Approximation by radial bases and neural networks

In this paper, we study approximation by radial basis functions including Gaussian, multiquadric, and thin plate spline functions, and derive order of approximation under certain conditions. Moreover, neural networks are also constructed by wavelet recovery formula and wavelet frames.     

Approximation by neural networks with sigmoidal functions

In this paper, we introduce a type of approximation operators of neural networks with sigmodal functions on compact intervals, and obtain the pointwise and uniform estimates of the ap- proximation. To improve the approximation rate, we further introduce a type of combinations of neurM networks. Moreover, we show that the derivatives of functions can also be simultaneously approximated by the derivatives of the combinations. We also apply our method to construct approximation operators of neural networks with sigmodal functions on infinite intervals.     

Approximation by series of sigmoidal functions with applications to neural networks

Annali di Matematica Pura ed Applicata (1923 -) - In this paper, we develop a constructive theory for approximating absolutely continuous functions by series of certain sigmoidal functions....     

Approximation on attraction domain of Cohen-Grossberg neural networks

In this paper, approximations of attraction domains of the asymptotically stable equilibrium points of some typical Cohen-Grossberg neural networks are achieved. Most Cohen-Grossberg neural networks are highly nonlinear systems which makes it difficult to approximate their attraction domain. Under some weak assumptions, we are allowed to employ the optimal Lyapunov method to obtain a Lyapunov function for asymptotically stable equilibrium points of a given Cohen-Grossberg neural network. With the help of this Lyapunov function, we approximate the corresponding attraction domain by the iterative expansion approach. Numerical simulations also illustrate that the approximation obtained is really part of the attraction domain.     

Approximation by ridge functions and neural networks with a bounded number of neurons

In this paper, we consider the problem of approximation of continuous multivariate functions by neural networks with a bounded number of neurons in hidden layers. We prove the existence of single-hidden-layer networks with bounded number of neurons, which have approximation capabilities not worse than those of networks with arbitrarily many neurons. Our analysis is based on the properties of ridge functions.     

Approximation of level continuous fuzzy-valued functions by multilayer regular fuzzy neural networks

This paper proves the capability of approximation by four-layer regular fuzzy neural networks on the set of all level continuous fuzzy-valued functions according to level convergence. A numerical example is given to illustrate the results.     

多层前向正则模糊神经网络的逼近能力

Abstract. Four-layer feedforward regular fuzzy neural networks are constructed. Universal ap-proximations to some continuous fuzzy functions defined on (R)“ by the four-layer fuzzyneural networks are shown. At first,multivariate Bernstein polynomials associated with fuzzyvalued functions are empolyed to approximate continuous fuzzy valued functions defined on eachcompact set of R“. Secondly,by introducing cut-preserving fuzzy mapping,the equivalent condi-tions for continuous fuzzy functions that can be arbitrarily closely approximated by regular fuzzyneural networks are shown. Finally a few of sufficient and necessary conditions for characteriz-ing approximation capabilities of regular fuzzy neural networks are obtained. And some concretefuzzy functions demonstrate our conclusions.     

Approximation by neural networks with a bounded number of nodes at each level

It is shown that the general approximation property of feed-forward multilayer perceptron networks can be achieved in networks where the number of nodes in each layer is bounded, but the number of layers grows to infinity. This is the case provided the node function is twice continuously differentiable and not linear.     

Max—min神经网络的一种有效学习算法

本文在Ａ．Ｂｌａｎｃｏ等人的算法的基础上，提出了ｍａｘ－ｍｉｎ神经网络的一种改进了的反馈学习算法，严格证明了该算法的迭代收敛性，理论分析及实例计算结果均表明，本文算法具有算法简单，收敛速度快，输出误差小等显著特点。     

Tidal level forecasting using functional and sequential learning neural networks

Prediction of tides is very much essential for human activities and to reduce the construction cost in marine environment. This paper presents two methods (1) an application of the functional networks (FN) and (2) sequential learning neural network (SLNN) procedures for the accurate prediction of tides using very short-term observation. This functional network model predicts the time series data of hourly tides directly while using an efficient learning process by minimizing the error based on the observed data for 30 days. Using the functional network, a very simple equation in the form of finite difference equation using the tidal levels at two previous time steps is arrived at. Sequential learning neural network uses one hidden neuron to predict the current tidal level using the previous four levels quite accurately. Hourly tidal data measured at Taichung harbor and Mirtuor coast along the Taiwan coastal region have been used for testing the functional network and sequential neural network model. Results show that the hourly data on tides for even a month can be predicted efficiently with a very high correlation coefficient.     

