Advanced Reasoning
Design verification of complex electro-mechanical systems
This project will design and develop tools based on qualitative reasoning that will present to engineers understandable representations of the systems that they have designed. For example, say a group of engineers had designed an airbag system for a car, the tool might present to the engineers a diagram showing all of the states that the airbag system could be in. This is important, because it will help the engineers understand the safety implications of the system that they have designed.
The project will involve both system modelling of the kind that our group has a long record of [1, 2, 3], and also the application of HCI techniques to present the results to the users in the most understandable forms.
Proposed Supervisor: Dr Neal Snooke. Also the work fits closely with some of the other research of the group, and so there will be a number of staff able to help and advise.
References
[1] C. J. Price and N. S. Taylor, Automated Multiple Failure FMEA, Reliability Engineering and System Safety 76(1), pp1-10, 2002.
[2] P. Struss and C. J. Price, Model-based systems in the automotive industry, AI Magazine special issue on Qualitative Reasoning, pp17-34, vol. 24(4), Winter 2003.
[3] N. Snooke, Simulating electrical devices with complex behaviour, AI Communications, vol 12(1-2), pp45-59, IOS Press, 1999.
Improving machine learning through qualitative reasoning
Model building is a difficult and time consuming process. A much more efficient alternative to building models by hand would be to learn models from observed data. This is still a very difficult machine learning challenge for complex domains. Qualitative reasoning can help with this by enabling us to derive quantitative models from qualitative models. In domains where qualitative models are known, but are not executable, qualitative reasoning should provide graphical ways of building executable models (e.g. [1]), and make clear the assumptions behind the models, enabling domain experts to compare their models on a like-for-like basis. Where models are known and data is available, this type of work should make it possible to develop accurate numerical models with known assumptions and limitations (e.g. [2]).
On this project we will be looking to apply these techniques to complex real world systems, such as the systems being researched by our computational biology research group (e.g. [3]).
Proposed Supervisors: Professor Qiang Shen, Professor Chris Price
References
[1] Forbus, K., Carney, K., Sherin, B., Ureel, L. 2004 VModel: a visual qualitative modeling environment for middle school children, In procs 16th Innovative Applications of Artificial Intelligence Conference, 2004.
[2] R. Bellazzi, R. Guglielmann, L. Ironi: How to improve fuzzy-neural system modelling by means of qualitative simulation, IEEE Trans. on Neural Networks, vol. 11, no. 1, 249-253, 2000.
[3] King, R.D., & Clare, A.J. (2004) Confirmation of data mining based predictions of protein function. Bioinformatics, 20 1110-1118
Reasoning about deviations in qualitative flow
Qualitative reasoning is very good at reasoning about physical systems in many circumstances. Our own research has concentrated on being able to reason about failures in electrical systems [1, 2, 3], and has been successful both technically and commercially. Qualitative reasoning has some shortcomings in reasoning about changes to components short of complete failure. This proposed research will look at how changes to the size of an influence is best be propagated qualitatively through a system. A good example of where this would be useful is in a hydraulic system - the kind of changes that are useful to reason about are "the size of a pipe is reduced" or "there is a leak which reduces the pressure within the system".
Proposed Supervisor: Professor Chris Price. Also the work fits closely with some of the other research of the group, and so there will be a number of staff able to help and advise.
References
[1] C. J. Price and N. S. Taylor, Automated Multiple Failure FMEA, Reliability Engineering and System Safety 76(1), pp1-10, 2002.
[2] P. Struss and C. J. Price, Model-based systems in the automotive industry, AI Magazine special issue on Qualitative Reasoning, pp17-34, vol. 24(4), Winter 2003.
[3] N. Snooke, Simulating electrical devices with complex behaviour, AI Communications, vol 12(1-2), pp45-59, IOS Press, 1999.
Using spatial aggregation for recognising significant medical problems
Our visualisation group has an excellent record of research in automating the recognition of malignant growths such as breast cancers (e.g. [1]). Medical advances mean that ever more detailed 3D data is available, often over time, giving a 4D data set. The most effective interpretation of such data requires skills that are possessed by very few experts. This project will investigate the application of qualitative reasoning (QR) methodologies for spatial/temporal reasoning [1,2] to support experts in identifying salient features in 4D data sets of this kind.
Proposed Supervisors: Dr Reyer Zwiggelaar and Professor Chris Price.
References
[1] R. Zwiggelaar, S.M. Astley, C.J. Taylor and C.R.M. Boggis, Linear structures in mammographic images: detection and classification, IEEE Transaction on Medical Imaging 23 (9), pp. 1077-1086 (2004)
[2] C. Bailey-Kellogg and F. Zhao. Qualitative spatial reasoning: extracting and reasoning with spatial aggregates. AI Magazine special issue on Qualitative Reasoning, 47-60, vol. 24(4), Winter 2003.
[3] L. Ironi, S. Tentoni: Towards automated electrocardiac map interpretation: an intelligent contouring tool based on Spatial Aggregation, Lecture Notes in Computer Science, 2810, 397-408, 2003.
Electrical circuit analysis tools using qualitative reasoning techniques
We have had considerable success with the application of qualitative reasoning methods applied to electrical systems. The main idea is that instead of calculating numeric voltages and currents, we use symbolic values (such as high/low or active/inactive) that allow engineers to find the most critical or significant parts of a circuit. This novel method [1, 2] has been developed into commercial quality software and is widely used in the automotive industry [3]. However there are many limitations and we are researching new ways of modelling circuits and enhancing the basic technique. One important modelling aspect is how the symbolic values relate to the real circuit and we are keen to improve our models and algorithms to give finer detail [4]. This project will investigate a new circuit model, build and test implementations and evaluate the ideas against more conventional methods. There are several other active researchers in this area, who also collaborate with relevant companies - these will provide a supporting context for the project.
