Supervised and unsupervised learning in radial basis functionclassifiers

The paper considers a number of strategies for training radial basis function (RBF) classifiers. A benchmark problem is constructed using ten-dimensional input patterns which have to be classified into one of three classes. The RBF networks are trained using a two-phase approach (unsupervised clustering for the first layer followed by supervised learning for the second layer), error backpropagation (supervised learning for both layers) and a hybrid approach. It is shown that RBF classifiers trained with error backpropagation give results almost identical to those obtained with a multilayer perceptron. Although networks trained with the two-phase approach give slightly worse classification results, it is argued that the hidden-layer representation of such networks is much more powerful, especially if it is encoded in the form of a Gaussian mixture model. During training, the number of subclusters present within the training database can be estimated: during testing, the activities in the hidden layer of the classification network can be used to assess the novelty of input patterns and thereby help to validate network outputs     

Environment-adaptation mobile radio propagation prediction usingradial basis function neural networks

This paper investigates the application of a radial basis function (RBF) neural network to the prediction of field strength based on topographical and morphographical data. The RBF neural network is a two-layer localized receptive field network whose output nodes from a combination of radial activation functions computed by the hidden layer nodes. Appropriate centers and connection weights in the RBF network lead to a network that is capable of forming the best approximation to any continuous nonlinear mapping up to an arbitrary resolution. Such an approximation introduces best nonlinear approximation capability into the prediction model in order to accurately predict propagation loss over an arbitrary environment based on adaptive learning from measurement data. The adaptive learning employs hybrid competitive and recursive least squares algorithms. The unsupervised competitive algorithm adjusts the centers while the recursive least squares (RLS) algorithm estimates the connection weights. Because these two learning rules are both linear, rapid convergence is guaranteed. This hybrid algorithm significantly enhances the real-time or adaptive capability of the RBF-based prediction model. The applications to Okumura's (1968) data are included to demonstrate the effectiveness of the RBF neural network approach     

基于自适应RBFN阵列的说话人识别

提出一种可用于说话人识别的自适应RBFN阵列。RBF网设计的核心在于确定网络中心的数目及位置，该自适应算法有效地融合了IOC与ROLS算法的优点，不仅能动态调节RBF网的隐节点数，还能使网络的数据中心自适应变化，很好地优化了网络的结构。用与文本无关的闭集说话人识别系统对该算法进行了验证，实验结果表明，该方法与传统的RBF算法相比，自适应RBF网具有较好的鲁棒性以及精简的网络结构等优点。     

Multigradient: a new neural network learning algorithm for patternclassification

The authors propose a new learning algorithm for multilayer feedforward neural networks, which converges faster and achieves a better classification accuracy than the conventional backpropagation learning algorithm for pattern classification. In the conventional backpropagation learning algorithm, weights are adjusted to reduce the error or cost function that reflects the differences between the computed and the desired outputs. In the proposed learning algorithm, the authors view each term of the output layer as a function of weights and adjust the weights directly so that the output neurons produce the desired outputs. Experiments with remotely sensed data show the proposed algorithm consistently performs better than the conventional backpropagation learning algorithm in terms of classification accuracy and convergence speed     

RBF神经网络在视频业务建模及预测中的应用

ＡＴＭ技术在关键在于业务控制,能否有效地实施业务控制则取决于对业务特征的了解和预测能力。传统的解析方法对视频业务进行预测的局限性较为明显。本文采用径向基函数神经网络对视频业务进行建模和预测,并提出了分别采用ＬＢＧ算法和Ｈｅｓｔｅｎｅｓ奇异值解进行隐含层神经元中心选择和输出神经腱权值计算的改进方法。     

An improved scheme for direct adaptive control of dynamical systems using backpropagation neural networks

This paper presents an improved direct control architecture for the on-line learning control of dynamical systems using backpropagation neural networks. The proposed architecture is compared with the other direct control schemes. In this scheme the neural network interconnection strengths are updated based on the output error of the dynamical system directly, rather than using a transformed version of the error employed in other schemes. The ill effects of the controlled dynamics on the on-line updating of the network weights are moderated by including a compensating gain layer. An error feedback is introduced to improve the dynamic response of the control system. Simulation studies are performed using the nonlinear dynamics of an underwater vehicle and the promising results support the effectiveness of the proposed scheme.     

基于自适应投影学习算法的RBF神经网络多用户检测方法

自适应投影学习算法是一种简单有效的构造和训练径向基函数神经网络的方法,该方法能迭代地确定径向基函数的个数,中心的位置以及网络的权系数。本文将基于自适应投影学习算法的径向基函数神经网络应用于CDMA系统多用户检测,仿真表明:这种方法对远近问题不敏感,具有良好的误码率性能和抗多址干扰性能。     

Input Resistance Calculation for Circular Microstrip Antennas Using Adaptive Neuro-Fuzzy Inference System

This paper presents a new method based on adaptive neuro-fuzzy inference system (ANFIS) to calculate the input resistance of circular microstrip patch antennas. The ANFIS is a fuzzy inference system (FIS) implemented in the framework of an adaptive fuzzy neural network. It combines the explicit knowledge representation of FIS with learning power of neural networks. A hybrid learning algorithm based on the least square approach and the backpropagation algorithm is used to optimize the parameters of ANFIS. The input resistance results predicted by ANFIS are in excellent agreement with the experimental results reported elsewhere.     

