Jun-Fen Chen

Multiple neural networks fusion model based on Choquet fuzzy integral

It is well recognized that the fuzzy measure plays a crucial role in fusion of multiple different classifiers using fuzzy integral. Many papers have focused on how to determine a fuzzy measure. Taking into account the intuitive idea that every classifier has different classification ability to the different class and the important role of the fuzzy integral in the process of information fusion, This work presents an optimization problem. By solving this optimization problem, the density function can be determined. Our study focuses on the Choquet fuzzy integral and the g-Lamda fuzzy measure. It shows that, in comparison with other fuzzy integrals such as Sugeno integral, the Choquet fuzzy integral and the corresponding g-Lamda fuzzy measure have the better performance for the system classification accuracy.

A comparative study of four fuzzy integrals for classifier fusion

Fuzzy Integral is widely accepted and applied in multi-classifier fusion to express the importance of individual classifiers and the interaction among classifiers. In this fusion model, there are two keys to determine. One is determining the fuzzy measure. Many researchers have done much work and proposed many types of fuzzy measure and methods to determine fuzzy measures. Another is selecting from four types of fuzzy integral: Sugeno integral, Choquet integral, upper integral and lower integral. Usually, the type of fuzzy integral is specified in advance. Choquet integral is often the choice. This paper is to compare comprehensively four fuzzy integrals in multiple-classifier fusion and hope to give the foundation for selecting Choquet integral. According the theoretical and experimental analysis, it gives the conclusion that Choquet integral is the best suitable for classifier fusion.

A parallel genetic algorithm for solving the inverse problem of support vector machines

Support Vector Machines (SVMs) are learning machines that can perform binary classification (pattern recognition) and real valued function approximation (regression estimation) tasks. An inverse problem of SVMs is how to split a given dataset into two clusters such that the maximum margin between the two clusters is attained. Here the margin is defined according to the separating hyper-plane generated by support vectors. This paper investigates the inverse problem of SVMs by designing a parallel genetic algorithm. Experiments show that this algorithm can greatly decrease time complexity by the use of parallel processing. This study on the inverse problem of SVMs is motivated by designing a heuristic algorithm for generating decision trees with high generalization capability.

Determine Fuzzy Measures in Multiple Classifiers Fusion Model

In finite set, Choquet fuzzy integral with respect to fuzzy measures can be transferred into linear combination of product, based on this fact we can choose standard optimization technical to determine fuzzy measures. This paper present linear programming and quadratic programming to determine fuzzy measures, the experiments demonstrate that classification accuracy of fuzzy integral with respect to fuzzy measure is better than the classification accuracies of majority voting and weighted average

A dynamic fuzzy measure for multiple classifier fusion

It has been shown that the fuzzy integral is an effective tool for the fusion of multiple classifiers. Of primary importance in the development of the system is the choice of the measure which embodies the importance of subsets of classifiers. In this paper we propose a method for a dynamic fuzzy measure which will change following the pattern to be classified (data dependent). This method uses the neural network which has good study ability. Our experiment results show that this method make the classification accurate improve.

Research on Chinese loanwords classification based on fuzzy set technology

Classification is necessary and basic to scientific research. The Chinese loanword has always been a hot spot of studies on Chinese linguistics, however the range of it remained unsettled. The paper will introduce fuzzy set technology into the discriminant process of Chinese loanwords to compose a reliable and efficient classifier. Simulations verify the efficiency and feasibility.

The inverse problem of support vector machines and its solution

Support vector machine (SVM) is novel type learning machine, based on statistical learning theory, which tasks involving classification, regression or novelty detection. This paper investigates an inverse problem of support vector machines (SVMs). The inverse problem is how to split a given dataset into two clusters such that the margin between the two clusters attains the maximum. Here the margin is defined according to the separating hyper-plane generated by support vectors. It is difficult to give an exact solution to this problem. In this paper, we design a genetic algorithm to solve this problem. Numerical simulations show the feasibility and effectiveness of this algorithm. This study on the inverse problem of SVMs is motivated by designing a heuristic algorithm for generating decision trees with high generalization capability.