Advances in Soft Computing

Soft Computing is a computing paradigm that is tolerant of imprecision, uncertainty and partial truth. The motivation behind the paradigm is human mind. Conventional computing is considered to be ‘hard’, because it is limited to represent human like ‘vagueness’. Soft computing provides the opportunity to represent ‘ambiguity’ in human thinking with the real life ‘uncertainty’. The concept is proved to be useful for many real life problems. The principal constituents of soft computing are Fuzzy, Neuro, Evolutionary and Probabilistic computing techniques. This is a partnership, where each technique contributes a unique methodology for addressing problems in its domain.

Modern engineering industry is relying more and more on the use of knowledge in product development. This paper advocates design repositories as a natural progression from traditional design databases to systems that are created to more actively support knowledge-based design. In contrast to traditional design databases, design repositories serve not only as archives, but as repositories of heterogeneous information that are designed to enable representation, capture, sharing, and reuse of corporate design knowledge. This paper describes the NIST Design Repository Project, an ongoing project within the Engineering Design Technologies Group at the National Institute of Standards and Technology (NIST). The project objectives are to develop a computational framework for the creation of design repositories, and a proof-of-concept prototype to demonstrate their benefits. A number of research issues associated with the envisioned role of design repositories in industry are addressed. The current state of the project and its implementation are presented.

The paper contains a discussion material and preliminary experimental results on a new approach to building on-line, adaptive decision making and control systems. This approach is called evolving connectionist systems (ECOS). ECOS evolve through incremental, on-line learning. They can accommodate any new input data, including new features, new classes, etc. New connections and new neurons are created during operation. The ECOS framework is illustrated here on a particular type of evolving neural networks - evolving fuzzy neural networks. ECOS are three to six orders of magnitude faster than the multilayer perceptrons, or fuzzy neural networks, trained with the backpropagation algorithm or with a genetic algorithm. ECOS are appropriate techniques to use for creating on-line, real-time, adaptive intelligent systems. This is illustrated on a case study problem of on-line, wastewater time-series flow prediction and control. Possible real world applications of this approach are discussed.

Evolutionary programming (EP) is one of the major branches of evolutionary computation. It has been applied to many learning and optimisation problems with success in recent years. This paper gives an overview of the latest results on evolutionary programming. In particular, the paper will analyse why the recently proposed fast EP performs better than classical EP for most benchmark functions, discuss the scalability of EP, and demonstrate how EP has been used to solve the function optimisation problem and neural network design problem.

Recently, it has been actively exploited to combine several softcomputing techniques such as fuzzy logic, neural networks and evolutionary computation to develop better systems that can match human ability, but more work is still required to be able to match human performance. In order to remedy this shortcoming, several approaches to incorporate human himself into the softcomputing framework have been devised, and among them is interactive evolutionary computation in which the genetic optimization is conducted interactively with human evaluation. This paper presents the potential of interactive evolutionary computation in an engineering problem, called content-based image retrieval. In this system, each individual in population is evaluated by user and the next generation is produced based on the user’s evaluation, by which it might incorporate human preference into the process of image retrieval. Experiments with a small database indicate that the interactive evolutionary computation is useful in content-based image retrieval.

This paper discusses how a genetic algorithm searches through a design space to generate highly fit designs for an example engineering design problem. The problem used herein is the design of a stiffened composite panel, combining discrete, integer and continuous variables. The GA process is examined and compared to an accepted view of creative engineering design practices. Examples from the stiffened composite panel problem are used to illustrate how the genetic algorithm may encourage creativity and / or act as a creative agent in engineering design. To be successful, a genetic algorithm search still requires intelligent, creative design engineers; this need is also discussed.

The paper deals with an abduction problem within the approach for handling probabilistic type of uncertainty in the framework of Constraint Logic Programming. The approach is intended for solving real-life logistic problems such as scheduling, manufacturing, planning, considering uncertainty of input information. The Monte-Carlo method used as background engine allows to combine logical inference with probability and straightforward statistical modeling features. We describe particular cases of abduction problem with different kinds of input and output data (probabilities of clauses and bounds of stochastic variables). We also discuss the problem of determining the dependence between input and output domain variables, and show that this dependence is multilinear in the particular case with independent uncertain clauses. Graph with uncertain edges exemplifies the discussion. At last, we demonstrate the use of some of described methods in the pilot application for solving the network based project planning problem in extended non-deterministic formulation.