Learning techniques to train neural networks as a state selectorfor inverter-fed induction machines using direct torque control

Neural networks are receiving attention as controllers for many industrial applications. Although these networks eliminate the need for mathematical models, they require a lot of training to understand the model of a plant or a process. Issues such as learning speed, stability, and weight convergence remain as areas of research and comparison of many training algorithms. This paper discusses the application of neural networks to control induction machines using direct torque control (DTC). A neural network is used to emulate the state selector of the DTC. The training algorithms used in this paper are the backpropagation, adaptive neuron model, extended Kalman filter, and the parallel recursive prediction error. Computer simulations of the motor and neural-network system using the four approaches are presented and compared. Discussions about the parallel recursive prediction error and the extended Kalman filter algorithms as the most promising training techniques is presented, giving their advantages and disadvantages     

Adaptive feedback linearization control of antilock braking systems using neural networks

Antilock braking systems are designed to control the wheel slip, such that the braking force is maximized and steerability is maintained during braking. However, the control of antilock braking systems is a challenging problem due to nonlinear braking dynamics and the uncertain and time-varying nature of the parameters. This paper presents an adaptive neural network-based hybrid controller for antilock braking systems. The hybrid controller is based on the well-known feedback linearization, combined with two feedforward neural networks that are proposed so as to learn the nonlinearities of the antilock braking system associated with feedback linearization controller. The adaptation law is derived based on the structure of the controller, using steepest descent gradient approach and backpropagation algorithm to adjust the networks weights. The weight adaptation is online and the stability of the proposed controller in the sense of Lyapunov is studied. Simulations are conducted to show the effectiveness of the proposed controller under various road conditions and parameter uncertainties.     

一种基于AGA的RBF网络混合学习算法

基于目前RBF网络学习方法中的一些不足,提出了一种基于AGA的混合学习方法,即应用AGA对网络隐单元RBF个数和宽度σ同时优选,并将最佳隐单元数作为K-均值聚类数得到隐单元中心,隐层到输出层的权值由LS法确定。针对K-均值聚类算法对初始值敏感的问题,算法在最后阶段对其执行多次运算,由此选择最佳结果。仿真结果表明,该方法在大样本情况下,训练得到的网络在精度和结构上得到了良好的结合。     

A neural-network-based controller for a single-link flexiblemanipulator using the inverse dynamics approach

This paper presents an intelligent-based control strategy for tip position tracking control of a single-link flexible manipulator. Motivated by the well-known inverse dynamics control strategy for rigid-link manipulators, two feedforward neural networks (NNs) are proposed to learn the nonlinearities of the flexible arm associated with the inverse dynamics controller. The redefined output approach is used by feeding back this output to guarantee the minimum phase behavior of the resulting closed-loop system. No a priori knowledge about the nonlinearities of the system is needed and the payload mass is also assumed to be unknown. The network weights are adjusted using a modified online error backpropagation algorithm that is based on the propagation of output tracking error, derivative of that error and the tip deflection of the manipulator. The real-time controller is implemented on an experimental test bed. The results achieved by the proposed NN-based controller are compared experimentally with conventional proportional-plus-derivative-type and standard inverse dynamics controls to substantiate and verify the advantages of our proposed scheme and its promising potential in identification and control of nonlinear systems     

δ-NARMA neural networks: a new approach to signal prediction

This paper presents a new connectionist architecture for stochastic univariate signal prediction. After a review of related statistical and connectionist models pointing out their advantages and limitations, we introduce the ϵ-NARMA model as the simplest nonlinear extension of ARMA models. These models then provide the units of a MLP-like neural network: the δ-NARMA neural network. The associated learning algorithm is based on an extension of classical backpropagation and on the concept of virtual error. Such networks can be seen as an extension of ARIMA and ARARMA models and face the problem of nonstationary signal prediction. A theoretical study brings understanding of experimental phenomena observed during the δ-NARMA learning process. The experiments carried out on three railroad-related real-life signals suggest that δ-NARMA networks outperform other studied univariate models     

A fast new algorithm for training feedforward neural networks

A fast algorithm is presented for training multilayer perceptrons as an alternative to the backpropagation algorithm. The number of iterations required by the new algorithm to converge is less than 20% of what is required by the backpropagation algorithm. Also, it is less affected by the choice of initial weights and setup parameters. The algorithm uses a modified form of the backpropagation algorithm to minimize the mean-squared error between the desired and actual outputs with respect to the inputs to the nonlinearities. This is in contrast to the standard algorithm which minimizes the mean-squared error with respect to the weights. Error signals, generated by the modified backpropagation algorithm, are used to estimate the inputs to the nonlinearities, which along with the input vectors to the respective nodes, are used to produce an updated set of weights through a system of linear equations at each node. These systems of linear equations are solved using a Kalman filter at each layer     

