Earticle

A novel neural network algorithm optimized by particle swarm optimization (PSO) for function approximation is proposed in this paper. The prior information extracted from the upper and lower bound of the approximated function is coupled into PSO. Since the prior information narrows the search space and guides the movement direction of the particles, the convergence rate and the approximation accuracy are improved. Experimental results demonstrate that the new algorithm is more effective than traditional methods.

For system analysis and design purposes, it is meaningful discussion to define control system pole status whether a pole is significant or not. If a pole is less important, it can be canceled from the transfer function of system and system order is reduced. In this paper a method for system order reduction of transfer function using Rationalized Haar functions based on approximation and transform algorithm is presented. The Haar function set forms a complete set of orthogonal rectangular functions such as Walsh and block pulse functions. But the Haar functions have some disadvantages of calculation because of including irrational numbers such as ± sp, (p = 1, 2, ..). The Rationalized Haar functions were introduced by M. Ohkita to overcome these disadvantages. The Rationalized Haar functions constitute of rational numbers only. The applied method to solve the system order reduction of transfer function problem is superior to conventional numerical methods.

Since introduction, the Support Vector Machines (SVM) has been popularly used in machine learning and data mining tasks due to their strong mathematical background and promising result. Nevertheless, they are noticeably slow in the prediction stage. The speed is influenced by number of support vectors determined in the training phase. Motivated by this fact, several studies are done to reduce the number of support vectors. The reduction should consider the degeneration of learning quality and preserve it at much as possible. Most previous methodologies either reduce the training set or apply a post-processing step to reduce the number of support vectors. In this paper, we proposed a new SVM cost function called Step Regularized Support Vector Machine (SRSVM), which is a standard SVM with extra constrained to reduce the number of support vectors, which can be defined by user. Experimental results are done to evaluate the efficiency and speed of proposed algorithm. SRSVM are also compared to other related SVM algorithms. The comparisons showed that the proposed method is effective in reducing number of support vectors while preserving the high performance of the classifier.

In allusion to such problems in the traditional face recognition methods as poor recognition accuracy and dissatisfactory processing effect for directivity and anisotropic characteristic in face data, lasso regularized Gabor shearlet face multivariate sparse function approximation algorithm is proposed in this article. Firstly, Gabor improved shearlet algorithm is adopted at the level of the face-image biological signals for the sparse expansion representation of the face data characteristics, and meanwhile this algorithm is also adopted to extract the geometrical characteristics of the expansion face with directivity and anisotropic characteristic. Secondly, in order to balance the algorithm effect, lasso regularization theory is introduced therein to control and weigh the relation between the fidelity and the smoothness of the face data. Finally, the corresponding simulation experiment is carried out to compare the proposed algorithm and the existing algorithms in the standard test database in order to verify the advantages of the proposed algorithm in the aspect of face recognition accuracy and efficiency.

오래 전부터 연료의 가격은 상승하고 있다. 제조업체는 보증을 통해 실용적인 대안을 찾고자 전기와 강력한 바이오 연료를 이용하여 차량의 성장가능을 연구하고 있다. 이제, 이러한 녹색 환경(emission) 관련된 보증은 보증기간이 확장되며, 이러한 보증을 '수퍼 보증" 이라 불린다. 본 논문의 주요 결과는 라돈 변환의 역행렬을 보증공간의 수치를 줄이기 위해 사용되며, 응용 프로그램 및 RBF 네트워크를 사용하여 대략적인 이변량의 보증 기능에 새로운 방법을 제시한다. 이 방법은 다음과 같은 단계로 구성되어 있다. 첫째, 라돈 변환을 이용하여, 이변량 보증 함수의 1 차원 함수를 줄일 수 있다. 둘째, 1 차원 함수의 각 신경 서브 네트워크와 신경 네트워크 기법을 사용하여 근사할 수 있다. 셋째, 이러한 신경 sub-networks 형태로 최종 근사 신경망 함께 결합 된다. 넷째, 라 돈 변환의 역함수 값을 사용 하여 최종 근사 신경 네트워크에 우리가 주어진 함수 근사화를 얻을 수 있다. 또한, 우리는 자동차 회사의 일부 그린 보증 데이터를 가지고 위의 방법을 적용한다.

