD research topics (or to propose any of your own ideas/suggestions), please contact Dr. Hussain or visit his webpage as below.
Prompting for Cognitive Impairment
The aim of this PhD research is to design a voice-based prompting system that is useful for those with cognitive impairment (e.g. dementia, brain injury and learning difficulties). This would be based on the existing Guide and Cress systems. Guide has been developed by Dr. Alex Gillespie (Psychology) for talking users through activities of daily living, and has been prototyped as a means of helping those with dementia to prepare food and to don a prosthetic limb. Cress has been developed by Prof. Ken Turner (Computing Science) for a designing a wide variety of services including Interactive Voice Response systems. The aim is to combine, extend and evaluate the strengths of both systems in a practical and worthwhile application. See the more detailed description (PDF) of this topic for additional background and work plan.
Dynamics and Control of Wildlife Diseases
1: Optimal control strategies in disease systems - In particular I have been studying vaccination in domestic dogs in Ethiopia carried out to protect the endangered Ethiopian wolves. In this case there is a high turnover in the dog population and pulse vaccination occurs in villages in geographically widely distributed villages. We will build a theoretical model of pulse vaccination under this scenario and then parameterise it for the system in Ethiopia and look at how frequently we should vaccinate and what the optimal percentage of dogs to vaccinate is. This project scales up from information about individual villages to the entire dog population. This would utilise the biological expertise of project partner Fiona Matthews from Exeter and colleagues in Wildcru at Oxford.
2: Predicting transient disease dynamics and how they might effect disease control strategies - Most models of infectious disease spread look at the long term behaviour of the system and not at the short term dynamics; however, the latter is what is more useful biologically, particularly when investigating control measures. However, studying short term dynamics is more mathematically challenging.
3: Emerging diseases and wildlife reservoirs - Emerging diseases are defined as those which are affecting new hosts or appearing in new geographical areas. It is often the case that these diseases are passed to humans from wildlife due to changes in land use which cause increased contacts between humans and wildlife. It is important to understand how and why diseases emerge and to be able to predict the emergence of new diseases. Determining how to control these diseases is also vital.
4: Models of bumble bee parasites - Stirling is the home of the Bumblebee conservation Trust and this project would make use of this fact and the expertise that we have here. The first part of the project would look at the available data and model the dynamics of a parasitic infection within a nest, we will then change scales and look at how to use the single nest model to determine between nest transmission and the population level impact of parasites on bumblebee populations. This is an important problem as bumblebee populations are declining and they are important pollinators of our agricultural crops.
Evolutionary and ecological systems
1: Control strategies in multi-host and shared pathogen systems - Biological systems where several host species share a common disease (for example, both cattle and badgers are vulnerable to Bovine TB, or those that cross from wildlife to human) is widespread, and with climate change and globalisation this is becoming an increasing problem. Here control (eradication) of the disease can be critical, however developing a strategy can be difficult. Problems may include situations where controlling (e.g. culling) all the species is not a viable option and so the question arises as to whether the disease can be removed from the system by only controlling a select number of host species. In addition, these problems naturally extend into optimisation problems with an aim of, for example, minimising the number of individuals culled/vaccinated or minimising the total cost.
2: Co-Evolution of Mating Conflict - Natural selection drives the evolution of species towards an optimum state with respect to their current environment. However sexual conflict over mating can often shift species from such optima as individuals attempt to gain their own way during reproductive encounters. Traits that evolve through sexual conflict can reduce the fitness of the opposite sex. For example, males can harass females into accepting unwanted and costly matings; in response females develop both morphological and behavioural traits that enable them to resist unwanted copulations. This antagonistic co-evolution of male and female traits can result in the exaggeration of traits that reduce the fitness of the species as a whole. The studentship will be in collaboration with School of Biological and Environmental Sciences and involve developing established models of co-evolution so that empirical data can be easily entered into them to test their robustness and gain new insights into sexual conflict, natural selection and their interaction.
