Towards a rational methodology for using evolutionary search algorithms

• Main Author: Sharpe, Oliver John
• Published: University of Sussex 2002
• Subjects: 629.892; Fitness landscape
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.250147

Evolutionary search algorithms (ESAs from now on) are iterative problem solvers developed with inspiration from neo-Darwinian survival of the fittest genes. This thesis looks at the core issues surrounding ESAs and is a step towards building a rational methodology for their effective use. Currently there is no such method of best practice rather the whole process of designing and using ESAs is seen as more of a black art than a tried and tested engineering tool. Consequently, many non-practitioners are sceptical of the worth of ESAs as a useful tool at all. Therefore the first task of the thesis is to layout the reasons, from computational theory, why ESAs can be a potentially powerful tool. In this context the theory of NP-completeness is introduced to ground the discussions throughout the thesis. Then a simple framework for describing ESAs is developed to form another cornerstone of these later discussions. From here there are two main themes of the thesis. The first theme is that the No Free Lunch result requires us to take a problem centric, as opposed to algorithm centric, perspective on ESA research. The second major theme is the argument that whole algorithms and traditional computer science problem classes are the wrong level of granularity for the focus of our research. Instead we should be researching empirical questions of search bias at the granularity of the components of search algorithms. Furthermore, we should be finding empirical evidence to demonstrate that our granularity of analytic class is such that one analytic class maps onto one search bias class. We will see that this can mean that we have to sub-divide our classic computer science problems classes into smaller sub-classes. The hope is that we can find analytic distinctions that will sub-divide the instances along lines that match the divisions between the various empirically discoverable search bias classes. The intention is to develop our knowledge until we get one analytic class to map into one empirical class. If we have strong empirical evidence to suggest that this has been achieved then we have good grounds on which to confidently use this knowledge to predict the effective search biases required for new problem instances. In the last two chapters of the thesis we demonstrated these ideas on various instances of the euclidean TSP problem class