Kwong-Sak Leung

Convergence analysis of cellular neural networks with unbounded delay

Cellular neural networks (CNNs) have been successfully applied in many areas such as classification of patterns, image processing, associative memories, etc. Since they are inherently local in nature, they can be easily implemented in very large scale integration. In the processing of static images, CNNs without delay are often applied whereas in the processing of moving images, CNNs with delay have been found more suitable. This paper proposes a more general model of CNNs with unbounded delay, which may have potential applications in processing such motion related phenomena as moving images, and studies global convergence properties of this model. The dynamic behaviors of CNNs, especially their convergence properties, play important roles in applications. This paper: (1) introduces a class of CNNs with unbounded delay; (2) gives some interesting properties of a networks output function; (3) establishes relationships between a networks state stability and its output stability; and (4) obtains simple and easily checkable conditions for global convergence by functional differential equation methods.

Fuzzy concepts in expert systems

The authors present a comprehensive expert-system building tool, called System Z-II, that can deal with exact, fuzzy (or inexact), and combined reasoning, allowing fuzzy and normal terms to be freely mixed in the rules and facts of an expert system. This fully implemented tool has been used to build several expert systems in the fields of student curriculum advisement, medical diagnosis, psychoanalysis, and risk analysis. System Z-II is a rule-based system that uses fuzzy logic and fuzzy numbers for its inexact reasoning. It uses two basic inexact concepts, fuzziness and uncertainty, which are distinct from each other in the system.<<ETX>>

A genetic algorithm for determining nonadditive set functions in information fusion

As a classical aggregation tool, the weighted average method is widely used in information fusion. It is the Lebesgue integral with respect to the weights essentially. Due to some inherent interaction among diverse information sources, the weighted average method does not work well in many real problems. To describe the interaction, an intuitive and effective way is to replace the additive weights with a nonadditive set function defined on the power set of the set of all information sources. Instead of the weighted average method, we should use the Choquet integral or some other nonlinear integrals, especially, the new nonlinear integral introduced by the authors recently. The crux of making such an improvement is how to determine the nonadditive set function from given input-output data when the nonlinear integral is viewed as a multi-input single-output system. In this paper, we employ a specially designed genetic algorithm to realize the optimization in determining the nonadditive set function.

Mechanical properties of normal skin and hypertrophic scars

The non-linear viscoelastic properties of skin tissue were characterized by modulus of elasticity E, which represents stiffness, and percentage extension (strain, xi) at load intensities of 20, 40 and 100 gm. The latter property is a measure of the extensibility for both normal skin and hypertrophic scar. A quasi-static extensometer applies a standard rate of extension to the skin and its mechanical properties were obtained from 15 Chinese patients with burn injuries of superficial to full skin thickness burns. Clinical evaluation of hypertrophic scar is qualitative and depends on colour, thickness and hardness or firmness. Using mechanical properties assists in the characterization by providing a quantitative indicator. Higher scar grading is synonymous with increased stiffness and decreased extensibility. Correlation with clinical assessment was achieved with these in vivo viscoelastic properties.

Using evolutionary programming and minimum description length principle for data mining of Bayesian networks

We have developed a new approach to learning Bayesian network structures based on the minimum description length (MDL) principle and evolutionary programming. It employs a MDL metric, which is founded on information theory, and integrates a knowledge-guided genetic operator for the optimization in the search process.

An expanding self-organizing neural network for the traveling salesman problem

The self-organizing map (SOM) has been successfully employed to handle the Euclidean traveling salesman problem (TSP). By incorporating its neighborhood preserving property and the convex-hull property of the TSP, we introduce a new SOM-like neural network, called the expanding SOM (ESOM). In each learning iteration, the ESOM draws the excited neurons close to the input city, and in the meantime pushes them towards the convex-hull of cities cooperatively. The ESOM may acquire the neighborhood preserving property and the convex-hull property of the TSP, and hence it can yield near-optimal solutions. Its feasibility is analyzed theoretically and empirically. A series of experiments are conducted on both synthetic and benchmark TSPs, whose sizes range from 50 to 2400 cities. Experimental results demonstrate the superiority of the ESOM over several typical SOMs such as the SOM developed by Budinich, the convex elastic net, and the KNIES algorithms. Though its solution accuracy is not yet comparable to some other sophisticated heuristics, the ESOM is one of the most accurate neural networks for the TSP in the literature.

