检索结果-内蒙古大学图书馆

A New Learning Algorithm for Recurrent Networks

IFAC Proceedings Volumes 1995年第24期28卷 323-328页

作者： Daniel Volovici Dept. of Electrical Engineering and Computer Science University of Sibiu str. Revolutiei no. 4 2400-Sibiu Romania

The most popular methods for modifying feedforward and recurrent neural networks’ weights, accordingly the backpropagation methods, actually are gradient methods. Due to die fact that the gradient is a local measure, the step towards the gradient’s direction, which step is given by the learning rate p , must be infinitesimal, which implies the choosing of a very small p . But this leads to a very slow convergence. Hence, still a large p is chosen. On the other hand, a too large p leads to strong oscillations of the aim function. Moreover, the values of p are relative to the problem to be solved.

关键词： neural networks backpropagation algorithms gradient methods learning algorithms least squares

来源：评论

学校读者我要写书评

暂无评论

On initialization of neural network

On initialization of neural network

引用

International Conference on Neural Networks

作者： P. Costa-Hirschauer P. Larzabal H. Clergeot LESiR-ENS CNRS Cachan France URA CNRS LESiR-ENS Cachan Cachan France LTSMM-UAG French Guiana France

The paper concerns the problem of initialization in neural networks. We focus on backpropagation networks with one hidden layer. The initialization of the weights is crucial: if the network is incorrectly initialized, it converges to local minima. So, the classical random initialization appears as a very bad solution. If we consider the problem's specificity and the non linearity of sigmoids, improvements can be very significant. We propose a new initialization scheme based on the search for an explicit solution to the problem. We study two cases with different hypotheses. Simulation results are presented showing that these original initializations allow us to avoid local minima, to reduce training time, to obtain a better generalization and to estimate the network size.

关键词： Neural networks Function approximation Linearity Linear approximation Neurons Electronic mail backpropagation algorithms Linear systems Parameter estimation Nonlinear equations

来源：评论

学校读者我要写书评

暂无评论

On the effect of the nonlinearity of the sigmoid function in artificial neural network classifiers

On the effect of the nonlinearity of the sigmoid function in...

引用

International Conference on Neural Networks

作者： S. Uchimura Y. Hamamoto S. Tomita Oshima Nat. Coll. of Maritime Technol. Japan Faculty of Engineering Yamaguchi University Ube Japan

Despite a considerable amount of recent directed towards pattern recognition applications of the artificial neural networks (ANNs), little quantitative information concerning the nonlinearity of the sigmoid function is available to a designer. In this paper, the authors study the effect of the nonlinearity of the sigmoid function on the generalization capability of the ANN classifiers trained with the backpropagation algorithm. Experimental results show that like the case of the network size, there exists the optimal degree of the nonlinearity for a particular problem.

关键词： Artificial neural networks Pattern recognition backpropagation algorithms

来源：评论

学校读者我要写书评

暂无评论

Utilization of hierarchical structure stochastic automata for the back propagation method with momentum

Utilization of hierarchical structure stochastic automata fo...

引用

International Conference on Neural Networks

作者： N. Baba H. Handa Information Science Osaka Kyoiku University Kashihara Osaka Japan

backpropagation (BP) is one of the most popular learning algorithms for multilayer networks, but it has limitations. Various modified BP methods have therefore been proposed. The BP method with momentum may be one of the most popular such modified algorithms. It has been reported that the BP method with momentum has been applied quite successfully to many practical problems. However, despite its effectiveness, this method involves the following serious problem: "Its learning performance depends heavily upon the selection of the value of momentum factor." Unfortunately, it seems that there has not so far been proposed an intelligent algorithm for determining an appropriate value of the momentum factor. In this paper, we suggest that hierarchical structure stochastic automata are quite helpful for finding an appropriate value of the momentum factor of the BP method with momentum. Several computer simulation results confirm our idea.

关键词： Stochastic processes Learning automata backpropagation algorithms Nonhomogeneous media Computer simulation

来源：评论

学校读者我要写书评

暂无评论

A sparse matrix approach to neural network training

A sparse matrix approach to neural network training

引用

International Conference on Neural Networks

作者： Fang Wang Q.J. Zhang Electronics Department Carleton University Ottawa ONT Canada

A new training technique, based on sparse matrix concept is developed for the training of multilayer perceptron. The proposed approach exploits the patterns of neuron activations in neural networks and substantially reduces the amount of computations in backpropagation. The proposed training algorithm is applied to word recognition with TI20 real speech data. Compared to techniques without using the sparse concept, same or better recognition accuracy is achieved and training speed is substantially improved.

关键词： Sparse matrices Neural networks Neurons backpropagation algorithms Multilayer perceptrons Speech recognition Computer networks Multi-layer neural network Standards development Feedforward neural networks

来源：评论

学校读者我要写书评

暂无评论

A learning strategy for multilayer neural network using discretized Sigmoidal function

A learning strategy for multilayer neural network using disc...

引用

International Conference on Neural Networks

作者： S. Anna Durai P.V. Siva Prasad A. Balasubramaniam V. Ganapathy Anna University Chennai Chennai Tamil Nadu IN Sch. of Comput. Sci. & Eng. Anna Univ. Madras India

In this paper a new approach for training the multilayer neural network (MNN), is proposed based on the scheme of discretization of the sigmoidal threshold activation function at regular intervals. The discretized values are stored in the form of a look up table (LUT). The weights are updated on a layer by layer basis using the values in the LUT. The proposed algorithm is applied to pattern classification problems and tested experimentally. It is found that the proposed algorithm takes lesser training time than the conventional backpropagation algorithm with continuous sigmoidal function. The proposed algorithm is more suitable for hardware implementation.