This paper examines the four main types of Evolutionary Design by computers: Evolutionary Design Optimisation, Evolutionary Art, Evolutionary Artificial Life Forms and Creative Evolutionary Design. Definitions for all four areas are provided. A review of current work in each of these areas is given, with examples of the types of applications that have been tackled. The different properties and requirements of each are examined. Descriptions of typical representations and evolutionary algorithms are provided and examples of designs evolved using these techniques are shown. The paper then discusses how the boundaries of these areas are beginning to merge, resulting in four new ‘overlapping’ types of Evolutionary Design: Integral Evolutionary Design, Artificial Life Based Evolutionary Design, Aesthetic Evolutionary AL and Aesthetic Evolutionary Design. Finally, the last part of the paper discusses some common problems faced by creators of Evolutionary Design systems, including: interdependent elements in designs, epistasis, and constraint handling.

The design of surfaces often involves multiple criteria, which are governed by a number of geometry constraints. These constraints satisfy certain demands relating to curvature, and make the optimisation of surfaces quiet complex. A flexible approach to design optimisation is essential to handle the diverse criteria and constraints. The paper presents a framework for flexible surface optimisation within a CAD/CAM environment using the robust evolutionary computing technique of Genetic Algorithms (GAs). Particular needs for optimisation in a CAD/CAM environment are discussed with relation to current state of the art research and practice in surface optimisation. A simple GA is integrated in an existing CATIA CAD/CAM system, to provide a foundation for the flexible optimisation of freeform curves and surfaces. The paper presents some initial results on surface optimisation. Directions for future research are also identified.

This paper proposes an inverse fuzzy-model-based controller. The gradient-descent algorithm can be used on-line to form adaptive fuzzy controllers. This ability allows the controller to be used in applications where the knowledge to control the process does not exist or the process is subject to changes in its dynamic characteristics. To demonstrate the feasibility of the method simulation and experimental control tests were evaluated in a non-linear level process, composed of mono-tank, which was submitted to reference change. The proposed adaptive Sugeno fuzzy logic controller is shown to be capable of compensating non-linear terms that affect the systems dynamics and providing better overall system performance than the optimal PI controller.

Structures and design issues of fuzzy PID (proportional-integral-derivative) controllers (FLC-PID’s) are presented and evaluated in this paper. Configuration and basic characteristic of several structures of FLC-PID based on models proposed in the literature — (PD + I), (PI + D conventional), incremental (PD + I), (PD + PI) — are here reviewed and implemented. FLC-PID’s are assessed on a horizontal balance process, consisting of two propellers driven by two DC motors. Such process offers control complexities and can become unstable by using classical controllers. Experimental results, robustness and performance of FLC-PID’s are illustrated and discussed.

This paper proposes an accurate COG (Center of Gravity) defuzzification method that improves both the system’s approximation behavior and the control performance of a fuzzy logic controller (FLC). The premise of this paper is that the defuzzifcation is a process of optimal selection and an appropriate selection of the defuzzification method can improve the approximation behavior and the control performance of the FLC greatly. The accuracy of the proposed COG defuzzifier is obtained by representing the output membership functions (MFs) with various design parameters such as the centers, widths, and modifiers of MFs and by adjusting these design parameters with Lamarckian co-adaptation of learning and evolution. This co-adaptation scheme allows to evolve much faster than the nonlearning case and gives a higher possibility of finding an optimal solution due to its wider searching capability. An application to the truck backer-upper control problem of the proposed co-adaptive design method of COG defuzzifier is presented. The approximation behavior and control performance of the proposed COG defuzzifier are compared with those of the conventionally simplified COG defuzzifier in terms of the system error and the average tracing distance, respectively.

