David John Nettleton

Evolutionary algorithms and a fractal inverse problem.

Over the past 30 years, algorithms that model natural evolution have generated robust search methods. These so-called evolutionary algorithms have been successfully applied to a wide range of problems. This paper discusses two types of evolutionary algorithms and their application to a problem in shape representation. Genetic algorithms and evolutionary programming, although both based on evolutionary principles, each place different emphasis on what drives the evolutionary process. While genetic algorithms rely on mimicking specific genotypic transformations, evolutionary programming emphasizes phenotypic adaptation. Results presented show the success of evolutionary programming in solving an example of a fractal inverse problem, but indicate that a genetic algorithm is not as successful. Reasons for this disparity are discussed.

The Durham Telephone Enquiry System

The Durham telephone enquiry system is a speaker/gender independent telephone enquiry service which operates in real-time. The system has been successfully applied to English and Italian telephone databases of up to 100,000 entries.The mechanism by which the telephone database is searched is a key feature of the enquiry system and is based upon a management-of-uncertainty methodology. The practical result of which is that there is no expectation that the users utterances will always be correctly interpreted by the speech recognition stage. This paper contains a description of the components of the system and how they are integrated. Issues which had to be addressed in ensuring that the system operates in real-time on ‘live’ input are also discussed.

Large Scale WSD Using Learning Applied to SENSEVAL

A word sense disambiguation system which is going to be used aspart of a NLP system needs to be large scale, able to beoptimised towards a specific task and above all accurate. This paperdescribes the knowledge sources used in a disambiguation system able toachieve all three of these criteria. It is a hybrid system combining sub-symbolic, stochastic and rule-based learning. The paper reportsthe results achieved in Senseval and analyses them to show the systemsstrengths and weaknesses relative to other similar systems.

Reductions in the search space for deriving a fractal set of an arbitrary shape

At present, the problem of finding a quick and efficient way of representing an arbitrary shape as a set of contraction mappings (an iterated function system) is unresolved. Such a representation is particularly useful in shape representation since the primitives used to construct the shape will automatically have the correct morphology. Several attempts have been made to solve this problem and some of these are discussed. The main difficulty with these approaches is the large size and great complexity of the search space. This paper examines several constraints, all of a low computational complexity, which can be placed on each of the mappings which make up a possible solution. These constraints reduce the search space of four of the six coefficients of a mapping by between 20% and 85%, and of the other two by between 75% and 95% (the size of the reduction depends only on the size of the bounding box of the shape). Since these constraints apply to each mapping of an IFS, their cumulative effect on the search space is substantial. It is anticipated that these reductions in the search space can be used to aid a variety of search algorithms.

Large ratios of mutation to crossover: the example of the traveling salesman problem

Genetic algorithms have recently been successfully applied to a wide range of problems. These often have search spaces that are very large, very complex, or both and are unsuitable for standard search algorithms such as hill climbing. The operators used in producing successive generations are usually those of crossover and mutation. The crossover operator is normally used in producing the majority of a generation while mutation acts as a background process. This paper examines the use of high amounts of mutation and gives the example of a genetic algorithm applied to the travelling salesman problem. This shows that high amounts of mutation need not ruin the algorithms convergence to optimal solutions.

Subsymbolic Processing using Adaptive Algorithms

Subsymbolic approaches have been adopted in attempting to solve many AI problems. In order to find a near optimal solution to the problem a procedure is needed by which the subsymbolic components can be manipulated. In searching all but the simplest of solution spaces algorithms such as hill climbing will often result in only suboptimal solutions being found. Often search algorithms do not make sufficient use of information acquired from previous evaluations of possible solutions. Several forms of adaptive algorithm have been developed in an attempt to overcome this problem and produce robust search mechanisms, e.g., evolutionary algorithms, classifier systems. This paper discusses some adaptive algorithms and presents initial work on a novel form of adaptive algorithm.

Search-space reductions in deriving a fractal set for an arbitrary shape

At present the problem of finding a quick and efficient way of representing an arbitrary shape as a set of contraction mappings (an iterated function system) is unresolved. The main problem that arises is the sheer size and complexity of the search space. This paper examines several constraints that can be placed on solutions, each of which has a low computational complexity. These constraints considerably reduce the search space in which the solutions exist and can be used to aid a variety of search algorithms.

Qualitative mathematical modelling of genetic algorithms

Genetic algorithms are adaptive search algorithms which generate and test a population of individuals where each individual corresponds to a solution. They have been successfully applied to a range of problems in both artificial intelligence research and industry. The selection of the optimal parameters for a genetic algorithm is often a problem. This is especially true if the genetic algorithm has a protracted run-time in which case the setting of the parameters by trial and error is often unrealistic. This paper proposes the use of probability distribution functions and random walks to model various operators used in genetic algorithms. In this way it is hoped that a qualitatively accurate model with a very short run-time can be produced.