International Symposium on Neural Networks 2006

Neuroinformatics combines neuroscience and informatics research to develop and apply the advanced tools and approaches that are essential for major advances in understanding the structure and function of the brain." (OECD MegaScience Forum report, 1999). Neuroinformatics has become increasingly important in neuroscience, providing an ideal opportunity to build stronger links between experimentalists and computer scientists, mathematicians, physicists and engineers. Among the techniques used, signal processing is becoming a vital tool to analyze the signals from brain such as EEG, MRI recordings. The analysis of neural signal is very helpful to understand the function of brain, for example coherence and synchronization of multiple EEG series can be successfully applied to understand the mechanism of epileptic seizures. In the 3th International Symposium on Neural Networks (ISNN 2006), we intend to bring together leading researchers in the area of neural signal processing to share their expert views and experiences.
We invite and welcome high quality contributions on a wide variety of topics relevant to neural signal processing. We will especially cover in-depth the following important topics:

Special Session on NN-based Optimization and Control for Complex Stochastic Processes

Organizers: Lei Guo, Southeast University, Nanjing, China (l.guo@seu.edu.cn)
Fei Liu, South Yangtze University, Wuxi, China (fliu@thmz.com)

Complex stochastic processes mainly include the stochastic processes with nonlinearity and non-Gaussian variables, for which the optimization and control for the error mean or variance will be insufficient. In this case, stochastic distribution functions and their entropies (in various definitions) can be used to characterize the stochastic property of the processes. Consequently, the entropy optimization also depends on the solutions of the output PDFs. However, it is well-known that the output probability density functions (PDFs) obey a nonlinear partial differential equation even for the so-called Ito equation. In practice, an analytical expression for the PDF of a random variable, which is necessary for the computation of the entropy, is not available in most cases.
In many practical processes (such as the batch processes), it is available to use neural networks (NNs) to model the output PDFs together with other tools such as Parzen windowing. With the NN expansions, one can set up the model-based analysis and synthesis for both the (non-Gaussian) signal processing and feedback optimization problems. In this case, it is possible to transform the infinite-dimensional optimization to a finite-dimensional problem.

This is a new direction in the control and signal processing fields, and has been shown great significance in many batch processes in engineering. Sub-topics include (but not limited to):

Organizers: Songcan Chen, Nanjing University of Aeronautics and Astronautics, China
(s.chen@nuaa.edu.cn)
Hujun Yin, The University of Manchester, UK (h.yin@manchester.ac.uk)

Self-Organisation is a widely observed phenomenon in natural organisms and bodies from neural networks to molecules and from societies to events. This topic and its associated learning paradigm has long interested and inspired theorists, researchers and practitioners in many fields to explore, advance and utilise in real-world applications. This session aims to gather recent advances and developments in all areas of self-organisation and applications. The sub-topics include but not limited to:

Self-Organisation Theory;
Unsupervised Learning;
Temporal Extension;
Hardware and Implementation;
Data Mining and Visualisation;
Bioinformatics and Biometrics Application;
Image and Vision Computing.

Special Session on Hybrid Neurocomputing in Finance Modeling and Forecasting

Organizers: Yuehui Chen, Jinan University, Shandong, Jinan, China (yhchen@ujn.edu.cn)
                       Ajith Abraham, Chung Ang University, Seoul, South Korea (ajith.abraham@ieee.org)
                      Chengling Gou, Beijing University of Aeronautics and Astronautics, China
                      (gouchengling@hotmail.com)

The special session aims to bring together professionals and the scientific community in the fields of financial engineering and hybrid neurocomputing in finance. Hybrid neurocomputing is a well-established paradigm, where new theories with a sound biological understanding have been evolving. Hybrid architectures like evolutionary neural networks, fuzzy neural networks, wavelet neural networks, flexible neural tree, multiple neural networks, hierarchical neural networks and so on, are widely applied for real-world problem solving. Hybrid neurocomputing techniques have the potential to impact many financial applications, from portfolio selection to proprietary trading to risk management. The special session greatly encourages new ideas/papers, combining two or more areas, such as evolutionary neural networks, fuzzy neural networks, wavelet neural networks, flexible neural tree, neural networks ensemble, multiple neural networks, hierarchical neural networks, etc. to be submitted. Topics of Interest include, but are not limited to any aspect of hybrid neurocomputing applications and theories that are involved in:

Artificial Stock Markets;
Behavioral Finance and Experimental Economics;
Financial Engineering and Financial Data Mining;
Trading and Hedging Strategies;
Portfolio Management;
Derivative Pricing;
Term Structure Models;
Financial Time Series Forecasting and Analysis;
Agent-Based Computational Finance;
Econophysics;
Program Trading.

