PROGRAM

8^th WSEAS International Conference on
EVOLUTIONARY COMPUTING
(EC '07)

Vancouver, Canada, June 18-19, 2007

Monday, June 18 2007

PLENARY LECTURE 1

Partitioning Capabilities of Multi-Layer Perceptrons

Assistant Professor Che-Chern Lin

Department of Industrial Technology Education

National Kaohsiung Normal University

Taiwan

Abstract: Recently, multi-layer-structured neural networks have been widely used in many areas. The back-propagation algorithm is one of popular training algorithms where a multi-layered network structure is designed to map inputs to outputs using well-trained weights. To minimize classification errors, it probably takes a lot of computational time in updating weights. Multi-layer perceptrons (MLPs) are different multi-layer-structured neural networks for classifications. This lecture theoretically discusses the partitioning capabilities of MLPs. We first explain how a MLP forms its decision region and how it performs mappings from inputs to desired outputs. A general discussion on the partitioning capabilities of a single-layer perceptron, two-layer perceptron, and three-layer perceptron is also given.  The implementation feasibilities on more complicated decision regions using MLPs are finally covered in this lecture.

PLENARY LECTURE 2

An Extension of the Crisp Ontology for Uncertain Information Modeling – Fuzzy Ontology Map

Associate Professor James Liu

Department of Computing

Hong Kong Polytechnic University

Abstract: In the current World Wide Web (WWW), users find it difficult to locate relevant information using search engines. This may be due to the fact that the current World Wide Web lacks semantic markup. One of the possible solutions for this problem is Semantic Web. In the latest Semantic Web technology, descriptive markup languages, such as Resource Description Framework (RDF) and Web Ontology Language (OWL), were proposed to model the web content in a machine-readable way which assists information gathering and automatic searching by software agents. Since these ontology markup languages deal with ‘hard’ semantics with the description and manipulation of crisp data, they are not capable of representing uncertain information while using current ontology representation.

This talk presents an extension of the current ontology representation which supports uncertain information modeling. The extension is called Fuzzy Ontology Map (FOM) which is based on the integration of fuzzy theory and graph theory. The FOM is a connection matrix which collects the membership values between classes in the ontology graph. Thus, a fuzzy ontology could be created by using the FOM and the ontology document (RDF/OWL). It is possible to use an FOCM for better knowledge management and information searching. We’ll focus on Web applications to develop systems that can deal with imprecise or vague information. Practical examples will be given and possible extension of the methodology incorporating the use of high level Petri nets will be provided for future work consideration.

PLENARY LECTURE 3

Applications of Meta-heuristics for Combinatorial Optimization Problems

Associate Professor Reza Tavakkoli-Moghaddam

Department of Industrial Engineering, Faculty of Engineering

University of Tehran, Tehran, Iran

Department of Mechanical Engineering

The University of British Columbia

Vancouver, Canada

Abstract: This plenary lecture deals with various applications of meta-heuristics to solve a number of the combinatorial optimization problems (COPs). It is divided into two main sections: (1) meta-heuristics and (2) optimization problems. In the original definition, meta-heuristics are stochastic/approximate solution methods that orchestrate an interaction between local improvement procedures and higher level strategies to create a process capable of escaping from local optima and performing a robust search of the solution space of the combinatorial optimization problems. With the development of complexity theory in the early 70's, it became clear that, since most of COPs were indeed NP-hard problems, there was little hope of ever finding efficient exact solution procedures for them. This realization emphasized the role of heuristics and meta-heuristics for solving such hard problems that were encountered in real-life applications and that needed to be tackled, whether or not they were NP-hard. Each meta-heuristic method balances the exploration and exploitation of the solution space using a specific strategy. The strategy is mainly inspired by the natural phenomena such as evolution, group cooperation, group competition, short/long-term memory, body immune system, self-replication, learning, and DNA/molecular computing or by other sciences such as annealing process, and quantum computations. Each strategy trades off between the effort and time spent to explore the new regions of the solution space and to exploit the explored regions. The solution representation and operator design are most significant aspects of implementing each meta-heuristic method. An important issue is that these two aspects are variety from a problem to another. In general, how to design operators is extremely depended on the structure of the solution representation and neighborhood definition. Thus, it is possible to consider the several ways for implementing a meta-heuristic method for a given problem. The closer the problems to the real-world situations, the harder the implementation of the meta-heuristics will be.

    The focus of this talk is on considering the above-mentioned aspects of the foregoing meta-heuristics by a number of various examples. These typical examples are: the dynamic cell formation problem, flexible flow lines scheduling problem, aggregate production planning problem, resource-constraint project planning problem, and vehicle routing problem. The considered examples are taken from the newly published work by the author.

PLENARY LECTURE 4

Industrial Applications using Artificial Intelligence and Statistical Techniques

Professor Anna Gabriela Perez de Rivas

Universidad de Los Andes

Facultad de Ciencias Economicas y Sociales

Escuela de Estadistica

Merida - Venezuela

Abstract: In this plenary it will be presented some industrial applications using artificial intelligence techniques as: Expert Systems, Neural Networks, Fuzzy Logic, Neo-Fuzzy Systems and Genetic Algorithms that have been used together with Statistical Techniques as: Data Analysis, Cluster Analysis, Time-series, Data Imputation, among others.

These applications include methodologies for designing virtual sensors, fault detection and isolation systems, classifier systems and controller Auto-Tuning systems. Some of these applications have been developed for oil companies.

Also, it will be included some research in neural networks that have been developed using variable structure control- based learning algorithms.

SESSION: Evolutionary Optimization Methods & Applications I

Chair: Qiang Hua, Igor Bernik

Tuesday, June 19 2007

SESSION: Evolutionary Computing Applications I

Chair: Jorge A. Tejedor, Haiyi Zhang

SESSION: Evolutionary Computing Applications II

Chair: Michael Rosenman, Rudolf Freund