As the price of traditional fuels soar, the alternatives are becoming more viable. And manufacturers are promoting the growing viability of electric and biofuel-powered vehicles through longer warranties. Now, these longer green environment (emission)warranties, sometimes called extended warranties or “super warranties,” have been adapted. The main result of this paper is to present a new method to approximate a bivariate warranty function by using Radial Basis Function Network with application of Radon Transform and its inverse which is used to reduce the dimension of the warranty space. This method consist of the following stages: First, by using the Radon Transform, the bivariate warranty function can be reduced to one dimensional function. Second, each of the one dimensional functions is approximated by using neural network technique into neural sub-networks. Third, these neural sub-networks are combined together to form the final approximation neural network. Four, by using the inverse of radon transform to this final approximation neural network we get the approximation to the given function. Also, we apply the above method to some green warranty data of automotive vehicle company.

We present a relation between the orthogonality of the constrained Legendre polynomials over the triangular domain and the BB ($B{\acute{e}zier}\;-Bernstein$) coefficients of the polynomials using the equivalence of orthogonal complements. Using it we also show that the best constrained degree reduction of polynomials in BB form equals the best approximation of weighted Euclidean norm of coefficients of given polynomial in BB form from the coefficients of polynomials of lower degree in BB form.

This paper presents function approximation based on nonparametric estimation. As an estimation model of function approximation, a three layered network composed of input, hidden and output layers is considered. The input and output layers have linear activation units while the hidden layer has nonlinear activation units or kernel functions which have the characteristics of bounds and locality. Using this type of network, a many-to-one function is synthesized over the domain of the input space by a number of kernel functions. In this network, we have to estimate the necessary number of kernel functions as well as the parameters associated with kernel functions. For this purpose, a new method of parameter estimation in which linear learning rule is applied between hidden and output layers while nonlinear (piecewise-linear) learning rule is applied between input and hidden layers, is considered. The linear learning rule updates the output weights between hidden and output layers based on the Linear Minimization of Mean Square Error (LMMSE) sense in the space of kernel functions while the nonlinear learning rule updates the parameters of kernel functions based on the gradient of the actual output of network with respect to the parameters (especially, the shape) of kernel functions. This approach of parameter adaptation provides near optimal values of the parameters associated with kernel functions in the sense of minimizing mean square error. As a result, the suggested nonparametric estimation provides an efficient way of function approximation from the view point of the number of kernel functions as well as learning speed.

Background: Histone deacetylase (HDAC) 8 is one of its family members catalyzes the removal of acetyl groups from N-terminal lysine residues of histone proteins thereby restricts transcription factors from being expressed. Inhibition of HDAC8 has become an emerging and effective anti-cancer therapy for various cancers. Application computational methodologies may result in identifying the key components that can be used in developing future potent HDAC8 inhibitors. Results: Facilitating the discovery of novel and potential chemical scaffolds as starting points in the future HDAC8 inhibitor design, quantitative structure-activity relationship models were generated with 30 training set compounds using genetic function approximation (GFA) and Bayesian algorithms. Six GFA models were selected based on the significant statistical parameters calculated during model development. A Bayesian model using fingerprints was developed with a receiver operating characteristic curve cross-validation value of 0.902. An external test set of 54 diverse compounds was used in validating the models. Conclusions: Finally two out of six models based on their predictive ability over the test set compounds were selected as final GFA models. The Bayesian model has displayed a high classifying ability with the same test set compounds and the positively and negatively contributing molecular fingerprints were also unveiled by the model. The effectively contributing physicochemical properties and molecular fingerprints from a set of known HDAC8 inhibitors were identified and can be used in designing future HDAC8 inhibitors.

In this paper, a fuzzy-neural interpolating network is proposed to efficiently approximate a nonlinear function. Specifically, basis functions are first constructed by Fuzzy Membership Function based Neural Networks (FMFNN). And the fuzzy similarity, which is defined as the degree of matching between actual output value and the output of each basis function, is employed to determine initial weighting of the proposed network. Then the weightings are updated in such a way that square of the error is minimized. To show the capability of function approximation of the proposed fuzzy-neural interpolating network, a numerical example is illustrated.