An efficient data mining method for learning Bayesian networks using an evolutionary algorithm-based hybrid approach

Given the explosive growth of data collected from current business environment, data mining can potentially discover new knowledge to improve managerial decision making. This paper proposes a novel data mining approach that employs an evolutionary algorithm to discover knowledge represented in Bayesian networks. The approach is applied successfully to handle the business problem of finding response models from direct marketing data. Learning Bayesian networks from data is a difficult problem. There are two different approaches to the network learning problem. The first one uses dependency analysis, while the second one searches good network structures according to a metric. Unfortunately, both approaches have their own drawbacks. Thus, we propose a novel hybrid algorithm of the two approaches, which consists of two phases, namely, the conditional independence (CI) test and the search phases. In the CI test phase, dependency analysis is conducted to reduce the size of the search space. In the search phase, good Bayesian network models are generated by using an evolutionary algorithm. A new operator is introduced to further enhance the search effectiveness and efficiency. In a number of experiments and comparisons, the hybrid algorithm outperforms MDLEP, our previous algorithm which uses evolutionary programming (EP) for network learning, and other network learning algorithms. We then apply the approach to two data sets of direct marketing and compare the performance of the evolved Bayesian networks obtained by the new algorithm with those by MDLEP, the logistic regression models, the na/spl inodot//spl uml/ve Bayesian classifiers, and the tree-augmented na/spl inodot//spl uml/ve Bayesian network classifiers (TAN). In the comparison, the new algorithm outperforms the others.

Applying fuzzy measures and nonlinear integrals in data mining

The paper gives an overview of applying fuzzy measures and relevant nonlinear integrals in data mining, discussed in five application areas: set function identification, nonlinear multiregression, nonlinear classification, networks, and fuzzy data analysis. In these areas, fuzzy measures allow us to describe interactions among feature attributes towards a certain target (objective attribute), while nonlinear integrals serve as aggregation tools to combine information from feature attributes. Values of fuzzy measures in these applications are unknown and are optimally determined via a soft computing technique based on given data.

An efficient self-organizing map designed by genetic algorithms for the traveling salesman problem

As a typical combinatorial optimization problem, the traveling salesman problem (TSP) has attracted extensive research interest. In this paper, we develop a self-organizing map (SOM) with a novel learning rule. It is called the integrated SOM (ISOM) since its learning rule integrates the three learning mechanisms in the SOM literature. Within a single learning step, the excited neuron is first dragged toward the input city, then pushed to the convex hull of the TSP, and finally drawn toward the middle point of its two neighboring neurons. A genetic algorithm is successfully specified to determine the elaborate coordination among the three learning mechanisms as well as the suitable parameter setting. The evolved ISOM (eISOM) is examined on three sets of TSP to demonstrate its power and efficiency. The computation complexity of the eISOM is quadratic, which is comparable to other SOM-like neural networks. Moreover, the eISOM can generate more accurate solutions than several typical approaches for TSP including the SOM developed by Budinich, the expanding SOM, the convex elastic net, and the FLEXMAP algorithm. Though its solution accuracy is not yet comparable to some sophisticated heuristics, the eISOM is one of the most accurate neural networks for the TSP.

TFBS identification based on genetic algorithm with combined representations and adaptive post-processing

MOTIVATION Identification of transcription factor binding sites (TFBSs) plays an important role in deciphering the mechanisms of gene regulation. Recently, GAME, a Genetic Algorithm (GA)-based approach with iterative post-processing, has shown superior performance in TFBS identification. However, the basic GA in GAME is not elaborately designed, and may be trapped in local optima in real problems. The feature operators are only applied in the post-processing, but the final performance heavily depends on the GA output. Hence, both effectiveness and efficiency of the overall algorithm can be improved by introducing more advanced representations and novel operators in the GA, as well as designing the post-processing in an adaptive way. RESULTS We propose a novel framework GALF-P, consisting of Genetic Algorithm with Local Filtering (GALF) and adaptive post-processing techniques (-P), to achieve both effectiveness and efficiency for TFBS identification. GALF combines the position-led and consensus-led representations used separately in current GAs and employs a novel local filtering operator to get rid of false positives within an individual efficiently during the evolutionary process in the GA. Pre-selection is used to maintain diversity and avoid local optima. Post-processing with adaptive adding and removing is developed to handle general cases with arbitrary numbers of instances per sequence. GALF-P shows superior performance to GAME, MEME, BioProspector and BioOptimizer on synthetic datasets with difficult scenarios and real test datasets. GALF-P is also more robust and reliable when further compared with GAME, the current state-of-the-art approach. AVAILABILITY http://www.cse.cuhk.edu.hk/~tmchan/GALFP/.