This paper is devoted to an intelligent approach towards the total glass production management in glass furnaces, and to a presentation of recent results. The global approach based on three-layered multiagent expert system architecture is explained, together with some details about various furnace parts control.

This paper presents an algorithm for predictive control using fuzzy models. This strategy is a Predictive Control Strategy for Nonlinear systems. The algorithm takes advantage of the already well known methods for linear systems (like Generalized Predictive Control). Analysis of the modified algorithm is presented and performance comparisons are shown. The performance comparisons are made using a model of a chemical process (Continuous Stirring Tank Reactor) where noise and uncertainties are included.

This paper presents an evolutionary design method for fuzzy logic controllers, which is based on a self-organizing process that learns the appropriate relationship between control input and output. Our approach employs an evolution strategy that operates on vectors of real numbers which correspond to the gain factors in the conclusion part of fuzzy rules. An incremental learning scheme gradually expands the genome and thereby refines the fuzzy knowledge base that acquires additional fuzzy rules.The paper describes the fuzzy control autopilot that constitutes the regulation layer of a hybrid flight vehicle management system, which governs the operation of an autonomous model helicopter. The autopilot is composed of four modules that control the longitudinal and lateral motion, altitude and heading. The modules are implemented by a hierarchy of fuzzy controllers. The evolutionary algorithm optimizes the fuzzy rule bases off-line. We compare two design approaches, learning the rule base starting without previous knowledge and tuning an predefined set of fuzzy rules.

A concept called the decomposition of multivariable control rules is presented. Fuzzy control is the application of the compositional rule of inference and it is shown how the inference of the rule base with complex rules can be reduced to the inference of a number of rule bases with simple rules. A fuzzy logic based controller is applied to a simple magnetic suspension system. The controller has proportional, integral and derivative separate parts which are tuned independently. This means that all parts have their own rule bases. By testing it was formed out that the fuzzy PID controller gives better performance over a typical operational range then a traditional linear PID controller. The magnetic suspension system and the contact-less optical position measurement system have been developed and applied for the comparative analysis of the real-time conventional PID control and the fuzzy control.

This paper evaluates methodologies of the computational intelligence applied to a nonlinear process identification. The different intelligent methodologies are: hybrid genetic algorithm with simulated annealing and modified Elman neural network. Experimental tests of identification are carried out using a laboratory scale wind tunnel plant. Experiments, for the estimation of the mathematical model of the process, are carried out using step function and pseudo-random binary sequence signals as process input. A white noise signal is used in the validation phase. The performance of each technique, in the estimation and validation phases of the process mathematical model, is presented and discussed.

Fuzzy sets are used in various fields. For those who have little knowledge on fuzzy set theory we find that improving usability of membership function identification method is really important. This paper aims to propose this new method improved in usability and precision, and then to verify the method through experiments. First, we propose the method to identify a trapezoidal membership function, BASE method (boundary asymptotic estimation method). The features of this method are ternary rating of membership grades, asymptotic estimation of boundaries of 1-level set and support set, and effective and recursive selection of elements using computers. Results from psychological experiments are showing that the BASE method is superior to the computerized fuzzy graphic rating scale both in usability for inexperienced users of fuzzy set theory and precision.

In this paper an automatic fuzzy rule generation problem through the artificial neural network (ANN) approach is considered. The unknown fuzzy relation reconstruction problem is treated as an optimization of the structure and parameters of the neural network. The functional equivalence between some classes of fuzzy systems and radial basis function networks (RBFNs), namely, their localized sensitivity to input value, is a background of the proposed approach. The improved structure and advanced learning feature RBFN is developed based on General Parameter (GP) method of complex system identification. The criterion of the GP RBFN (General Parameter Radial Basis Function Network) structure optimality is derived using the GP steady state statistics. The derived criterion is used then for the development of the GP RBFN structure self-organization procedure. As a result, an Adaptive Fuzzy System (AFS) with capability to extract fuzzy If-Then rules from input and output sample data is proposed. Simulation examples are given.