A lot of researches have been conducted to estimate the volatility smile effect shown in the option market. This paper proposes a method to approximate an implied volatility function, given noisy real market option data. To construct an implied volatility function, we use Gaussian Processes (GPs). Their output values are implied volatilities while moneyness values (the ratios of strike price to underlying asset price) and time to maturities are as their input values. To show the performances of our proposed method, we conduct experimental simulations with Korean Equity-Linked Warrant (ELW) market data as well as toy data.

In this paper, genetic algorithm (GA) is the technique to search for the optimal structures (i,e., the kind of neural network, the number of hidden neuron, ..) of the neural networks which are used approximating a given nonlinear function, In this paper, we used multi layer feed-forward neural network. The decision method of synapse weights of each neuron in each generation used back-propagation method. In this study, we simulated nonlinear function approximation in the temperature control system.

This paper describes novel methods to construct fuzzy inference rules with gradient descent. The present methods have a constructive mechanism of the rule unit that is applicable in two parameters: the central value and the width of the membership function in the antecedent part. The first approach is to create the rule unit at the nearest position from the input space, for the central value of the membership function in the antecedent part. The second is to create the rule unit which has the minimum width, for the width of the membership function in the antecedent part. Experimental results are presented in order to show that the proposed methods are effective in difference on the inference error and the number of learning iterations.

Based on an introduced optimal trajectory planning method, this paper mainly deals with the accuracy analysis during the function approximation process of the optimal trajectory planning method. The basis functions are composed of Hermit polynomials and Fourier series to improve the approximation accuracy. Since the approximation accuracy is affected by the given orders of each basis function, the accuracy of the optimal solution is examined by changing the combinations of the orders of Hermit polynomials and Fourier series as the approximation basis functions. As a result, it is found that the proper approximation basis functions are the $5^{th}$ order Hermit polynomials and the $7^{th}-10^{th}$ order of Fourier series.

This paper presents observer-based controller design for interval type-2 fuzzy system with staircase membership approximation. In type-2 fuzzy case, membership function is itself fuzzy set itself. Thus, type-2 fuzzy system can deal with parametric uncertainties of nonlinear system by capturing the uncertainties in membership function. Likewise, stabilization condition of type-2 fuzzy system is derived from quadratic Lyapunov function, and it goes to linear matrix inequality. Furthermore, in this paper, to relax the conservativeness of stabilization condition, staircase membership function approximating method is applied. Observer-based control method is adopted to control system which has some unmeasurable states. To prove suitability of our proposed method, numerical example is presented.

We study the neural network approximation to a function in $C^1$[0, 1] and its derivative. In [3], we used even trigonometric polynomials in order to get an approximation order to a function in $L_p$ space. In this paper, we show the simultaneous approximation order to a function in $C^1$[0, 1] using a Bernstein polynomial and a feedforward neural network. Our proofs are constructive.

This paper applies approximation of the renewal function for Hjorth model and Dhillon model which show the trend change in its aging properties. We obtain the renewal function for Hjorth model and Dhillon model by a numerical solution of an approximate integral. We observe the influence of each parameter in these models. The results of the computation are described and their corresponding graphs are provided.

We investigate the approximation order to a function in $L_p$[-1, 1] for $0{\leq}p<{\infty}$ by generalized translation networks. In most papers related to neural network approximation, sigmoidal functions are adapted as an activation function. In our research, we choose an infinitely many times continuously differentiable function as an activation function. Using the integral modulus of continuity and the divided difference formula, we get the approximation order to a function in $L_p$[-1, 1].

We obtain the approximation order to a smooth function on a compact subset of $\mathbb{R}$ by generalized translation networks. In our study, the activation function is infinitely many times continuously differentiable function but it does not have special properties around ${\infty}$ and $-{\infty}$ like a sigmoidal activation function. Using the Jackson's Theorem, we get the approximation order. Especially, we obtain the approximation order by a neural network with a fixed threshold.

In this paper, we actually construct the simultaneous approximation by neural networks to a differentiable function. To do this, we first construct a polynomial approximation using the Fejer sum and then a simultaneous neural network approximation with the squashing activation function. We also give numerical results to support our theory.