Proposed Supervisors: Prof Mark Lee, Prof Qiang Shen
References
[1] M. H. Lee, Qualitative Circuit Models in Failure Analysis Reasoning,, AI Journal. Vol. 111, pp239-276, 1999.
[2] M. H. Lee, Qualitative Modelling of Linear Networks in Engineering Applications, in Proc. ECAI'2000, 14th European Conf. on Artificial Intelligence, Berlin, August 19th - 25th, pp161-165.
[3] C. J. Price and N. S. Taylor, Automated Multiple Failure FMEA, Reliability Engineering and System Safety, 76(1), pp1-10, 2002.
[4] M. H. Lee, On Models, Modelling and the Distinctive Nature of Model-Based Reasoning, AI Communications, 12 (3), pp127-137, 1999.
Crime Risk Prediction
This project will examine real-crime risk and the fear of crime from a modelling perspective [1]. It will develop and utilise fuzzy systems methodologies to generate crime-risk surfaces (i.e. diagrams that put estimated risk in a geometric/spatial context) and then examine the relationships between these and fear of crime. The research will help improving the utility of current Geographical Information Systems (GIS) [2].
These aims will be achieved through a rethinking of the kinds of ways GIS and risk-models can generate, store, and manipulate data [3]. In particular, the project will develop models of risk, induced from real data with respect to expert advice; build tools for capturing perceived and/or vernacular spatial entities associated with risk perception (e.g. ``high crime areas''; ``bad part of town''); and investigate and develop techniques that may be used to manipulate, compare and represent such risk surfaces, perceived entities, and real data.
Proposed supervisor: Professor Qiang Shen
References
[1] Farrall, Bannister, Ditton and Gilschrist. Social psychology and the fear of crime: Re-examining a speculative model. British Journal of Criminology, 40:399-41, 2000.
[2] Evans and Waters. Mapping vernacular geography: Web-based GIS tools for capturing ``Fuzzy'' or ``Vague'' entities. International Journal of Technology, Policy and Management, 7(2):134-150, 2007.
[3] Shen, Keppens, Aitken, Schafer and Lee. A scenario driven decision support system for serious crime investigation. Law, Probability and Risk, 5(2):87-117, 2006.
Qualitative and fuzzy reasoning for automated diagnostic systems.
Recent work at Aberystwyth has resulted in tools that are able to produce diagnostic symptoms for electrical and fluid flow based systems such as those found in theautomotive and aerospace applications [2]. This work is based on qualitative simulation and automated Failure Modes and Effects Analysis (FMEA), and provides results that are very general and easy to interpret by an engineer. An example might be "When the pump is switched on the flow in pipe X is low". For real time diagnostic, monitoring, and prognostic applications these qualitative symptoms must be mapped onto system observations. The simple approach of thresholding observed values into qualitative ranges such as high, low, normal has several practical disadvantages and can lead to undesirable diagnostic artefacts. The work being proposed will initially aim to investigate the possibility of using other techniques to improve the diagnosis and prognosis of faults. One technique of particular interest is the use of fuzzy values and fuzzy reasoning [1] to allow a fault ranking to be made when a number of observations and possibly conflicting symptoms are available. The automated generation of symptom sets for complex systems, and analysis of the sensor placement for maximum diagnostic utility and or fault isolation are also areas of research that are closely related and may form part of the work. This work is likely to require some mathematical and programming interest and ability.
Proposed supervisors: Dr Neal Snooke, Professor Qiang Shen, Professor Chris Price.
References
[1] T. J. Ross. Fuzzy Logic with Engineering Applications. Wiley, 2005.
[2] N. A. Snooke, C. J. Price, Integrating Reliability Analysis and Diagnostics for Complex Technical Systems, Proc. IMechE, Part O Journal of Risk and Reliability, Vol 221(2), pp.153-159 July 2007.
Hybrid Evolutionary Computation Techqniues for Feature Selection
One of the main obstacles facing many intelligent systems applications is that of high dataset dimensionality. To enable such systems to be effective, a redundancy-removing step is usually carried out beforehand, with an aim for those and only those input features that are most predictive of a given outcome to be identified for use. Approximation-based techniques have proven useful in implementing such feature selection mechanisms (e.g. [1, 2]).
Whilst having been generally recognised as a multi-objective search problem, feature selection has not been extensively researched using alternative computation intelligence approaches. This project proposes to investigate and develop a hybrid variant of conventional evolutionary computing methods [3] to improve both effectiveness and efficiency of feature selection. This work will have a significant potential in a wide range of application areas, including a variety of problems such as gene sequence, web content and medical image classification.
Proposed supervisors: Professor Qiang Shen, Dr Jun He
References
[1] Q. Shen and R. Jensen, "Selecting Informative Features with Fuzzy-Rough Sets and its Application for Complex Systems Monitoring,"
Pattern Recognition, vol. 37, no. 7, pp. 1351-1363, 2004.
[2] R. Jensen and Q. Shen, "Fuzzy-Rough Data Reduction with Ant Colony Optimization," Fuzzy Sets and Systems, vol. 149, no. 1, pp. 5-20, 2005.
[3] H. Dong, J. He, H. Huang and W. Hou. Evolutionary programming using a mixed mutation strategy. Information Sciences. 177 (1): 312-327, 2007.