Jinlong An

An Improved LSSVM Regression Algorithm

Support Vector Machine (SVM) is a new and valid machine-learning algorithm developed on statistical learning theory, and it has been used for classification, function regression, and time series prediction. Recently an extension of traditional SVM named LSSVM has been introduced. Compared with the Support Vector Machine, the Least Squares Support Vector Machine (LSSVM) lose the sparseness, which would influence the efficiency of relearning. To conclude a sparse solution, in this paper we present an improved algorithm for Least Squares Support Vector Machine-XS-LSSVM, and prove its effect by an simulation experiment.

A SVM function approximation approach with good performances in interpolation and extrapolation

Function approximation estimation and prediction are used widely in many fields such as control and signal processing. The merit and shortcoming of existing methods of function approximation and regression are analyzed, and a new function approximation and regression approach which is based on the combination of SVMs (support vector machines) is presented. The new approach fully exerts the merit of SVM, and overcomes the shortcoming in extrapolation of function approximation and regression. The experiment demonstrates that the new approach improves the precision of SVM function approximation greatly in both interpolation and extrapolation.

EEG source localization of ERP based on multidimensional support vector regression approach

A new integrated multi-method system is presented to estimate the location and moment of equivalent current dipole sources of event-related potentials (ERP). In order to handle the large-scale high dimension problems efficiently and quickly, the ISOMAP algorithm was used to find the low dimensional manifolds from recorded EEG. Then, based on reduced dimension data, multidimensional support vector regression (MSVR) with similar iterative re-weight least square (IRWLS) was used to discover the relationship between the observation potentials on the scalp and the internal sources within the brain. In our experiments, the two current dipole sources with four-shell concentric sphere model were reconstructed. Our experiments demonstrate that MSVR based on the support vector machine can obtain more robust estimations for EEG source localization problem.

Study on SVM On-Line Function Regression Method for Mass Data

In order to overcome the problems that the SVM training time is too long for a large number of samples and that SVM cannot be trained online when the samples increase dynamically, a new approach of SVM online function regression for mass samples is put forward in this paper. And the validity of this method is proved by simulation experiment.

Study on the Decision-Making Technique Based on Fuzzy Support Vector Machine

At present, classification method based on support vector machine can not be used as a decision-making technique yet, but an auxiliary tool for decision-making in practice because the forecasting correct rate of which is difficult to reach as high as one hundred percent. At the same time, SVM method is very sensitive to noise and outlier points. In order to overcome the disadvantages above, a fuzzy support vector machine method for decision-making is presented. By giving different fuzzy factor to different classification samples in SVM, the decision-making hyperplane of each class samples can be obtained, on which the decision can be made in practical system of decision. The simulation demonstrates the effectiveness of this approach

Study on the solving method of electromagnetic field forward problem based on support vector machine

The solving methods of high-order ordinary differential equations and partial differential equations with various boundary conditions are introduced. A new method based on support vector machine is proposed to solve high-order ordinary differential equations and partial differential equations with various boundary conditions. The construction method of approximate solution function is given in detail. And the validity of the new solving method which is based on support vector machine to solve electromagnetic field forward problem is verified.

Study on the method of fault diagnosis in analog circuits based on new multi-class SVM

Fault diagnosis in analog circuits is a comparatively front research topic. Firstly, the characteristics and the difficulties of fault diagnosis in analog circuits are introduced in this paper. Secondly, to overcome the defections of existing methods of SVM multiclass classification, a new method of SVM multiclass classification based on binary tree is presented. Aiming at the characteristics of fault diagnosis with finite samples and the difficulties of traditional mode identifying method based on gradual-close theory faces in fault pattern classifier, we used our new method of SVM multiclass classification to fault diagnosis of analog circuits. Finally, we also simulate on the fault diagnosis examples with the same training and test samples, and compare the results with that of neural networks method. The simulation results show the new method is efficient.

The Method of Multidimensional Support Vector Regression for Moving Dipole Localization of Face Expression

Brain signal source localization is a process of inverse calculation from electroencephalogram (EEG) signal. A new method of Multidimensional Support Vector Regression (MSVR) with similar iterative re-weight least square (IRWLS) is firstly used in source localization of face expression. In order to discover the relationship between sensor information and internal source in the brain, the moving dipole with four-shell concentric sphere model was reconstructed. Its location parameters and components were fitted in a series of time points. EEG signals of face expression were adopted in our experiments. Satisfactory results demonstrate that MSVR based on the support vector machine can obtain more robust estimations for EEG inverse problem.

Study of Increasing Computational Efficiency in Element-Free Galerkin Method

The element-free Galerkin method (EFGM) is based on moving least square (MLS) approximations and uses only a set of nodal points to formulate the discrete model. The EFGM does not require any elements and has not the element deformation problems. Therefore, the EFGM simplifies pre-processing and post-processing and can solve problems with higher accuracy. However, the computational efficiency of the EFGM is very low, which caused mainly by the Gaussian integration and finding the inverse matrix of the coefficient matrix. To overcome this shortcoming, the orthogonal basic functions are given, the preconditioning method of equations and the support domain of weight functions are studied thoroughly in this paper