Zbigniew Skolicki

The influence of migration sizes and intervals on island models

A need for solving more and more complex problems drives the Evolutionary Computation community towards advanced models of Evolutionary Algorithms. One such model is the island model which, although the subject of a variety of studies, still needs additional fundamental research. In this paper we have experimentally studied the influence of various migrations sizes and intervals on island models using a set of special functions. One of the surprising observations from these experiments is that the migration interval seems to be a dominating factor, with migration size generally playing a minor role with regard to the best solution found. Additional experiments measuring genetic diversity show that too frequent migrations cause islands to dominate others and lose global diversity before they are able to exchange solutions to produce better results. Also, we observe that even small migrations already make a significant impact on the behavior of an island model and therefore the effects are comparable to those of bigger migrations. On the other hand rare migrations cause a degraded performance due to the slow convergence. Collectively, these observations provide useful guidance for island model applications.

An analysis of island models in evolutionary computation

A need for solving more and more complex problems drives the Evolutionary Computation community towards advanced models of Evolutionary Algorithms. One such model is the island model which, although the subject of a variety of studies, still needs additional fundamental research. In my Ph.D. thesis I am aiming at studying the behavior of island models with regard to the amount of cooperation between islands, the level of heterogeneity and the difficulty of the problem being solved. This paper presents the main ideas and gathers preliminary results.

Improving Evolutionary Algorithms with Multi-representation Island Models

We present an island model that uses different representations in each island. The model transforms individuals from one representation to another during migrations. We show that such a model helps the evolutionary algorithm to escape from local optima and to solve problems that are difficult for single representation EAs. We illustrate this approach with a two population island model in which one island uses a standard binary encoding and the other island uses a standard reflective Gray code. We compare the performance of this multi-representation island model with single population EAs using only binary or Gray codes. We show that, on a variety of difficult multi-modal test functions, the multi-representation island model does no worse than a standard EA on all of the functions, and produces significant improvements on a subset of them.

Intelligent Agents in Design

In this paper we argue that learning or adaptation ability should be included in the basic set of features characterizing an intelligent agent in design. We propose a collection of attributes describing agents, which are grouped into several categories. Next, we present the results of a detailed study of all agents for design, which were discussed during the First International Workshop on Agents in Design at MIT in August of 2002. A statistical analysis of their attributes has been conducted and its results are reported to suggest future evolution of agents in design. Finally, we briefly overview the topic of Directed Evolution and use its paradigms to predict further development of agents in design. The paper also provides our initial conclusions and suggests further research.Copyright

Co-evolution of terrorist and security scenarios for water distribution systems

Identification of vulnerabilities of water distribution systems and identification of appropriate counter-measures are important components of homeland security. These are difficult and time consuming tasks. This paper provides a new approach to resolve these problems in complex infrastructure systems. It is based on the use of co-evolutionary computation for the generation of both terrorist and security scenarios. The basic concepts of co-evolutionary computation are briefly explained. The concept of co-evolutionary generation of terrorist and security scenarios is introduced in the context of a hypothetical water distribution system for a small town. A tool developed at George Mason University is used for a number of experiments that reveal a variety of emerging security patterns. The experiments show that these patterns may be helpful in effectively protecting the network.

The importance of a two-level perspective for island model design

Our theoretical understanding of island models (IMs) is much worse than of single-population evolutionary algorithms (EAs). As a consequence there is relatively little guidance available to a practitioner for even the most basic aspects of IM design such as choosing the size and number of the islands. In this paper we improve on this situation by showing how a particular two-level perspective can in fact provide guidance for IM design.

Intelligent agent for designing steel skeleton structures of tall buildings

The paper discusses a study on the application of intelligent agents (IAs) to conceptual designing. It provides an overview of the state-of-the-art in the areas of ontologies and IAs. Next, the system Disciple, a learning intelligent agent shell, and system Inventor 2001, evolutionary design support tool, both developed at George Mason University, are briefly presented. Further, the paper introduces the developed ontology for a class of steel skeleton structures of tall buildings. The ontology was used to build an IA for the selection of initial parent design concepts in evolutionary designing. A description of the developed agent is provided as well. Finally, examples of design concepts proposed by the agent are presented. The paper also contains conclusions and recommendations for further research.

Linkage in Island Models

Island models (IMs) have multiple sub-populations which periodically exchange a fraction of individuals. This complex setup results in distinct dynamics of evolution and therefore, IMs are characterized by several interesting properties of IMs with regard to gene linkage, which are presented in this chapter. Traditional single-population evolutionary algorithms (EAs) suffer from a relatively quick and often random gene fixation, due to evolutionary selection, drift and physical linkage of recombination operators. In IMs, additional, slower inter-island level of evolution together with a recurring local evolution inside islands make searching for optimal allele configurations more systematic. In this chapter, it is shown that IM dynamics may counteract hitch-hiking. Further, multiple building blocks may be better identified and combined in IMs, consequently supporting compositional evolution, which is studied on the H-IFF function. Finally, a discussion follows on how the repeated fixation of genes in islands might be treated as a linkage learning process.

Water Supply Threat Reduction Using Evolutionary Approaches

The potential threats to US water supply systems have changed fundamentally in the past four years. Prior to 2001, the major threats were natural causes, accidents, and some malicious behavior by a small group of individuals. Against these threats, water supply agencies have done a truly remarkable job of ensuring a safe, dependable supply. However, threats posed by an organized group of actors may represent a new and different challenge to the security of water supplies. A tool to assist in identifying possible attacks and simultaneously providing remedies to counter these attacks has been developed. The tool uses evolutionary computation as the optimization method, and EPANET as the system simulator. Preliminary examples of its use to identify optimal attacks against a realistic but hypothetical pipe network and corresponding counter measures are presented.