PhD projects

The aim of this project is to investigate methods for robustly tracking interest points through 2D and 3D video sequences. Such techniques are highly important to a range of core computer vision and computer graphics techniques both in academic research and in industry for video games and film production. Examples include 3D camera movement extraction and markerless motion-capture.

The project will be based at the Media Technology Research Centre (MTRC), Department of Computer Science, in the University of Bath, and will have potential for strong overlap with industry via the recently funded Centre for Digital Entertainment.

Dr Darren Cosker

The aim of this project is to investigate methods to automatically animate faces in a range of styles using just an actors voice. The research will contribute to a range of research areas, including machine learning, speech analysis, animation and computer vision.  There are strong potential applications in video game and film production, such as automatic cartoon and photorealistic character animation. The project will be based at the Media Technology Research Centre (MTRC), Department of Computer Science, in the University of Bath, and will have potential for strong overlap with industry via the recently funded Centre for Digital Entertainment.

Dr Darren Cosker

The aim of this project is to acquire detailed 3D measurements of deformable objects using 2D video and computer vision techniques. Acquisition of 3D scene data is a highly important problem for a range of tasks including object classification, animation and rendering.   Such techniques also have strong value in industrial applications such as markerless motion-capture and object modelling for video games and films. The project will be based at the Media Technology Research Centre (MTRC), Department of Computer Science,  in the University of Bath, and will have potential for strong overlap with industry via the recently funded Centre for Digital Entertainment.

Dr Darren Cosker

Dynamic textures - water, clouds, trees, fire and such like are amongst the most ubiquitous yet least understood visual phenomena. They are of interest to both computer vision and computer graphics because their diversity and complexity offer significant a intellectual challenge. They are of interest to industry, who want to control them for films, games, architerual and landscape modelling and other applications.

The PhD will analyse video of real dynamic texture so as to aquire a model from which new instances can be readily created and animated in a wide variety of styles; see the show-reel for an example. The student will be part of the Media Technology Research Centre and collaborate with the industrially aligned Centre for Digital Entertainment. Sound programming skills and strong mathematical acumen are prerequisites.

Dr Peter Hall

Modelling virtual cities is becoming easier, with commercial software now able to help with this. Building models of particular buildings is harder. Capturing the inside as well as the outside is beyond current software - simple things like doors which open. Building models of cities which function - working traffic lights, traffic flows varying with time of day, office lights which go on and off, rivers which flow - is also not well-addressed. There is scope here for several research projects. We also have active links with major game companies and heritage organisations to help us.

Professor Phil Willis

The computer analysis of sound is a subject that stretches back for very many years. There are many ideas that have been tried and discarded as being to impractical: various transformations that are nice but can't be done fast enough (ideally real time) to be useful. For example, rather than using a standard Discrete Fourier Transform to analyse a sound we would like to use the more expensive psycho- acoustically "Tuned Q" transform that is not so mathematically convenient, but matched the human ear much more closely.

The continued march of Moore's Law means that multicore computers are becoming cheaper and more accessible. Massively parallel machines are becoming available at a reasonable cost. This is a continuing project on "HiPAC", or "High Performance Audio Computing", where we apply modern hardware to old (and new!) ideas in audio analysis.

This will require you to:

a) be comfortable with basic mathematical ideas like Fourier Transforms
b) be comfortable with programming parallel machines Having an appreciation of the theory of Music (and Audio in general) is a bonus

Dr Russell Bradford

In the near future we will have intelligent homes that will assist us in our daily lives. We can imagine robot or avatar helpers assisting in our care-giving tasks for our children and other infirm relatives. We would like digital assistants that can maintain our homes as well as our appointments. Some of us would like better opponents in AI games, or more companionship from our digital memorials to lost friends.

All of these applications require significant expertise in their specific domain. Such expertise is neither cheap nor readily available. Thus another complementary need will be vast demand for customisation to individuals and households. Machine learning alone cannot do handle problems of this magnitude. Rather, we need a Systems AI approach which takes advantage of automated learning and planning but also the software engineering skills of developers. Just like plumbing, AI will need to be installed and maintained, ideally by people who are domain experts (e.g. nurses or creative artists) rather than people with first-class university degrees in computer science.

