Emma Hart

Hyper-Heuristics: An Emerging Direction in Modern Search Technology

This chapter introduces and overviews an emerging methodology in search and optimisation. One of the key aims of these new approaches, which have been termed hyperheuristics, is to raise the level of generality at which optimisation systems can operate. An objective is that hyper-heuristics will lead to more general systems that are able to handle a wide range of problem domains rather than current meta-heuristic technology which tends to be customised to a particular problem or a narrow class of problems. Hyper-heuristics are broadly concerned with intelligently choosing the right heuristic or algorithm in a given situation. Of course, a hyper-heuristic can be (often is) a (meta-)heuristic and it can operate on (meta-)heuristics. In a certain sense, a hyper-heuristic works at a higher level when compared with the typical application of meta-heuristics to optimisation problems, i.e., a hyper-heuristic could be thought of as a (meta)-heuristic which operates on lower level (meta-)heuristics. In this chapter we will introduce the idea and give a brief history of this emerging area. In addition, we will review some of the latest work to be published in the field.

A Comparison of Dominance Mechanisms and Simple Mutation on Non-stationary Problems

It is sometimes claimed that genetic algorithms using diploid representations will be more suitable for problems in which the environment changes from time to time, as the additional information stored in the double chromosome will ensure diversity, which in turn allows the system to respond more quickly and robustly to a change in the fitness function. We have tested various diploid algorithms, with and without mechanisms for dominance change, on non-stationary problems, and conclude that some form of dominance change is essential, as a diploid encoding is not enough in itself to allow flexible response to change. Moreover, a haploid method which randomly mutates chromosomes whose fitness has fallen sharply also performs well on these problems.

An Overview of Artificial Immune Systems

The immune system is highly distributed, highly adaptive, self-organising in nature, maintains a memory of past encounters and has the ability to continually learn about new encounters. From a computational point of view, the immune system has much to offer by way of inspiration to computer scientists and engineers alike. As computational problems become more complex, increasingly, people are seeking out novel approaches to these problems, often turning to nature for inspiration. A great deal of attention is now being paid to the vertebrae immune system as a potential source of inspiration, where it is thought that different insights and alternative solutions can be gleaned, over and above other biologically inspired methods. Given this rise in attention to the immune system, it seems appropriate to explore this area in some detail. This survey explores the salient features of the immune system that are inspiring computer scientists and engineers to build Artificial Immune Systems (AIS). An extensive survey of applications is presented, ranging from network security to optimisation and machine learning. However, this is not complete, as no survey ever is, but it is hoped this will go some way to illustrate the potential of this exciting and novel area of research.

Producing robust schedules via an artificial immune system

This paper describes an artificial immune system (AIS) approach to producing robust schedules for a dynamic job-shop scheduling problem in which jobs arrive continually, and the environment is subject to change due to practical reasons. We investigate whether an AIS can be evolved using a genetic algorithm (GA), and then used to produce sets of schedules which together cover a range of contingencies, both foreseeable and unforeseeable. We compare the quality of the schedules to those produced using a genetic algorithm specifically designed for tackling job-shop scheduling problems, and find that the schedules produced from the evolved AIS compare favourably to those produced by the GA. Furthermore, we find that the AIS schedules are robust in that there are large similarities between each schedule in the set, indicating that a switch from one schedule to another could be performed with minimal disruption if rescheduling is required.

Learning a procedure that can solve hard bin-packing problems: a new GA-based approach to hyper-heuristics

The idea underlying hyper-heuristics is to discover some combination of familiar, straightforward heuristics that performs very well across a whole range of problems. To be worthwhile, such a combination should outperform all of the constituent heuristics. In this paper we describe a novel messy-GA-based approach that learns such a heuristic combination for solving one-dimensional bin-packing problems. When applied to a large set of benchmark problems, the learned procedure finds an optimal solution for nearly 80% of them, and for the rest produces an answer very close to optimal. When compared with its own constituent heuristics, it ranks first in 98% of the problems.

Evolutionary Scheduling: A Review

Early and seminal work which applied evolutionary computing methods to scheduling problems from 1985 onwards laid a strong and exciting foundation for the work which has been reported over the past decade or so. A survey of the current state-of-the-art was produced in 1999 for the European Network of Excellence on Evolutionary Computing EVONET—this paper provides a more up-to-date overview of the area, reporting on current trends, achievements, and suggesting the way forward.

Some Observations about GA-Based Exam Timetabling

Although many people have tried using genetic algorithms (GAs) for exam timetabling, far fewer have done systematic investigations to try to determine whether a GA is a good choice of method or not. We have extensively studied GAs that use one particular kind of direct encoding for exam timetabling. Perhaps not surprisingly, it emerges that this approach is not very good, but it is instructive to see why. In the course of this investigation we discovered a class of solvable problems with interesting properties: our GAs would sometimes fail to solve some of the moderately-constrained problems, but could solve all of the lightly-constrained ones and all of the highly-constrained ones. This is despite the fact that they form a hierarchy: those erratically-solved problems are subproblems of the easily-solved but highly-constrained ones. Moreover, some other non-evolutionary approaches also failed on precisely the same sets. This, together with some observations about much simpler graph-colouring methods based on the Brelaz algorithm, suggest some future directions for GA-based methods.

Application areas of AIS: the past, the present and the future

After a decade of research into the area of Artificial Immune Systems, it is worthwhile to take a step back and reflect on the contributions that the paradigm has brought to the application areas to which it has been applied. Undeniably, there have been a lot of successful stories — however, if the field is to advance in the future and really carve out its own distinctive niche, then it is necessary to be able to illustrate that there are clear benefits to be obtained by applying this paradigm rather than others. This paper attempts to take stock of the application areas that have been tackled in the past, and ask the difficult question “was it worth it ?”. We then attempt to suggest a set of problem features that we believe will allow the true potential of the immunological system to be exploited in computational systems, and define a unique niche for AIS.

GAVEL - a new tool for genetic algorithm visualization

This paper surveys the state of the art in evolutionary algorithm visualization and describes a new tool called GAVEL. It provides a means to examine in a genetic algorithm (GA) how crossover and mutation operations assembled the final result, where each of the alleles came from, and a way to trace the history of user-selected sets of alleles. A visualization tool of this kind can be very useful in choosing operators and parameters and in analyzing how and, indeed, whether or not a GA works. We describe the new tool and illustrate some of the benefits that can be gained from using it with reference to three different problems: a timetabling problem, a job-shop scheduling problem, and Goldberg and Horns long-path problem. We also compare the tool to other available visualization tools, pointing out those features which are novel and identifying complementary features in other tools.

Solving a real-world problem using an evolving heuristically driven schedule builder

This work addresses the real-life scheduling problem of a Scottish company that must produce daily schedules for the catching and transportation of large numbers of live chickens. The problem is complex and highly constrained. We show that it can be successfully solved by division into two subproblems and solving each using a separate genetic algorithm (GA). We address the problem of whether this produces locally optimal solutions and how to overcome this. We extend the traditional approach of evolving a permutation + schedule builder by concentrating on evolving the schedule builder itself. This results in a unique schedule builder being built for each daily scheduling problem, each individually tailored to deal with the particular features of that problem. This results in a robust, fast, and flexible system that can cope with most of the circumstances imaginable at the factory. We also compare the performance of a GA approach to several other evolutionary methods and show that population-based methods are superior to both hill-climbing and simulated annealing in the quality of solutions produced. Population-based methods also have the distinct advantage of producing multiple, equally fit solutions, which is of particular importance when considering the practical aspects of the problem.