Special Session on Intelligent Semiconductor Design and Manufacturing

Organizers: Tae Seon Kim, Catholic University of Korea (tkim@catholic.ac.kr)
Yean-Der Kuan, Northern Taiwan Institute of Science and Technology, Taiwan
(ydkuan@ntist.edu.tw)

Recently, the use of neural networks for modeling, optimization, and control of semiconductor manufacturing processes has becoming very popular and yielded very impressive results. However, neural networks based intelligent semiconductor manufacturing technologies are not matured yet since they are still at early stage level. For this reason, practical deployment of intelligent semiconductor manufacturing technologies is not yet achieved. The objective of this special session is to share various states of the art intelligent semiconductor manufacturing technologies with other researchers. Also, it can act as a catalyst for practical implementation of developed technologies.

Topics of interest:

Equipment/Process Modeling;
Device Design, Modeling & Analysis;
Process Optimization;
Process Control;
Chip Test & Measurement Techniques;
Thermal Management;
Yield Modeling & Reliability Analysis;
Production Planning and Job Scheduling.

Special Session on Extreme Learning Machine

Organization Chairs: Guang-Bin Huang and Meng-Hiot Lim, Nanyang Technological University, Singapore, ( EGBHuang@ntu.edu.sg, EMHLIM@ntu.edu.sg)

It is clear that the learning speed of neural networks is in general far slower than required and it has been a major bottleneck in their applications for past decades. Two key reasons behind may be: 1) the slow gradient-based learning algorithms are extensively used to train neural networks, and 2) all the parameters of the networks are tuned iteratively by using such learning algorithms. Although Support Vector Machine can produce better generalization performance, it faces two problems as well: 1) the intensive computation involved in its training which is at least quadratic with respect to the number of training examples; 2) large network size generated for large complex applications. In addition, some trivial works such as manually tuning parameters have to be done by users in the applications of these two technologies. A new emergent technology called extreme learning machine (ELM) which in theory tends to provide good generalization performance at extremely fast learning speed can overcome these problems easily. ELM can produce good generalization performance in most cases and can learn thousands of times faster than conventional popular learning algorithms for neural networks and support vector machines. More and more researchers have been conducting ELM related research. This session will provide a good platform for researchers to share their ideas and new results in this emergent area.

Special Session on Fuzzy neural networks for feedback control systems

Organizers: Wen Yu, CINVESTAV-IPN, Mexico (yuw@ctrl.cinvestav.mx)
Yu Tang, National University of Mexico, Mexico (tang@servidor.unam.mx).

    Both neural networks (NN) and fuzzy logic systems (FLS) are universal estimators. Resent results show that the fusion procedure of these two different technologies has significant advantages over standard feedback controllers for unknown nonlinear systems. Mostly, a neural network or a fuzzy logic system is used to approximate the nonlinearity of the system to be controlled and a controller is synthesized based on universal function approximators (indirect control), or a control law is directly designed using NN, or FLS based on stability theories. Another approach to feedback control design relies on using fuzzy neural networks to approximately solve various nonlinear controller design equations.
    In addition to the classical feedback control theory, adaptive control and robust control are effective techniques to treat system uncertainties but generally suffers from the disadvantage of being able to achieve asymptotical convergence of the tracking error, also the on-line computation load is usual heavy. In robust control designs, a fixed control law based on a prior information on the uncertainties (usually bounds on these uncertainties) is designed to compensate their effects, and exponential convergence of the tracking error to a (small) ball centered at the origin is obtained.
    There is a gap between control system community and computational intelligence (e.g. neural networks and fuzzy systems) society. The purpose of this session is to bring together fuzzy neural networks and feedback control design techniques.

The sub-topics include but not limited to fuzzy neural networks approaches in the following areas:

Feedback control using neural networks, fuzzy logic and fuzzy neural networks;
Robust neural (fuzzy) control;
Compensation of nonlinearities with (fuzzy) neural networks;
Identification and observers via (fuzzy) neural networks;
Applications of neural (fuzzy) control.

Special Session on Biomimetic Pattern Recognition

Organizers: Wenming Cao, Institute of Semiconductors, Chinese Academy of Sciences, Beijing, China ( wmcao@semi.ac.cn)
Shoujue Wang, Institute of Semiconductors, Chinese Academy of Sciences, Beijing, China (wsjue@red.semi.ac.cn )

This special session is focused on Biomimetic Pattern Recognition theory and applications in machine learning, speech recognition, speaker identification or verification, and face recognition. Biomimetic Pattern Recognition (BPR) , first proposed by Wang Shoujue as a new model for pattern recognition , is based on “matter cognition” instead of “matter classification”, so is thought closer to the function of human cognition than traditional statistical Pattern Recognition using “optimal separating” as its main principle. The method used by Biomimetic Pattern Recognition is “High-Dimensional Space Complex Geometrical Body Cover Recognition Method”, which studies some kinds of samples’ distribution in feature space and gives a reasonable cover, so the samples can be “recognized”. BPR has been used in a number of fields such as rigid object recognition, multi-camera face identification, DOA estimation and speech recognition and the results have shown its superiority.