We are looking for PhD students to extend our existing work on programmability for AI into any of the domains mentioned above, or another area matching the students own expertise or interests. Prospective students should have strong interest in an application domains, a good familiarity with real-time autonomous AI systems and / or strengths in machine learning, HCI and GUI design.

Dr Joanna Bryson

Do cultures solve problems collectively? If so, why do human cultures vary? How many species exploit strategies like these? How do eusocial species collective decision making strategies differ from ordinary social behaviour? Why are societies often characterisable by relatively few "factors" of variation between individual behaviour, and what determines these axes of behaviour?

Bath's Artificial Models of Natural Intelligence provides a unique environment for exploring the computational and evolutionary consequences of social behaviour, exploiting both mathematics and complex spatial and algorithmic simulation modelling, and grounding our models in a vast network of collaborations with empirical scientists. We are looking for PhD students willing to join our quest to understand intelligence. Areas of particular interest include:

• the impact of different species of macaque's varied social structures on their biological evolution and their spread of intelligence
• understanding how ecological constraints and rates of change may account for regulated limits in the size of culture and individual memory capacities
• looking at the computational properties of documented oscillations between imagistic and doctrinal religious behaviours
• understanding mobile genetic elements as a model of eusocial learning in bacteria
• understanding and exploiting the collective measurement and computation shown by ants

Candidates should be strong programmers with good knowledge of theoretical computer science and/or have masters degrees in evolutionary or mathematical biology. They should be avid readers willing to work hard to maintain currency across a variety of disciplines, yet be capable of sufficient focus and drive to become domain experts in one particular sub-problem area. They should be also be sufficiently social to form and maintain good collaborative relationships across disciplines.

Dr Joanna Bryson

Norms describe what, when and where an agent may do or say (communicate) something.  A (probably consistent) collection of norms constitutes an institution or an organization. An agent may choose to play a role in a particular organization because it should enable the achievement of the agent's goal.  Research issues: building effective models of existing organizations; authoring and verifying new organization descriptions; theories, tools and methodologies for organizational modelling.  Application areas (at Bath) include: situational awareness, business processes, energy management and music performance.

Dr Marina De Vos and Dr Julian Padget

Norms describe what, when and where an agent may do or say (communicate) something. A collection of norms constitutes an institution or an organization.

The norms are not necessarily fixed for ever. Certainly a programmer could change them, but it is perhaps more appropriate that the framework should monitor itself and consequently identify norms for revision (deletion, replacement). Research issues include: how to modify the norms in a deployed system (previously acceptable actions may be rendered unacceptable) and how to identify what to change and how to change it. Application areas (at Bath) include: situational awareness, sensor networks, business processes

Dr Marina De Vos and Dr Julian Padget

Norms describe what, when and where and agent may do or say (communicate) something.  A (probably consistent) collection of norms constitutes an institution or an organization and an agent may choose to play a role in a particular organization because it should enable the achievement of the agent's goal.  Research issues: building agents that can utilize or even understand the normative framework(s) under which they are operating.  Application areas (at Bath) include: situational awareness, business processes, energy management and music performance.

Dr Marina De Vos and Dr Julian Padget

These are systems that can change their behaviour in response to changes in their environment for the purpose of maximising some (or even more than one) objective function.  Research issues include: learning relationships between system indicators; detecting significant events; integrating new events; discarding old events; effective filtering of indicators to identify what really matters; coping with change in relevance over time.  Application areas (at Bath) include: intelligent manufacturing, sensor networks, large-scale self-managing software systems.

Multi-tasking is an everyday reality for many people.  Studies of information workers have revealed that they switch between tasks – because they are interrupted, or because they choose to interrupt themselves -  incredibly frequently.   Yet we understand relatively little about how people manage multiple-tasks, and how they may be supported by technology.

My particular interest is in “discretionary interleaving”  - the propensity of people to switch back and forth between tasks, despite the cognitive costs.  Having done some laboratory-based work using simple, artificial tasks, I’m now interested in carrying the ideas into real work environments.  I would be interested in supervising students who wish to understand how workers manage multiple tasks, how they choose when to give up one task in favour of another, and how computer systems could be designed to facilitate such working patterns.