Fuzzy modeling is a promising technique to describe input-output relationships of nonlinear systems with fuzzy if-then rules. It is easy for us to understand the linguistically expressed nonlinear relationships. This paper presents a new dividing method of input spaces for a hierarchical fuzzy modeling method using Fuzzy Neural Network (FNN) and Genetic Algorithm (GA). Uneven division of input space for each submodel in the hierarchical fuzzy model can be achieved with the proposed method. The obtained hierarchical fuzzy model are probable to be more concise and more precise than those identified with the conventional methods.

Recent developments have shown the usefulness of combinations of classifiers. This paper presents a 3-Level Stacked Generalizer consisting of evolutionary created Artificial Neural Networks with different architectures and weight settings and a nearest neighbour classifier. The components of the Stacked Generalizer have been obtained by an Evolutionary Algorithm to optimize the architecture of an Artificial Neural Network.Several combination and new selection methods are discussed and their usefulness is shown in practical examples. The use of Ensembles allows it, to obtain better classifiers in earlier generations of the evolutionary optimization process, reducing the necessary computing time. It can be shown that it is more efficient to use information obtained from the complete population, enriched by classical classifiers, rather than taking the best individual.

The goal of multi-dimensional fuzzy analysis consists in discovering different properties in multi-dimensional fuzzy distributions represented either extensionally (database) or intensionally (knowledge base). In this paper we show how this approach can be applied to such problems as decision making and knowledge discovery in databases. For uniform and efficient representation of fuzzy knowledge and data we propose a technique of sectioned matrices. For carrying out logical inference we use a new operation of fuzzy resolution which is a generalization of the conventional resolution. With the help of this operation we find fuzzy prime disjunctions which later are used for making decisions in concrete situations. For discovering hidden dependencies in data a new fuzzy covering method is used.

We suggest a novel, unified approach to automating the entire process of Knowledge Discovery in Databases (KDD). The approach builds upon Shannon’s Information Theory, statistical estimation methods, and Fuzzy Logic. The KDD stages to be automated include: dimensionality reduction, discovering informative rules (patterns), predicting values of unknown attributes, and cleaning a dataset from lowly reliable data.The relational database model is extended by partitioning the relation scheme into a subset of candidate input (predictive) attributes and a subset of target (classification) attributes. A multilevel information-theoretic connectionist network is constructed to evaluate the mutual information between input and target attributes. The optimal network structure is found by a stepwise gradient procedure. The network connection weights are used to extract informative rules and to calculate reliability degrees of target attributes.The approach is applied to a real-world manufacturing database containing typical data on production batches at a semiconductor factory.

The number of electronic catalogs has grown rapidly during the past few years. Most of these catalogs use standard databases for storing and retrieving product information. Using ordinary databases for product catalogs, however, has the major drawback that it is often very difficult to find the products desired: very often, the database does not return a matching product at all or it returns many products that have to be examined manually. To overcome this problem, we propose the use of Case-Based Reasoning(CBR) techniques as an approach to requirement-oriented retrieval of products. CBR incorporates product knowledge into the database by means of a similarity measure. Recently, a number of commercial electronic catalogs based on CBR have been realized. We take a closer look at Analog Devices’ on-line catalog of operational amplifiers, which helps an engineer to find a suitable amplifier for her specific requirements.

Data mining consists of the efficient discovery of knowledge from databases. This paper presents a new genetic algorithm designed for discovering a few interesting, high-level prediction rules from databases, rather than discovering classification knowledge (often a large rule set) as usual in the literature. Three important data mining issues addressed by our algorithm are the interestingness of the discovered knowledge, the computational efficiency of the algorithm, and the trade-off between representation expressiveness and efficiency.

This paper presents the extension of the Neural Evolutional Strategy System (Nessy) to the multiobjective optimization case. The neural architecture of the Nessy algorithm is extended by using more than one output neuron, one neuron for each objective. The learning law of Nessy is modified according to the presence of multiple measures of performance. Each hidden neuron of the generation layer randomly selects an objective for one cycle of the network. From this, the multiobjective ranking of the population (or neurons of the solutions layer) is stochastically approximated. The modified Nessy algorithm (Monessy) is able to search for the Pareto set of a multiobjective optimization problem. A test function from literature with well-known Pareto and trade-off set is examined. The newly proposed algorithm effectively searches for the Pareto set by switching between explorational and exploitational search phases. This was compared with random search, which did not hit the Pareto set as nearly as often as the Monessy algorithm. Also, the replacement of a weighted-sum matching measure with multiple matching measures in a framework for texture filter design is considered as a second example.

