Natural and Evolutionary Computation Group
Members:
Academic Staff in Computing and Electronic Systems:
Poli, Tsang, Lucas, Zhang, Gan, Hagras, Scott, Clark
Affiliate Members:
Academic Staff in Accounting, Finance and Management:
Abdel-Kader, Haven
Academic Staff in Mathematical Sciences:
Langdon,
Salhi
Academic Staff in Economics:
Markose
Group's Research Interests:
Natural Computation is the study of computational systems that
use ideas and get inspirations from natural systems, including
biological, ecological and physical systems. Examples of nature
inspired algorithms include evolutionary algorithms (see below),
neural networks, ant systems (click on the picture on the right for
an illustration of an ant system solving a travelling salesman
problem), etc. Our research covers a variety of areas of natural
computation.
Genetic and Evolutionary Computation (GEC) is a field the main
objective of which draws inspiration from genetics and natural
selection to solve engineering problems such as optimisation, search
and machine learning. Genetic Algorithms (GAs) and Genetic
Programming (GP) are the most famous G EC techniques. GAs use
vectors of numbers (chromosomes) to represent the solutions of
problems (adult individual). GAs maintain a population of such
solutions and evolve it by applying an artificial selective pressure
on the solutions (survival of the fittest) and by making them mate
(crossover). GP is an evolutionary technique for the automatic
discovery of programs. GP is a special GA in which solutions are
syntax trees representing computer programs (click on the picture on
the right for an illustration). The syntax trees are executed by an
interpreter/compiler and their performance (fitness) is evaluated.
The fittest programs survive and mate. Our work has focused both on
the applications of evolutionary technology and on the theoretical
foundations of genetic programming and genetic algorithms.
Estimation of distribution algorithms. Population-based algorithms using estimation of distribution, often called estimation of distribution algorithms (EDAs), are another major paradigm in evolutionary computation. Some EDA-like algorithms have achieved state-of-the-art performance in applications. However, most of the existing EDA-like algorithms have been developed on an ad-hoc basis. Both theory and implementation in this area are far from complete. This project will work towards establishing a sound theory for characterizing and explaining EDA-like algorithms. We will extend our previous results and seek to obtain global convergence conditions for the factorised distribution algorithm and ant-colony optimisation algorithms. With a better understanding of the working mechanisms of EDA-like algorithms, we will embed experimental design methods in EDA to develop more powerful and statistically sound search and optimisation algorithms. We will also exploit the potential benefits of combining EDA-like algorithms with GLS (Guided Local Search). We propose to use continuous optimisation problems and quadratic assignment problems as our test problems. This project is funded by the EPSRC and University of Essex.
Example Applications:
Finance, Prediction, Classification and Modelling. Click on the
picture on the right to see how a genetic programming system can
discover an expression that bests fit a set of datapoints (symbolic
regression).
Automatic Engineering Design. Click on the picture on the right
to discover how an evolutionary algorithm can routinely discover
parallel digital circuits implementing a set of user-specified
requirements
Image and Signal Processing. Many image analysis
tasks can be
seen as hard optimisation problems. Evolutionary algorithms are
natural candidates to solve this kind of problems. The objective of
this research stream is the development of new algorithms for the
solution of some of the main problems in image understanding such as
image enhancement and segmentation, content-based image retrieval,
2-D and 3-D shape representation. In previous research we have
successfully applied evolutionary techniques, in particular genetic
algorithms and genetic programming, to the analysis of medical
images and signals and to image enhancement problems. Click on the
pictures on the right to discover how genetic programming can
routinely discover image processing algorithms for a variety of
complex tasks.
Control, Robotics. The Associative Experience Engine (AEE) (British
Patent 99-10539.7) is a fuzzy genetic system (click on the diagram
on the right to see its
structure) that has been applied to
difficult and challenging problems such as controlling a mobile
robot in an outdoor environment, controlling intelligent building
and even controlling large diesel engines (see image on the left).
For more information contact:
Professor R Poli
Department of Computing and Electronic Systems, University of Essex, Colchester CO4
3SQ UK
Tel: +44 1206 872338
Fax: +44 1206 872788
Email: rpoli @ essex . ac . uk