Professor Steve Payne

It is widely recognized that design of interactive technologies must be sensitive to the needs of all users.  This creates challenges for design when groups with particular cognitive profiles are concerned, such as the elderly.  Most work on Accessible design for older users focuses on the cognitive and perceptual declines that are often associated with ageing.  My interest is rather on motivational changes, and the challenges these pose for design.

It is now widely recognized that older adults are less motivated to seek variety or novelty in their consumption experiences. They are more inclined to seek familiar, emotionally predictable experiences.  But much work remains to be done to understand the causes and consequences of this general bias,  and I believe this can be fruitfully explored in the context of exploratory learning of new technological devices. I have begun to explore these issues in collaboration with the BBC concerning the design of interactive TV services.  I would be interested in supervising students who wish to understand the psychology of motivational changes during ageing, and how design of interactive technologies might be responsive to this issue.

Professor Steve Payne

Modern networked system architectures have created enormous potential for collaborative and resilient communities of people and organisations sharing information, knowledge and ideas. How much of that potential is actually being realised?

The development of social networking and similar environments for sharing and exchanging information, knowledge and ideas show that it is possible to bring large collections of people together. However these technologies do not readily allow people to come together as a creative and collaborative community. The fluidity of virtual organisations, their memberships and information flows, challenge our traditional means of providing the policies and infrastructure that generate confidence in their resilience to failures and threats.

How can we create collaborative systems that enable people to creatively engage with each other, to produce solutions to some of the complex problems we face, to increase the creative capability of communities and to bring about a greater level of social inclusion?

What must we do to overcome the cultural, organizational and technological barriers that prevent communities of people from obtaining “collaborative flow”? How can we develop such flexible systems so that they exhibit the dependability and resilience that we require?

Professor Peter Johnson

Distributed teams work routinely together through intermittent-use technologies such as email and instant messaging, Google docs and Microsoft Sharepoint. However, there have been some efforts to create always-on technological connections between physical locations in which they work, to promote spontaneous interactions and feelings of mutual presence. These are have been described as virtual office- shares or media spaces, and are closely related to telepresence systems. However, there is a trade-off to be made between personal privacy in these technological 'panopticons' and the connectedness for which they have been designed. This project will investigate attentionally triggered mechanisms for controlling availability to manage that trade-off.

Dr Leon Watts

Wiki technologies allow permanent tracking of the history of a document. They record diff's in versions as well as who made changes and when they were effected. This project will examine technical and empirical methods for constructing wiki-type collaborative environments that are intended to support the evolution of decision- making processes in groups. It will in particular examine how group members understand ownership and the chronology of decision-making as a function of the facilities and evidence provided by their collaborative environment.

Dr Leon Watts

Smart grid technologies can help electricity generators and transmitters to decrease the power they are obliged to supply at moments of peak load, repair networks after unexpected failure and reduce the prices charged to end-users. More specifically, by increasing generator interaction with a) customers and b) computerised actors representing those customers, smart grid technologies have the potential to enable collaboration between those stakeholders that results in the shifting of demand away from moments of peak load and away from periods of component or network failure. For those benefits to accrue, however, both the collaborations between human and computerised (or 'automated") collaborators and the interactions, which support those collaboration must be carefully designed. Our recent work on human-human, human-autonomous system and autonomous system-autonomous system collaboration provides a basis for investigating the requirements of this design, providing initial insight into the importance of group, task and resource structure, the development of collaboration over time and the costs of collaboration. Extending this research to the point that we can model, support and assess the requirements of electricity grid technologies is, however, a non-trivial exercise.

The person undertaking this PhD will undertake this extension by considering the potential benefits of increased collaboration between electricity network stakeholders, modelling the capabilities and interactions needed to support those increased collaborations and the identifying the relative costs of non-,
partial- or complete-implementation of the supporting interactions. 

That work will draw upon previous research into:

• human-computer interaction
• collaboration
• complex systems

It will address specific problems within the electricity domain but will also contribute to a wider understanding of activity and group management in domains involving human-autonomous system activity (e.g. the provision of other utilities, aviation services and health care).

Prof Peter Johnson and Dr Rachid Hourizi

The fundamental result from which categorical logic sprang in the 1960's is an equivalence between the category theoretic notion of Lawvere theory and the logical notion of equational theory. The notion of Lawvere theory was also seen to be equivalent to that of finitary monad on the category Set.

