This group changed its name as it evolved. It started, back in the 1980s, when the robotics research group realized that software models could be very useful for analysing errors and diagnosing failures. The robotics group was trying to build robot control software that could automatically diagnose errors and find ways of recovering from them. This was in the context of automated assembly: programming robots to assemble industrial components from a collection of small parts. Rule-based approaches were very popular then (and commercially lucrative) and were used to capture rules of human expertise (often known as Expert Systems) but we found that too much knowledge had to be programmed into the system rather than being generated by the system itself. However if the system maintained a model of the component being assembled then it could compare the state of the component with the state of the ideal as seen in the model. This seemed a very promising approach and there were several people in the department interested in model-building so the group formed and took the initial name, the Model Based Systems group.

Models can be very complicated, often with lots of equations and mathematics. However human experts tend to think in more abstract terms and we explored the ideas of Qualitative Reasoning (QR) as a basis for our models. QR refers to techniques that use qualitative variables like 'high', 'medium' or 'low' instead of quantitative values. This level of abstraction provided ideal for representing relations that capture the essence of a system without the complexity of traditional numerical models.

Following much QR work on modelling physical objects, like gear assemblies, simple door-locks, and other basic mechanisms, we found that electrical circuits could be modelled with the same techniques we had developed and, rather surprisingly, were easier to model than mechanical devices. Through a series of grants the group built up a range of experience in developing and applying model based reasoning for electrical design analysis and diagnosis. This was during the period of the Alvey program and the group members were very active in the Deep Knowledge Based Systems Workshops that were held in the UK. The group became well known internationally and its most notable early successes were in diagnostic and failure analysis techniques for circuit design in the automotive industry.

We automated the standard industry testing process, known as Failure Mode Effects Analysis (FMEA) by our technique, giving a twenty-fold speedup. A complete scientific transfer progression can be seen in which fundamental blue-skies research was taken through all the stages, from theoretical ideas, to working prototypes, to commercialisation via a spin-off company and finally exploited as products now in use by several leading automotive manufacturers. During this period, Chris Price worked with Ford and Jaguar to develop our research into commercial software and established a company to sell the results to industry. The company, FirstEarth, grew to employ 12 people before being sold to Mentor Graphics Limited in 2003. Mentor Graphics (now owned by Siemens) was a major player in graphical tools, and AutoSteve became their second most popular product. The Ford Motor Company, who were supporters of our research, reported that our FMEA software had saved them $20 million in one year.

Other work on QR model-based reasoning includes investigations into generic diagnostic and monitoring problems in continuous process plant (Corus Group, formerly British Steel). In 1999, Chris Price, produced an authoritative practitioner monograph: “Computer based diagnostic systems”.

Group members became heavily involved in the international research community in QR, including programme chairs and committee members for the regular workshops and conferences, particularly the international workshops on Principles of Diagnosis, (DX), and on Qualitative Reasoning (QR), and the workshops Model-based systems and Qualitative Reasoning at ECAI conferences (the main European AI conference series).

This collaboration led to an application to fund MONET, the European Network of Excellence in Model-Based Systems and Qualitative Reasoning. Funded through the ESPRIT Long Term Research Programme, Networks of Excellence were groupings of industrial and academic research teams with common long term technological goals, and coordinated research, training and technology transfer policies. 1n 1997 John Hunt and Mark Lee were awarded a grant to set up and run the MONET network for three years. This employed John McCardle as manager and Suzy Shipman as administrator.

The objectives of the network were to establish a greater understanding among industry and public sector organisations of the contribution model-based reasoning can make towards the solution of complex problems. The network ran a regular newsletter, organised events, and maintained and supplied documents and resources. See a sample flyer and a report on a network-organised summer school. MONET grew to comprise more than 80 industrial and academic research laboratories from twelve countries in Europe with associated members in the USA and Japan. In 2001 George Coghill became a director and MONET 2 was funded, again for three years, and Iain Russell and Janet Thomas were the appointed administrators. For more on Monet click here.

The scope of our work in QR extended to cover other automotive problems (e.g. sneak circuit analysis, multiple failure analysis), functional modelling of components, design tools to detect errors at design time, and the modelling of electronic control units. George Coghill brought interests in fuzzy reasoning and had worked on combining fuzzy methods with QR in a fuzzy qualitative simulator. This strand of research was extended substantially in 2004 when Qiang Shen and a major part of his team joined Aberystwyth from Edinburgh. Qiang had pioneered work on integrating approximate inference with model-based reasoning.