In this paper a new optimization algorithm is presented. It is inspired by the natural behavior of human or animal scout in exploring unknown geographical regions, and in their ability to exploit information coming from past experience. The algorithm has been applied to the maximum clique problem, and favorably compared with other more specialized technique.

This paper presents an iterative method that performs a succession of box searches. For multimodal functions, provided that the size of the box is larger than that of the valleys of the objective function, the method is able to jump across those valleys without getting trapped in local optima. To carry out each box search we propose using a Genetic Algorithm, which are known to be robust and efficient search methods. Our implementation of the Genetic Box Search concept uses floating point-coded individuals, convex linear combination crossover and gaussian mutation. Computational experiences show that this approach is very robust with respect to the starting point, is very effective at converging to the global optimum and has low computational requirements.

This paper investigates the applicability of the improvement of simple genetic algorithm (SGA) method for solving the uncapacitated warehouse location problem (UWLP). Function for computing the item’s objective value is improved depending upon the number of established warehouses. It is efficiently implemented, giving excellent results in specified environment. Mutation rate is also changed and now depends on test problem size. Duplicate item strings in population are discarded, which makes the population more diversified.Overall performance of implementation is finally tuned by caching SGA. Through caching technique relatively smaller profit in performance is obtained when compared to previous techniques, but it is a general technique, and can be directly applied to other problems, not only to UWLP.Computational experience with given problem examples indicates that, when compared to code in [1], the implementation of the modified algorithm is faster several times. For large size test problems the increase of computational speed may even exceed factor 10, and quality of obtained solutions is also significantly better.

In this paper we investigate the effective design of an appropriate neural network model for time series prediction based on an evolutionary approach. In particular, the Breeder Genetic Algorithms are considered to face contemporaneously the optimization of (i) the design of a neural network architecture and (ii) the choice of the best learning method. The effectiveness of the approach proposed is evaluated on a standard benchmark for prediction models, the Mackey-Glass series.

The Evolution Strategies algorithm for maximum entropy image restoration is proposed. Evolutionary algorithms are widely used to search for solutions of “ill-posed” optimization problems. These are based on simulating the natural evolution process within a population of individuals. The statistical properties of the mutation operator are discussed. The approximate formulas to choose the reasonable mutation parameters for a population in the case of a maximum entropy objective function are presented. The algorithm is tested on a real and simulated astrophysical data. For the purpose of comparison, restoration results, based on Evolution Strategies MEM and gradient MEM for synthetic images are shown.

A programme to automate the finite element method is discussed, and the implementations of three specific sub-problems (node numbering, mesh placement, and adaptive meshing) are described. It is also argued that the overall architecture of an “intelligent finite element package” can serve as a “test-bed” for many soft-computing techniques.

This paper presents a new tabu search approach to the solution of a mixed integer linear programming formulation of the tool assignment and machine loading problem in flexible manufacturing systems. Our heuristic search method uses candidate list strategy, diversification, intensification and strategic oscillation elements. To test our heuristic approach, a set of test problems has been randomly generated and pseudoptimally solved by a mathematical programming software package. Computational results disclose that our tabu search approach is most effective, in terms of both solution quality and efficiency, for larger problem instances.

This paper examines the application of an evolutionary algorithm (GENO-COP III) to the problem of fitting surfaces and lines to both 2D synthetic and real 3D range data. The fitting is performed with both non-linear (domain) constraints and with non-linear (geometric and relational) constraints. Example fittings are given as well as an explanation of experimantal caveats that should be addressed when using this kind of approach. We have discovered, for example, that the time to generate starting reference points on the solution manifold is a significant part of the experiment time.