Topics of interest:

Machine Learning Based on BPR;
Speech Recognition Based on BPR;
Speaker Identification or Verification Based on BPR;
Face Recognition Based on BPR;
Theory of BPR.

Special Session on NN Applications to Image Processing and Computer Vision

Organizers : Hoon Kang, Chung-Ang University, Seoul, Korea(hkang@cau.ac.kr)
Jin-Young Choi, Seoul National University, Seoul, Korea (jychoi@snu.ac.kr)

Neural networks (NNs) and computational Intelligence (CI) techniques can be applied to image processing and computer vision widely in the fields of pre-processing, feature extraction, segmentation, registration, tracking and recognition. The intelligent paradigms include associative memories, vector quantization, multilayer perceptron, fuzzy inference engine, evolutionary computations, and so on.
Visual data acquired from CCD or CMOS sensors are difficult to be dealt with, due to uncertainties and complexities such as luminous and color changes, occlusion, and the cluttered environment. In this session, it is pursued that these uncertain problems may be solved or circumvented, not by 'ad-hoc' prescriptions but by 'robust' treatments of invariant and/or adaptive solutions of NN and CI techniques to computer vision. In this context, the NN and CI-based approaches are quite challenging because they provide some real-time testbeds as well as intelligent interfaces to higher-level decision in computer vision. Detailed sub-topics related to NN and CI in image processing and computer vision are as follows (but not limited to):

Principal Component Analysis (PCA), Independent Component Analysis (ICA) in Computer Vision;
Evolutionary Particle Filter (EPF) in Visual Tracking & Recognition;
Competitive Learning or Vector Quantization (VQ) in Computer Vision;
Visual Tracking and Recognition in Evolutionary Neural Networks (ENN);
Neural Network-based Matching in Scale Invariant Feature Transform (SIFT);
Radial Basis Function (RBF) Networks in Computer Vision;
Associative Memories in Visual Recognition.

Special Session on Machine Learning Methods in Bioinformatics

Organizers: Xue-wen Chen and Ya Zhang, Department of Electrical Engineering and Computer
Science, The University of Kansas, USA (xwchen@ku.edu and yazhang@ittc.ku.edu)

The goal of this special session is to present cutting edge pattern recognition and machine learning methods with applications to bioinformatics. While such research is of interdisciplinary nature, this session will focus on computational aspects of bioinformatics research, especially novel learning methods such as neural network methods, kernel methods, support vector machines, and Bayesian data analysis. The aim of this session is to bring together an interdisciplinary group of researchers from pattern recognition, machine learning, and life science to discuss problems and to identify the opportunities and challenges in applying machine learning and pattern recognition to bioinformatics. High-quality, unpublished papers in the relevant areas will be solicited.

Relevant topics in bioinformatics for this session include, but not limited to:

Protein function analysis;
Protein structure classification;
Sequence alignment;
Microarray data analysis;
Genomic data analysis;
Gene and regulatory signal discovery;
Biological system pathways and networks.

Special Session on Security Issues on Neural Networks

Organizer: Tai-hoon Kim, Security Engineering Research Group, Defence Security Command, Korea
(taihoonn@empal.com)

We welcome all papers describing new and original results in application of security to Neural Networks. Topics of interest will focus on:

Applications of security engineering to NN applications or systems;
Applications of security engineering to NN systems development processes and operational environments;
Security Testing and Evaluation of NN systems;
Other applications of security engineering to NN systems.

Important Notice:
Selected papers in this session will be asked to extend and to be included in the Special Issue of Journal AJIT (Asian Journal of Information Technology). Please visit http://www.medwellonline.net/call4papers.html for further details.

Special Session on Neural Networks for Knowledge Discovery and Data Mining

Organizers: Ying Tan, Department of Computer Science and Technology, University of Science and
                       Technology of China, China (ytan@ustc.edu.cn)
                       Xufa Wang, Department of Computer Science and Technology, University of Science
                      and Technology of China, China (xfwang@ustc.edu.cn)

This special session is to present neural network methods for knowledge discovery and data mining. Although it has found a wide range of applications in science and engineering, neural networks, recently, receive a great successful application in knowledge discovery and data mining, and are regarded as versatile useful tools. A variety of neural networks, such as multilayer neural network, SOM, recurrent network, is used for association rule extraction, pattern search and discovery as well as other data mining tasks. The purpose of this session is to bring together active researchers from neural networks, data mining, and knowledge discovery to discuss some key problems of how to apply neural networks to knowledge discovery and data mining.

The sub-topics for this session, but not limited to, are as follows:

Association rule extraction and analysis;
FP-related algorithmic variants;
Clustering analysis and algorithm;
Pattern search and discovery;
Outlier analysis;
Trend analysis and forecast of financial data;
Visualization.