Only a few years ago, terms like genetic algorithms, evolution strategies, classifier systems, genetic programming, etc. did not mean anything to most scientists let alone practitioners in industry, business, and the general public. However, these research fields, which are united now under the name evolutionary computation, have existed for a long time. Three of its most important roots are generally recognized. One is the work of John Holland in the USA, the second the work of Ingo Rechenberg in Germany, and the third is the work of Lawrence Fogel, again in the USA. Fogel, Rechenberg, Holland, and some others started already 30 years ago to investigate systems that evolve artificially. Some remarkable theoretical and practical results were achieved already at that time, but they had to wait for their broad acknowledgement.

Today, the scene has changed considerably. Even some newspapers have reported on artificially evolving systems and their application to solve practical problems. There are two main reasons for this change: the apparent simplicity of evolutionary algorithms coupled with the availability of powerful computers, which allow to handle more realistic and thus more complex models now. The basic idea of such algorithms is easily described: For any optimization problem, generate a number of solutions. Repeatedly, let the worst die out and modify the others at random, until a sufficiently good solution has been found. Since anybody can understand such algorithms immediately, David Goldberg called them once CP-easy. However, large evolving populations of solutions combined with a long evolution time require high computing power. Nowadays, computers providing that power are available nearly everywhere, in every lab, at home, at school, etc. It is thus very easy to write a short program and have first success.

Yet, artificial evolution is far from being well understood. Until now, there is no comprehensive theory for any type of evolutionary algorithm - only the most simple ones have been analyzed thoroughly, and even that for very simple problem situations only. For most practical applications, the algorithms' parameters have to be crafted manually. In contrast to this lack of understanding, evolutionary algorithms and algorithms based on other natural paradigms have been practically applied with remarkable success. One reason for that is the very high robustness of such algorithms and the fact that they can be applied in many situations, also where classical algorithms fail.

Within the last years, we see an explosion of interest in evolutionary computation and the application of other natural metaphors like neural networks, simulated annealing, and so forth. The at first relatively small communities of researchers in the USA and Europe who made progress in these fields during the last decades have grown exponentially, and new ones have been formed, such as the one in Japan. After many very difficult years, governments and industry have recognized the potential behind such algorithms, and funding sources for research have become available. Nonetheless, in all aspects there are so many open questions that every progress made could increase the efficiency and effectivity of such algorithms dramatically and thus open up new fields of application.

ICEC/PPSN tries to advance the state-of-the-art by presenting high-quality original research papers in a number of different topics. Five papers were selected that discuss general aspects of Darwinian and Lamarckian evolution, the relations between genotype and phenotype, the control of population diversity, and co-evolution. Since theory can best improve the applicability of new algorithms, considerable space was allocated to it. Eleven papers were selected that present new theoretical results. They consider either a specific type of algorithm and try to cover as many applications as possible, or focus on a specific aspect of a general class of algorithms. Examples are the convergence rate of evolutionary algorithms and the optimal population size. Despite of the lack of a complete theoretical understanding, researchers have altered the basic algorithms, either by incorporating application specific knowledge or introducing new ideas, and reported improved performance. PPSN presents nine papers of this type. Classifier systems and genetic programming have received less attention than genetic algorithms and evolution strategies, although very promising results have been reported recently. Three papers concentrate on learning in these rule-based systems, one of them in combination with fuzzy logic. Four papers focus on various theoretical and practical aspects of genetic programming. Unfortunately, only a few of the submitted papers deal with other natural metaphors. Four have been selected that have been subsumed under the term ``emergent computation'', even if this is a little arbitrary. They range from aspects of artificial life to cellular automata. Two papers compare different algorithms or different parameter settings with each other. They are of particular interest to practitioners that want to get a feeling for the behaviour of such algorithms under various conditions. A further topic relates to the combination of different paradigms. Two of the three papers accepted deal with parallel simulated annealing guided by the application of genetic operators. Here and in the following section, at least a few other natural metaphors are presented. Six papers deal with various aspects of the combination of neural networks with evolutionary algorithms.

Whereas the former sections are more on the theoretical side, the following sections relate directly to practical applications. One section considers the parallel implementation of evolutionary algorithms. The four papers presented concentrate on the efficiency gain obtained by implementing genetic algorithms on parallel processors of various type or on a cluster of distributed processors. In one of them a hardware implementation of an evolutionary algorithm is described that increases the processing speed dramatically. The last large section is devoted to specific applications. Up to now, evolutionary computation has been applied to a great variety of optimization problems, and this variety is mirrored in the selected eight papers. The applications range from optimization of VLSI chips over time tables to the bi-partitioning problem. Finally, one paper presents a software tool that facilitates the application of evolutionary algorithms.

Here is a short list of the topics covered:

We are sure that these papers presented at ICEC/PPSN are of interest to many of the conference participants, the researchers, and the users of evolutionary computation.

Jerusalem,Yuval Davidor, Conference Chair
Mannheim,Reinhard Männer, Programme Co-Chair
Dortmund,Hans-Paul Schwefel, Programme Co-Chair
July 1994
Mon Jul 10 16:32:49 MET DST 1995