The group name then changed to the Advanced Reasoning Group (ARG) to reflect the coverage of qualitative and approximative reasoning, methods of knowledge representation, and model-based problem solving. Currently, the ARG covers all major areas of Computational Intelligence (CI) and is one of the world-leading groups in CI, particularly renowned for its invention of semantics-preserving approximation techniques for explicit knowledge model formulation and simplification, and for its ground-breaking work on data-driven decision support with increased levels of automation, efficiency and reliability.

Grants and applications have built intelligent decision support systems, especially in crime detection and prevention, engineering design analysis, and computer-based diagnosis. Other work includes: model-based whole lifecycle automated system analysis; qualitative model-based learning; knowledge extraction over high dimensional data sets; compositional modelling and preference handling; and modelling sparse data and reasoning with sparse knowledge. We have led and contributed to organising a large number of international and national events in the subject area, including chairing many prestigious events (e.g., the 2007 IEEE International Conference on Fuzzy Systems, the EPSRC Workshop on the Future of Fuzzy Systems Research, and the 20th anniversary of the Annual Workshop on Computational Intelligence in 2021).

Recent EPSRC grants on approximate qualitative MB approaches to crime investigation and intelligence data analysis. Scientifically, this research for the first time, integrates modern qualitative model-based techniques and classical uncertaincy reasoning methodologies (e.g., Bayesian networks). Qiang Shen and Richard Jensen introduced fuzzy-rough techniques for knowledge model formulation and simplification. This resulted in the very first version of a data-driven fuzzy-rough feature selection tool on noisy and vague, continuous data, known as FRFS. It was systematically applied to the problem of complex industrial plant monitoring, which has since become a benchmark in applied feature selection. Jensen and Shen co-authored "Computational Intelligence and Feature Selection: Rough and Fuzzy Approaches".

Beyond feature selection, Richard Jensen and Neil MacPartthalain have also developed robust data mining approaches to tackle significant real-world problems that challenge traditional machine learning techniques (e.g., instance selection, missing data imputation, and object classification), as well as those in the more general areas like knowledge representation and decision support. Jensen and MacParthalain have also contributed many programmes implementing data-driven model learning and feature selection algorithms to Weka, the world’s largest open-access repository for machine learning and data-mining software.

Our research has had far-reaching applications, ranging from face and motion recognition, human expression analysis, bioinformatics and healthcare (including animal husbandry), material analysis and lithology modelling, webpage recommendation and digital documentation processing, environmental impact prediction, to plant monitoring and fault diagnosis. A much-celebrated example is the work with industry that led to the development of algorithms to detect usability issues for e-commerce sites, saving significant amounts of money (over $86 million). For intelligent systems that perform tasks such as classification, prediction and regression, the classical view has been that a system can only work when its underlying knowledge (be it provided by experts or learned from empirical data or a mixture of both) covers a given observation. Traditional probabilistic and fuzzy approaches relax such assumptions, but still require partial coverage. In reality, a system’s knowledge base is often incomplete and does not always match, even partially, the observation. Our group has made seminal contributions to tackling this ubiquitous problem in automated decision-making within uncertain environments. The most significant is fuzzy interpolative reasoning through knowledge or rules transformation (known as FRI). This work departs radically from previous approaches that frequently lead to mathematically ill-defined interpolated results (which do not have any meaningful interpretation, destroying the beauty of reasoning transparency that rule-based systems would otherwise enjoy). This exceptional advancement received the 2009 IEEE Outstanding Transaction Paper prize. Since then, fuzzy rule interpolation has rapidly left its footprint on R&D activities internationally, including a sequence of best paper awards or annual spotlight articles that our research staff and students received at the foremost international conferences.

As a globally leading CI group, ARG also acts as a major host for exchanges of international experts as well as for the training and development of future generation of specialists in the discipline. For instance, over the last REF period (2014-2021), a large number of world-authorities visited us from China, Europe, Singapore, and the US. We were awarded five Sêr Cymru II COFUND Research Fellows and one Marie Skłodowska-Curie Fellow. The highly competitive Sêr Cymru II COFUND programme was introduced by the Welsh Government to attract world class talent to Wales, and ARG has attracted the largest number of such outstanding fellows. All these colleagues now hold senior academic positions in China, France and the UK, including two becoming full-professors and one Head of Department. Recently two of our PhD graduates have been appointed to senior lecturer in the UK and two others appointed as Heads of Department in leading Chinese universities.

Other research topics include evolutionary methods for optimisation (Jun He, and others - names please) More here from Thomas, Christine, Neal, ….?