For storage of gases pressure vessels are used. The pressure, material, manufacturing method, sizes,type and number of vessels are quantities that a designer has to decide on. Usually the cost is a function one wants to minimize. In this paper a general structure for the various variables is proposed. This structure can be regarded as an expert system for a certain class of pressure vessels. The problem of selecting the various variables and making the right choices can be formulated as an optimization problem with mixed discrete/continuous variables and with variables structured in tables. The optimization procedure is integrated in the whole design process. An existing optimization program based on genetic algorithms that is able to handle those kind of variables is used to arrive at a well-balanced choice. A case study is given to illustrate the method.

Three different levels of sophistication of genetic algoriithms (GAs) are described. (1) The first level are GAs, where the chromosome fitness is determined exactly by the chromosome position on the fitness landscape. This first level of sophistication corresponds to standard GAs, where chromosomes are directly mapped on positive real numbers. (2) In the second level the chromosome fitness is determined by the nearest neighborhood of the chromosome position on the fitness landscape, i.e. chromosomes are capable of learning. The role of learning in evolution theory called the Baldwin effect was first studied by GAs in 1987 by Hinton and Nowlan. (3) In the third level the chromosome fitness is determined not only by the nearest neighborhood of the chromosome position on the fitness landscape but also by the so-called meme that determines an information that is able to increase the fitness of chromosomes. The idea of memes was introduced to evolutionary biology by Dawkins. The concept of chromosome is enlarged to a complex of chromosome and meme, both of them determine a fitness of chromosomes itself. All the above three different levels of GAs may be formally considered as evolutionary steps of Darwin’s evolution. The purpose of this communication is to present a GA simulation of learning and Dawkins’ memes.

The paper is devoted to one possible approach to a generalization of formulation of structure synthesis problem by the genetic algorithms for some applications. The approach includes a substitution of design parameter representation in chromosome by representation of rules for choice of parameter values in accordance with Heuristics Combination Method. Number and sense of parameters are defined from information model of application similar IDEF1X model. A transition from one application to another requires replacement of rules (heuristics) set and algorithm of fitness function calculation. Also the paper includes the results of solving some problems such as OSSP, VRPTW and distribution of subnetworks in VLAN (application distribution problem).

Altruism belongs to “social attributes” of individuals in a population. Model of “kinship” altruism is studied, where an altruistic individual is more likely to help its relative.In our model simulations by a modified genetic algorithm the chromosomes were defined by a presence or an absence of an altruistic disposition, affiliation to a group (kinship relation), and a fitness, initially random. No objective function is used, new fitness is created in a reproduction process from an arithmetic mean of parental fitness modified by a random number with normal distribution.Altruistic process increases fitness of a randomly chosen chromosome-recipient, and decreases fitness of a randomly chosen chromosome with an altruistic allele-donor. The increase/decrease is unproportional to an absolute value of the difference of their affiliated groups. Results suggest, that an altruistic behavior in the initial stage was created by a random genetic drift in a subgroup and initial division of a population into subgroups is crucial for its evolutionary (meta)stability.The presented simulation can be useful for evolutionary biology as well as for a modification of genetic algorithms (e.g. in artificial agents) towards natural systems.

The paper deals with the Bivariate Marginal Distribution Algorithm (BMDA). BMDA is an extension of the Univariate Marginal Distribution Algorithm (UMDA). It uses the pair gene dependencies in order to improve algorithms that use simple univariate marginal distributions. BMDA is a special case of the Factorization Distribution Algorithm, but without any problem specific knowledge in the initial stage. The dependencies are being discovered during the optimization process itself. In this paper BMDA is described in detail. BMDA is compared to different algorithms including the simple genetic algorithm with different crossover methods and UMDA. For some fitness functions the relation between problem size and the number of fitness evaluations until convergence is shown.

A set-theoretic framework is proposed for granular computing. Each element of a universe is associated with a nonempty family of neighborhoods. A neighborhood of an element consists of those elements that are drawn towards that element by indistinguishability, similarity, proximity, or functionality. It is a granule containing the element. A neighborhood system is a family of granules, which is the available information or knowledge for granular computing. Operations on neighborhood systems, such as complement, intersection, and union, are defined by extending set-theoretic operations. They provide a basis of the proposed framework of granular computing. Using this framework, we examine the notions of rough sets and qualitative fuzzy sets.

