Martin J. Oates

The Pareto Envelope-Based Selection Algorithm for Multi-objective Optimisation

We introduce a new multiobjective evolutionary algorithm called PESA (the Pareto Envelope-based Selection Algorithm), in which selection and diversity maintenance are controlled via a simple hyper-grid based scheme. PESAs selection method is relatively unusual in comparison with current well known multiobjective evolutionary algorithms, which tend to use counts based on the degree to which solutions dominate others in the population. The diversity maintenance method is similar to that used by certain other methods. The main attraction of PESA is the integration of selection and diversity maintenance, whereby essentially the same technique is used for both tasks. The resulting algorithm is simple to describe, with full pseudocode provided here and real code available from the authors. We compare PESA with two recent strong-performing MOEAs on some multiobjective test problems recently proposed by Deb. We find that PESA emerges as the best method overall on these problems.

A Preliminary Investigation of Modified XCS as a Generic Data Mining Tool

Wilsons XCS classifier system has recently been modified and extended in ways which enable it to be applied to real-world benchmark data mining problems. Excellent results have been reported already on one such problem by Wilson, while other work by Saxon and Barry on a tunable collection of machine learning problems has also pointed to the strong potential of XCS in this area. In this paper we test a modified XCS implementation on twelve benchmark machine learning problems, all real-world derived. XCS is compared on these benchmarks with C4.5 and with HIDER (a new and sophisticated GA for machine learning developed elsewhere). Results for both C4.5, HIDER and XCS on each problem were tenfold cross-validated, and in the case of HIDER and XCS a modest amount of preliminary parameter investigation was done to find good results in each case. We find that XCS outperforms the other techniques in eight of the twelve problems, and is second-best in two of the remaining three. Some investigation is then done of the variance in XCS performance, and we find this to be verging on significant, either when varying the data fold composition, or the algorithmic random seed. We also investigate variation of several XCS parameters around well-known default settings. We find the default settings to be generally robust, but find the mutation rates and GA selection scheme to be particularly worthy of exploration with a view to improved performance. We conclude that XCS has the potential to be a powerful general data mining tool, at least for databases without too many fields, but that considerable research is warranted to identify rules and guidelines for parameter and strategy setting.

A Ruleset Reduction Algorithm for the XCS Learning Classifier System

XCS is a learning classifier system based on the original work by Stewart Wilson in 1995. It is has recently been found competitive with other state of the art machine learning techniques on benchmark data mining problems. For more general utility in this vein, however, issues are associated with the large numbers of classifiers produced by XCS; these issues concern both readability of the combined set of rules produced, and the overall processing time. The aim of this work is twofold, to produce reduced classifier sets which can more readily be understandable as rules, and to speedup processing via reduction of classifier set size during operation of XCS. A number of algorithmic modifications are presented, both in the operation of XCS itself and in the post-processing of the final set of classifiers. We describe a technique of qualifying classifiers for inclusion in action sets, which enables classifier sets to be generated prior to passing to a reduction algorithm, allowing reliable reductions to be performed with no performance penalty. The concepts of ‘spoilers’ and ‘uncertainty’ are introduced, which help to characterise some of the peculiarities of XCS in terms of operation and performance. A new reduction algorithm is described which we show to be similarly effective to Wilson’s recent technique, but with considerably more favourable time complexity, and we therefore suggest that it may be preferable to Wilson’s algorithm in many cases with particular requirements concerning the speed/performance tradeoff.

Real-World Applications of Evolutionary Computing

This book constitutes the refereed proceedings of six workshops on evolutionary computation held concurrently as EvoWorkshops 2000 in Edinburgh, Scotland, UK, in April 2000. The 37 revised papers presented were carefully reviewed and selected by the respective program committees. All in all, the book demonstrates the broad application potential of evolutionary computing in a variety of fields. In accordance with the individual workshops, the book is divided into sections on image and signal processing; systems, controls, and drives in industry; telecommunications; scheduling and timetabling; robotics; and aeronautics

On the Assessment of Multiobjective Approaches to the Adaptive Distributed Database Management Problem

In this paper we assess the performance of three modern multiobjective evolutionary algorithms on a real-world optimization problem related to the management of distributed databases. The algorithms assessed axe the Strength Pareto Evolutionary Algorithm (SPEA), the Pareto Archived Evolution Strategy (PAES), and M-PAES, which is a Memetic Algorithm based variant of PAES. The performance of these algorithms is compared using two distinct and sophisticated multiobjective-performance comparison techniques, and extensions to these comparison techniques are proposed. The information provided by the different performance assessment techniques is compared, and we find that, to some extent, the ranking of algorithm performance alters according to the comparison metric; however, it is possible to understand these differences in terms of the complex nature of multiobjective comparisons.

On Fitness Distributions and Expected Fitness Gain of Mutation Rates in Parallel Evolutionary Algorithms

Setting the mutation rate for an evolutionary algorithm (EA) is confounded by many issues. Here we investigate mutation rates mainly in the context of large-population-parallelism. We justify the notion that high rates achieve better results, using underlying theory which notices that parallelization favourably alters the fitness distribution of a mutation operator. We derive an expression which sets out how this is changed in terms of the level of parallelization, and derive further expressions that allow us to adapt the mutation rate in a principled way by exploiting online-sampled landscape information. The adaptation technique (called RAGE - Rate Adaptation with Gain Expectation) shows promising preliminary results. Our motivation is the field of Directed Evolution (DE), which uses large-scale parallel EAs for limited numbers of generations to evolve novel proteins. RAGE is highly suitable for DE, and is applicable to large-scale parallel EAs in general.

Investigating Evolutionary Approaches to Adaptive Database Management Against Various Quality of Service Metrics

The management of large distributed databases is becoming more complex as user demand grows. Further, global access causes points of geographic contention to ‘follow the sun’ during the day giving rise to a dynamic optimisation problem where the goal is to constantly maximise the quality of service seen by the database users. A key quality criterion is to optimise the quality of service perceived by the worst-served user by finding a choice of client-server mapping which best balances issues such as exploitation of fast servers and communications links, and the degradation in response-time due to over-use of such servers/links. Any approach to solving the problem must be fast (so that results remain applicable) and successful over a variety of different database usage scenarios and quality of service metrics. This paper investigates the effectiveness of several local and evolutionary search approaches to this problem, focusing on the variations in performance across a range of QoS metrics.

Encouraging Compact Rulesets from XCS for Enhanced Data Mining

Learning Classifier Systems (LCSs) are increasingly being found to be effective machine learning systems that can address a variety of real world problems (as testified by several chapters in this book). Based on seminal ideas due to Holland (1976, 1980), they gradually evolve rulesets, which either model a static dataset, or model actions (and chains of actions) in an environment. Comprehensive tutorial and survey material on this rapidly growing field is now provided in many places, but we particularly point to Holland et al (2000), Holland (2000) and Lanzi and Riolo (2000), as well as the introductory material in this volume.