The definition of finitary monad immediately makes sense on a wide variety of categories other than Set, in particular (subject to a mild extension from finiteness to countability) on the category of omega-cpo's, which has been fundamental to the study of denotational semantics. So one wonders whether the ideas of Lawvere theory and equational theory can be extended in a way that one can readily use in practice and that extends the above equivalences.

In fact, over recent years, the definition of Lawvere theory has been extended in a numbers of ways that retain the equivalence with monads and do indeed play a helpful role in denotational semantics, in particular in regard to modelling computational effects that appear in otherwise functional languages, such being for example side-effects and exceptions in ML.

But what about extending the definition of equational theory? An answer is not entirely straightforward. The key problem is the notion of arity: in ordinary equational theories, each operation has an arity given by a natural number. Extending that to account for countability is routine. But for an axiomatic treatment, the most immediate natural notion of arity has more complex structure than that of a natural number, and that plays havoc with the ideas of composite operation and equation between composite operations.

I believe that is now resolvable by focusing on what are called discrete Lawvere theories. The proposal is to see this idea through to resolution.

Dr John Power

The derivative of a function at a point is defined as the slope, or more formally as the Cauchy Limit, and so on. But, when we actually calculate derivatives, we behave like a computer algebra system and push symbols around (Chain Rule, Product Rule etc.). Similarly, we regard 'log' as a function from the complexes to themselves, such that its derivative is 1/z, and therefore log(ab)=log(a)+log(b) and other nice algebraic properties, and conveniently ignore the fact that there is NO continuous function with these properties, and a brnach cut has to intervene somewhere.

Previous work has shown that it is helpful to look at the decomposition of the complexes as a two-dimensional real plane, and analyse the geometry of the branch cuts. This work would build on [1] [2] and the work of the supervisor during his sabbatical at Waterloo (Ontario).

Professor James Davenport

We all know that sqrt(x^2) = x, so what is wrong with 1 = sqrt(1) = sqrt((-1)^2) = -1 ? It so happens that sqrt(x^2) = -x when x is a negative real number.

There are many "rules" we assume are true that are in fact false. For example, it is not true that log(xy) = log(x) + log(y) for all x and y.

This is a continuing project on the correct simplification of algebraic expressions: when, precisely, is sqrt(x^2) = x, and when is it not? Computer Algebra systems traditionally shy of this aspect of computational algebra.

It is a surprisingly subtle subject that requires you to have:

a) a solid understanding of degree-level mathematics
b) a healthy amount of experience of programming in order to implement the algorithms that arise

Exercise: One of the below is correct. Which one?

1. sqrt(x-1)sqrt(x+1) = sqrt(x^2-1)
2. sqrt(1-x)sqrt(1+x) = sqrt(1-x^2)

Dr Russell Bradford

Game semantics is an approach to mathematical modelling of programming languages which interprets a computation as a play in a simple two-player game. In the last 15 years, game semantics has demonstrated itself to be a flexible and powerful approach.

Most games models are purely qualitative: they model what happens, but not how good the outcome is. Some models include a notion of "winning": a simple value judgement stating that a program execution is good in some way, typically meaning error-free and terminating.

This project is to investigate how further quantitative information can be incorporated into games models, reconciling the separate fields of game semnatics and game theory. Preliminary investigation has shown that simple forms of information security can be analysed by adding payoffs to semantic games, corresponding to security levels. In this project, we will study this model further, attempting to extend it with more detailed notions of payoff, and analyse notions from traditional game theory such as optimal and equilibrium strategy to identify their computational significance.

Professor Guy McCusker

"Higher-Order" references, which can be used to store functions or objects, are a powerful programming language feature. This expressiveness cna make it difficult to write correct programs, or reason about them. A particular problem is interference between variables evident in effects such as aliassing --- two names which refer to the same location.  One way to avoid these problems would be a form of "interference control" for higher-order references; in the case of references holding basic data such as integers, elegant type theories have been designed which prevent interference whilst retaining useful expressive power. However, these theories do not translate straightforwardly to higher-order references (essentially because they allow function calls out of scope).

Game semantics has been used to give precise denotational models for many sophisticated programming language features, including higher-order references, and (separately) interference control for integer state.

