Feature subset selection is an important problem in knowl- edge discovery, not only for the insight gained from deter- mining relevant modeling variables but also for the improved understandability, scalability, and possibly, accuracy of the resulting models. In this paper we consider the problem of feature selection for unsupervised learning. A number of heuristic criteria can be used to estimate the quality of clusters built from a given featuresubset. Rather than combining such criteria, we use ELSA, an evolutionary lo- cal selection algorithm that maintains a diverse population of solutions that approximate the Pareto front in a multi- dimensional objectiv espace. Each evolved solution repre- sents a feature subset and a number of clusters; a standard K-means algorithm is applied to form the given n umber of clusters based on the selected features. Preliminary results on both real and synthetic data show promise in finding Pareto-optimal solutions through which we can identify the significant features and the correct number of clusters.

While search engines have become the major decision support tools for the Internet, there is a growing disparity between the image of the World Wide Web stored in search engine repositories and the actual dynamic, distributed nature of Web data. We propose to attack this problem using an adaptive population of intelligent agents mining the Web online at query time. We discuss the benefits and shortcomings of using dynamic search strategies versus the traditional static methods in which search and retrieval are disjoint. This paper presents a public Web intelligence tool called MySpiders, a threaded multiagent system designed for information discovery. The performance of the system is evaluated by comparing its effectiveness in locating recent, relevant documents with that of search engines. We present results suggesting that augmenting search engines with adaptive populations of intelligent search agents can lead to a significant competitive advantage. We also discuss some of the challenges of evaluating such a system on current Web data, introduce three novel metrics for this purpose, and outline some of the lessons learned in the process.

Evolutionary algorithms have been appied to the synthesis of neural architectures...

Feature subset selection is important not only for the insight gained from determining relevant modeling variables but also for the improved understandability, scalability, and possibly, accuracy of the resulting models. Feature selection has traditionally been studied in supervised learning situations, with some estimate of accuracy used to evaluate candidate subsets. However, we often cannot apply supervised learning for lack of a training signal. For these cases, we propose a new feature selection approach based on clustering. A number of heuristic criteria can be used to estimate the quality of clusters built from a given feature subset. Rather than combining such criteria, we use ELSA, an evolutionary local selection algorithm that maintains a diverse population of solutions that approximate the Pareto front in a multi-dimensional objective space. Each evolved solution represents a feature subset and a number of clusters; two representative clustering algorithms, K-means and EM, are applied to form the given number of clusters based on the selected features. Experimental results on both real and synthetic data show that the method can consistently find approximate Pareto-optimal solutions through which we can identify the significant features and an appropriate number of clusters. This results in models with better and clearer semantic relevance. 1.

. Evolutionary algorithms are becoming increasingly valuable in solving large-scale, realistic engineering multiobjective optimization (MO) problems, which typically require consideration of conflicting and competing design issues. The new procedure, Constraint Method-Based Evolutionary Algorithm (CMEA), presented in this paper is based upon underlying concepts in the constraint method described in the mathematical programming literature. Pareto optimality is achieved implicitly via a constraint approach, and convergence is enhanced by using beneficial seeding of the initial population. CMEA is evaluated by solving two test problems reported in the multiobjective evolutionary algorithm (MOEA) literature. Performance comparisons based on quantitative metrics for accuracy, coverage, and spread are presented. CMEA is relatively simple to implement and incorporate into existing implementations of evolutionary algorithm-based optimization procedures. 1

Abstract — Swarm Intelligence (SI) is a relatively new paradigm being applied in a host of research settings to improve the management and control of large numbers of interacting entities such as communication, computer and sensor networks, satellite constellations and more. Attempts to take advantage of this paradigm and mimic the behavior of insect swarms however often lead to many different implementations of SI. The rather vague notions of what constitutes self-organized behavior lead to rather ad hoc approaches that make it difficult to ascertain just what SI is, assess its true potential and more fully take advantage of it. This article provides a set of general principles for SI research and development. A precise definition of self-organized behavior is described and provides the basis for a more axiomatic and logical approach to research and development as opposed to the more prevalent ad hoc approach in using SI concepts. The concept of Pareto optimality is utilized to capture the notions of efficiency and adaptability. A new concept, Scale Invariant Pareto Optimality is described and entails symmetry relationships and scale invariance where Pareto optimality is preserved under changes in system states. This provides a mathematical way to describe efficient tradeoffs of efficiency between different scales and further, mathematically captures the notion of the graceful degradation of performance so often sought in complex systems. Index Terms — swarm intelligence, self-organization, multiobjective optimization, Pareto optima, finite-state machines

In this paper we are concerned with finding the Pareto optimal front or a good approximation to it. Since non-dominated solutions represent the goal in multiobjective optimisation, the dominance relation is frequently used to establish preference between solutions during the search. Recently, relaxed forms of the dominance relation have been proposed in the literature for improving the performance of multiobjective search methods. This paper investigates the influence of different fitness evaluation methods on the performance of two multiobjective methodologies when applied to a highly constrained two-objective optimisation problem. The two algorithms are: the Pareto archive evolutionary strategy and a population-based annealing algorithm. We demonstrate here, on a highly constrained problem, that the method used to evaluate the fitness of candidate solutions during the search affects the performance of both algorithms and it appears that the dominance relation is not always the best method to use.

The World Wide Web creates opportunities for search systems using adaptive distributed agents. This paper presents a threaded implementation of InfoSpiders, a client-based system that uses an evolving population of intelligent agents to browse the Web at query time. We consider different fitness functions based on network resource consumption and show that taxing agents in proportion to latency results in better efficiency without penalties in the quality of the retrieved documents. The tool

Some recent successful techniques to solve multiobjective optimisation problems are based on variants of evolutionary algorithms and use recombination and self-adaptation to evolve the population. We present an approach that incorporates a population of solutions into a hybrid metaheuristic with no recombination. The population is evolved using self-adaptation, a mutation operator and an information-sharing mechanism. Since the main component in our approach is a simulated annealing algorithm, the cooling schedule for the whole population becomes critical. A common cooling schedule for the whole population is determined based on an evolutionary process. Results are presented using a real-world multiobjective combinatorial optimisation problem, namely space allocation with two conflicting criteria. These results suggest that this approach is a suitable alternative not only for combinatorial multiobjective optimisation problems, but also for obtaining a population of locally optima solutions in singleobjective optimisation problems. 1

Traditional approaches to designing multi-agent systems are offline, in simula-tion, and assume the presence of a global observer. Artificial Physics (AP) or physicomimetics (Spears and Gordon 1999) can be used to self-organize swarms of mobile robots into formations that move towards a goal. Using an offline ap-proach, we extend the AP framework to moving formations through obstacle fields. We provide important metrics of performance that allow us to (a) compare the utility of different generalized force laws in the artificial physics framework, (b) examine trade-offs between different metrics, and (c) provide a detailed method of comparison for future researchers in this area. In the online, real world, a global observer may be absent, performance feedback may be delayed or perturbed by noise, agents may only interact with their local neighbors, and only a subset of agents may experience any form of performance feed-back. Under these constraints, designing multi-agent systems is difficult. We present a novel approach called“Distributed Agent Evolution with Dynamic Adaptation to Local Unexpected Scenarios ” or DAEDALUS to address these issues, by mimicking