基于人工免疫RBF神经网络的时间序列预测方法研究

研究了一种基于免疫识别原理的径向基函数神经网络学习算法,该算法将所识别的数据作为抗原,抗体为抗原的压缩映射并作为神经网络模型的隐层中心,采用最小二乘法确定权值,提高了RBF神经网络收敛速度和精度.将人工免疫RBF神经网络应用于时间序列预测中,实例仿真结果证明了算法的有效性和可行性,为时间序列预测提供了一种新途径.     

Machine learning based optimized live virtual machine migration over WAN links

Live virtual machine migration is one of the most promising features of data center virtualization technology. Numerous strategies have been proposed for live migration of virtual machines on local area networks. These strategies work perfectly in their respective domains with negligible downtime. However, these techniques are not suitable to handle live migration over wide area networks and results in significant downtime. In this paper we have proposed a Machine Learning based Downtime Optimization (MLDO) approach which is an adaptive live migration approach based on predictive mechanisms that reduces downtime during live migration over wide area networks for standard workloads. The main contribution of our work is to employ machine learning methods to reduce downtime. Machine learning methods are also used to introduce automated learning into the predictive model and adaptive threshold levels. We compare our proposed approach with existing strategies in terms of downtime observed during the migration process and have observed improvements in downtime of up to 15 %.     

Hardware Implementation of a Pulse Density Neural Network Using Simultaneous Perturbation Learning Rule

The choice of the learning scheme is very important in the implementation of neural networks to take advantage of their learning ability. Usually, the back-propagation method is widely used as a learning rule in neural networks. Since back-propagation requires so-called error back propagation to update weights, it is relatively difficult to realize hardware neural networks using the back-propagation method. In this paper, we present a pulse density neural network system with learning ability. As a learning rule, the simultaneous perturbation method is used. The learning rule requires only forward operations of networks to update weights instead of the error back-propaga- tion. Thus, we can construct the network system with learning ability without the need for a complicated circuit that calculates gradients of an error function. Pulse density is used to represent the basic quantities in this system. The pulse system has some attractive properties which includes robustness against a noisy environment. A combina- tion of the simultaneous perturbation learning rule and the pulse density system results in an interesting architec- ture of hardware neural systems. Results for the exclusive OR problem and a simple identity problem are shown.     

Three machine learning techniques for automatic determination of rules to control locomotion

Automatic prediction of gait events (e.g., heel contact, flat foot, initiation of the swing, etc.) and corresponding profiles of the activations of muscles is important for real-time control of locomotion. This paper presents three supervised machine learning (ML) techniques for prediction of the activation patterns of muscles and sensory data, based on the history of sensory data, for walking assisted by a functional electrical stimulation (FES). Those ML's are: 1) a multilayer perceptron with Levenberg-Marquardt modification of backpropagation learning algorithm; 2) an adaptive-network-based fuzzy inference system (ANFIS); and 3) a combination of an entropy minimization type of inductive learning (IL) technique and a radial basis function (RBF) type of artificial neural network with orthogonal least squares learning algorithm. Here we show the prediction of the activation of the knee flexor muscles and the knee joint angle for seven consecutive strides based on the history of the knee joint angle and the ground reaction forces. The data used for training and testing of ML's was obtained from a simulation of walking assisted with an FES system [39]. The ability of generating rules for an FES controller was selected as the most important criterion when comparing the ML's. Other criteria such as generalization of results, computational complexity, and learning rate were also considered. The minimal number of rules and the most explicit and comprehensible rules were obtained by ANFIS. The best generalization was obtained by the IL and RBF network.     

Neural-network-based predictive learning control of ram velocity in injection molding

In this paper, we develop a predictive learning controller for ram velocity of injection molding based on neural networks. We first introduce a model of describing the injection molding, including the time horizon and the batch index. The feedback control plus biased function is proposed for controlling this plant. More specifically, a radial basis function (RBF) network is used to approximate the biased function based on the time horizon. The weights in the RBF are determined by a predictive control scheme based on the batch index. For this algorithm, relevant convergence is investigated. Simulation results reveal that the proposed control can achieve our claims.     

基于递归正交最小二乘的径向基函数网络人脸识别

本文提出了一种基于递归正交最小二乘的径向基函数(RBF)网络人脸识别算法,该算法首先使用主成分分析(PCA)提取输入图像特征,将提取的特征作为RBF网络的输入进行识别,在求取网络权值时采用递归正交最小二乘(ROLS)算法。实验表明,该算法能明显地缩短训练时间同时具有较高的识别率。