Recent research has studied the structure of these models, leading to new categorical semantics and type theories  for describing information flow in them. The aim of the project is to build on this work, to describe a model of interference control for higher-order references and to use it to discover typing rules guaranteeing interference control.

Dr Jim Laird

Consider the following, broad problems, the first of which is practical and the second theoretical:

1) It would be nice to do research in mathematics by interacting with computers, leaving to them the menial tasks, like verifying the correctness of proofs, or applying routine techniques. This is probably achievable in the next 20 years, but it will be necessary to improve on our ability of representing proofs, because right now the computer cannot make the difference between two proofs employing the same ideas, but differing in inessential details.

2) We know very little about the size of proofs. For example, given big proofs of certain statements, can we compress them to manageable size? A negative answer to this question would imply that P is different from NP, and this is obviously a very difficult problem. 

However, many subproblems of a similar kind are not so difficult, and will all require to improve on our current knowledge of proof size.

We are making progress in these two very different research areas (and others) by studying a new representation of proofs that is based on geometry rather than syntax. This has been made possible by recent advances in the area of deep inference, and, in particular, by the achievement of full locality in proof theory. Several projects are available, for bright computer scientists and mathematicians.

Ours is an international, growing, well financed, cooperative, young and enthusiastic research community.  I encourage you to get in touch with my students and ask them about their experience with deep inference.

Alessio Guglielmi

Today's computer networks require concurrent programming languages.  These languages coordinate the activity of asynchronous entities, and by their nature induce programs that are difficult to design and verify. For example, it is currently impossible to guarantee that an operating system will not crash. One of the challenges of computer science is to get concurrency under control. This can be achieved by inventing languages with rich mathematical structure and a good theory.

A great success in non-concurrent language design has been the discovery of a correspondence between the lambda calculus and the mathematical theory of formal proofs, called proof theory. This has led to the creation of functional languages, in particular through the concept of data type, which is now recognised as one of the fundamental tools of computer science.

We think that this correspondence can be extended to the more challenging case of concurrent languages, so helping to solve, perhaps in a crucial way, the problem of controlling concurrency.

The proof-theory area of deep inference (recently conceived by myself and my colleagues) comes equipped with some key features that already led to considerable progress towards solving the problem of capturing concurrent computation via proofs. I propose to improve on these earlier results and achieve a complete, useful correspondence. Several projects towards this broad objective are available, for bright computer scientists and mathematicians.

Ours is an international, growing, well financed, cooperative, young and enthusiastic research community.  I encourage you to get in touch with my students and ask them about their experience with deep inference.

Alessio Guglielmi

Answer Set Programming (ASP) is a declarative programming paradigm. Like other declarative programming languages, the programmer specifies what needs to be achieved, rather than how it should be achieved.  In ASP, programs are written in AnsProlog and describe the requirements for the solutions of certain problem.

AnsProlog is increasingly being used to solve practical problems both in and out of the academic domain. To advance the field the programmer needs adequate support.

Comprehensive programming methodologies have been developed for other programming paradigms. It is not clear how, or even if, such methodologies can be applied to ASP.

Dr Marina De Vos

In Mobile and Pervasive Computing, we are interested in understanding, designing, implementing and evaluating truly pervasive systems. As is common practice, we use the terms Pervasive and Ubiquitous Computing more or less interchangeably but we prefer Pervasive as it better conveys our ultimate vision: enabling and empowering services pervading our lives, our environments and our societies.

In this research, we are interested in developing fundamental theory, principles, design tools and methods by taking into account a range of technical, social and other factors. Our goals include understanding, designing, building and evaluating complex interactive systems involving many people and many technologies. Within the mobile and pervasive arena we ask questions such as:

• How do we design for usability when the human-computer interface is dispersed and interwoven throughout our environment?
• How can we understand and account for the web of influences amongst society, environment and technology?
• How do we interact successfully with and through devices and networks with many form factors?
• How do we design these devices, networks and services, in terms of both process and product?

We are particularly focused on the relationships between mobile and pervasive technologies and innovative interaction methods, including 3D interfaces and gestural interaction. We are also interested in making pervasive systems context-aware, developing fundamental theory for context-aware systems and reflecting this in working architectures and applications.

Eamonn O'Neill