This paper sketches a new approach to the use of fuzzy concepts in natural language processing (NLP). The main issue addressed is the fuzzification of the „automaton-driven“ parser, which works in the stratificational knowledge representation system (SKRS).The proposed self-learning analyzer of a dialog system can learn the grammatical rules from the gained sentences imitating human behavior.

This paper presents a new approach to acquisition of comprehensible fuzzy rules for fuzzy modeling from data. For the accuracy of the model, the identified model is probable to have many membership functions which are overlapped with each other. From the viewpoint of knowledge acquisition, it is desirable that the model has a smaller number of membership functions without the overlaps. Considering that the precision and the clarity of the fuzzy model are the trade-off, we propose an acquisition method of comprehensible fuzzy rules from the identified model which satisfies the desired accuracy. The precise model is to be obtained by, for example, fuzzy neural network(FNN) and using evolutionary algorithm. This paper proposes to use evolutionary programming(EP) to extract comprehensible rules from the fuzzy model. A numerical experiment is done to show the feasibility of the proposed method.

“Blind” signal processing techniques have attracted a great deal of interest from the signal processing community in recent years. The term “blind” refers to the fact that very little information is known about the signals or processing in question. This paper deals with blind signal separation, the separation of unknown signals that have been mixed in an unknown way. In this paper the linear mixture model has been adopted; the signals in question are assumed to be linear mixtures of stationary sources. Techniques that address this problem using higher-order statistics rely on the accurate estimation of higher-order moments, either implicitly or explicitly. This can cause difficulties in real-time adaptive systems as anomalous outliers in the data can have disproportionate effects on the system. We introduce a novel on-line method for separating super-Gaussian signal mixtures that does not rely on higher-order moments, called “zero-point probability” (ZPP).

Since the Adleman’s experimental demonstration of its feasibility, DNA computing has been applied to a number of combinatorial optimization problems. Several experimental results have shown that DNA strands can be used to compute solutions to NP-complete problems. Usually they employ random codes to represent candidate solutions in DNA. However, some codes have better error-tolerance than others since current bio-lab experiment techniques are involved with reaction errors. In this paper we present an evolutionary method for optimizing the DNA codes for solving a given problem. The method uses a genetic algorithm to find best codes by emulating chemical reaction processes before actual biological experiments begin. Simulations have been performed to solve maximum clique problems, an NP-complete search problem. The results show that the optimized codes improve the reliability of DNA computing and reduce time and costs that may be caused by repeated bio-experiments.

This tutorial is intended to provide a comprehensive and broad overview of the main techniques, algorithms, application issues and software in relation with the growing area of evolutionary computation (EC). It includes in an on-line fashion a flexible access to different chapters containing explanations, examples and demos on the different algorithm families embedded in the EC-like search mechanisms. The tutorial is highly structured in chapters and sections, and also it offers a bibliography on the matter, as well as descriptions of concrete applications and theoretical details on evolutionary algorithms. The dynamic nature of such an info suggests the advantages of an on-line format. This allows a continuous improvement by the addition of results, graphs and new links to related material.

The tutorial concerns automatic control of an inverted pendulum, especially rule based control by means of fuzzy logic. A ball balancer, implemented in a software simulator in Matlab, is used as a practical case study. The objectives of the tutorial are to teach the basics of fuzzy control, and to show how to apply fuzzy logic in automatic control. The tutorial is distance learning, where students interact one-to-one with the teacher using e-mail.

The papers presented in the first nine parts of this book comprise of the papers presented in the 3rd On-line World Conference on Soft Computing in Engineering Design and Manufacturing (WSC3). This part summarises the discussion about the papers during the Conf rence. The summary is divided into nine sections according to the first nine parts in the book, and most of the sections have been prepared by the session chair(s). The name of the person who has prepared the summary is given at the beginning of each section.