关键词： Multi-layer neural network Neural networks Table lookup Computer science Signal processing algorithms backpropagation algorithms Error correction Equations Neurons Testing

来源：评论

学校读者我要写书评

暂无评论

A self-organizing system for the development of neural network parameter estimators

A self-organizing system for the development of neural netwo...

引用

IEEE Workshop on Neural Networks for Signal Processing

作者： M.T. Manry Department of Electrical Engineering University of Texas Arlington Arlington TX USA

The design an optimal neural network estimator from training data is difficult because: 1) the required complexity of the estimation network is unknown, 2) existing training algorithms for multilayer perceptrons (MLPs) are inefficient, in terms of training time and use of free parameters, 3) existing bounds on neural network estimation error assume noiseless inputs and are not practical to calculate, 4) there is no generally accepted procedure for finding the best subset of input features to be used in optimal estimation, and 5) a method for automatically developing optimal estimators from training data is not available. In this paper, we present a methodology for attacking these problems. We describe three separate processing blocks which attempt to solve problems (1), (2), and (3). These blocks are then assembled into larger compound systems or blocks which attempt to solve the remaining problems. Examples of multilayer perceptron (MLP) estimators, designed using the proposed system, are given.

关键词： Neural networks Parameter estimation Training data Polynomials Clustering algorithms Multi-layer neural network Multilayer perceptrons backpropagation algorithms Character generation Filters

来源：评论

学校读者我要写书评

暂无评论

A hybrid RBF-HMM system for continuous speech recognition

A hybrid RBF-HMM system for continuous speech recognition

引用

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： W. Reichl G. Ruske Institute for Human-Machine-Communication Munich University of Technology Munchen Germany

A hybrid system for continuous speech recognition, consisting of a neural network with radial basis functions and hidden Markov models is described in this paper together with discriminant training techniques. Initially the neural net is trained to approximate a-posteriori probabilities of single HMM states. These probabilities are used by the Viterbi algorithm to calculate the total scores for the individual hybrid phoneme models. The final training of the hybrid system is based on the 'minimum classification error' objective function, which approximates the misclassification rate of the hybrid classifier, and the 'generalized probabilistic descent' algorithm. The hybrid system was used in continuous speech recognition experiments with phoneme units and shows about 63.8% phoneme recognition rate in a speaker-independent task.

关键词： Speech recognition Hidden Markov models Neural networks Viterbi algorithm backpropagation algorithms Maximum likelihood estimation Pattern classification Radial basis function networks Error probability Training data

来源：评论

学校读者我要写书评

暂无评论

Traditional and neural networks algorithms: applications to the inspection of marble slab

Traditional and neural networks algorithms: applications to ...

引用

IEEE International Conference on Systems, Man and Cybernetics

作者： V.G. Hernandez P.C. Perez L.G.G. Perez L.M.T. Balibrea H. Puyosa Pina Grupo de Visión y Robótica Departamento de Automática Electricidad y EIectronica IndustrialEscuela Técnica Superior de Ingenieros Industriales Universidad de Murcia Spain Grupo de Visiony Robotica Departamento de Automática Electricidad y EIectronica IndustrialEscuela TCcnica Superior de Ingenieros Industriales Universidad de Murcia Spain Grupo de Vision y Robotica Murcia Univ. Cartagena Spain

This article describes the traditional algorithms of deterministic and stochastic cluster analysis and classification (a priori and with learning), by comparing the results obtained using the classifiers based on neural networking techniques in supervised learning (multilayer perceptron with backpropagation). The selected samples, chosen to check the algorithms have been marble slabs of "Sierra de la Puerta" type.

关键词： Neural networks Inspection Slabs Color Surface texture Clustering algorithms backpropagation algorithms Service robots Robot vision systems Statistics

来源：评论

学校读者我要写书评

暂无评论

An artificial neural network to linearize a G (tungsten vs. tungsten 26% rhenium) thermocouple characteristic in the range of zero to 2000/spl deg/C

An artificial neural network to linearize a G (tungsten vs. ...

引用

IEEE International Symposium on Industrial Electronics (ISIE)

作者： M. Attari F. Boudjema M. Heniche Institute of Electronics Department of Instrumentation University of Sciences and Technology Houari-Boumediene Algiers Algeria Inst. of Electron. Houri Boumediene Univ. of Sci. & Technol. Algiers Algeria Department of Electrical Engineering Ecole Nationale Polytechnique Algiers Algeria

An alternative way to correct sensors in order to exhibit good linearity is proposed. This paper describes the design and behavior of an artificial neural network (ANN) with two hidden layers of twelve neurons each for linearizing a static characteristic of a G thermocouple ranged from zero to 2000/spl deg/C. The use of interpolation methods to perform such corrections is also discussed. In the training phase of the ANN, backpropagation and random optimization algorithms have jointly been used to adjust the weights of the network for a desired final error. After training the ANN, it is then used as a neural linearizer to generate the primary variable (temperature) from the thermocouple's output voltage. The accuracy of this method is compared with those obtained by interpolation.

关键词： Tungsten Artificial neural networks Temperature sensors Interpolation Plasma temperature Thermal sensors Read only memory Instruments Sensor phenomena and characterization backpropagation algorithms

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：