Large deviations and mean-field theory for asymmetric random recurrent neural networks

 In this article, we study the asymptotic dynamics of a noisy discrete time neural network, with random asymmetric couplings and thresholds. More precisely, we focus our interest on the limit behaviour of the network when its size grows to infinity with bounded time. In the case of gaussian connection weights, we use the same techniques as Ben Arous and Guionnet (see [3]) to prove that the image law of the distribution of the neurons' activation states by the empirical measure satisfies a temperature free large deviation principle. Moreover, we prove that if the connection weights satisfy a general condition of domination by gaussian tails, then the distribution of the activation potential of each neuron converges weakly towards an explicit gaussian law, the characteristics of which are contained in the mean-field equations stated by Cessac-Doyon-Quoy-Samuelides (see [4–6]). Furthermore, under this hypothesis, we obtain a law of large numbers and a propagation of chaos result. Finally, we show that many classical distributions on the couplings fulfill our general condition. Thus, this paper provides rigorous mean-field results for a large class of neural networks which is currently investigated in neural network literature. Received: 10 January 2000 / Revised version: 15 June 2001 / Published online: 13 May 2002     

Synchronization of heteroclinic circuits through learning in coupled neural networks

The synchronization of oscillatory activity in neural networks is usually implemented by coupling the state variables describing neuronal dynamics. Here we study another, but complementary mechanism based on a learning process with memory. A driver network, acting as a teacher, exhibits winner-less competition (WLC) dynamics, while a driven network, a learner, tunes its internal couplings according to the oscillations observed in the teacher. We show that under appropriate training the learner can “copy” the coupling structure and thus synchronize oscillations with the teacher. The replication of the WLC dynamics occurs for intermediate memory lengths only, consequently, the learner network exhibits a phenomenon of learning resonance.     

Approximation theory and optimization problems

In the first part of the article (Secs. 1, 2), a short historical survey is given of the development of analyses in the approximation theory of functions, distinguishing the most important stages and the basic papers, which stimulated investigations at each stage. In the second part (Secs. 3, 4), the basic aspects of the contemporary state of approximation theory and some tendencies of its further development are illuminated, and new statements of problems are formulated, connected with the optimization of approximation methods.Translated from Ukrainskii Matematicheskii Zhurnal, Vol. 42, No. 5, pp. 579–593, May, 1990.     

Constructive approximate interpolation by neural networks

We present a type of single-hidden layer feedforward neural networks with sigmoidal nondecreasing activation function. We call them ai-nets. They can approximately interpolate, with arbitrary precision, any set of distinct data in one or several dimensions. They can uniformly approximate any continuous function of one variable and can be used for constructing uniform approximants of continuous functions of several variables. All these capabilities are based on a closed expression of the networks.     

Classification by ensembles of neural networks

We introduce a new procedure for training of artificial neural networks by using the approximation of an objective function by arithmetic mean of an ensemble of selected randomly generated neural networks, and apply this procedure to the classification (or pattern recognition) problem. This approach differs from the standard one based on the optimization theory. In particular, any neural network from the mentioned ensemble may not be an approximation of the objective function.     

Pattern memory analysis based on stability theory of cellular neural networks

In this paper, several sufficient conditions are obtained to guarantee that the n-dimensional cellular neural network can have even (?2ⁿ) memory patterns. In addition, the estimations of attractive domain of such stable memory patterns are obtained. These conditions, which can be directly derived from the parameters of the neural networks, are easily verified. A new design procedure for cellular neural networks is developed based on stability theory (rather than the well-known perceptron training algorithm), and the convergence in the new design procedure is guaranteed by the obtained local stability theorems. Finally, the validity and performance of the obtained results are illustrated by two examples.     

Structural optimization by wavelet transforms and neural networks

An efficient methodology is presented to achieve optimal design of structures for earthquake loading. In this methodology a combination of wavelet transforms, neural networks and evolutionary algorithms are employed. The stochastic nature of the evolutionary algorithms makes the slow convergence. Specially, when earthquake induced loads are taken into account. To reduce the computational burden, a discrete wavelet transform is used by means of which the number of points in the earthquake record is decreased. Then, by using a surrogate model, the dynamic responses of the structures are predicted. In order to investigate the efficiency of the proposed methodology, two structures are designed for optimal weight. The numerical results demonstrate the computational advantages of the proposed hybrid methodology to optimal dynamic design of structures.