Scheduled System Maintenance:
Some services will be unavailable Sunday, March 29th through Monday, March 30th. We apologize for the inconvenience.
By Topic

Neural Networks Proceedings, 1998. IEEE World Congress on Computational Intelligence. The 1998 IEEE International Joint Conference on

Date 4-9 May 1998

Go

Filter Results

Displaying Results 1 - 25 of 158
  • The 1998 IEEE International Joint Conference on Neural Networks Proceedings [front matter]

    Publication Year: 1998 , Page(s): i - xxxvi
    Save to Project icon | Request Permissions | PDF file iconPDF (2194 KB)  
    Freely Available from IEEE
  • Author's index

    Publication Year: 1998 , Page(s): A
    Save to Project icon | Request Permissions | PDF file iconPDF (466 KB)  
    Freely Available from IEEE
  • Handwritten digit recognition by neural `gas' model and population decoding

    Publication Year: 1998 , Page(s): 1727 - 1731 vol.3
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (484 KB)  

    In this paper, we present a handwritten digit recognition scheme using a topology representation model called neural gas. Instead of applying the model only for feature extraction, we train test separate gas models which aim to describe the data submanifolds respectively in the ten classes. A modular classification system is proposed based on the idea of population decoding, as a topographic map essentially provide a kind of population code for the input. Experiment results show a fast learning and a high recognition rate View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Application of neural networks in spatio-temporal hand gesture recognition

    Publication Year: 1998 , Page(s): 2116 - 2121 vol.3
    Cited by:  Papers (2)  |  Patents (5)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (556 KB)  

    Several successful approaches to spatio-temporal signal processing such as speech recognition and hand gesture recognition have been proposed. Most of them involve time alignment which requires substantial computation and considerable memory storage. In this paper, we present a neural-network-based approach to spatio-temporal pattern recognition. This approach employs a powerful method based on hyperrectangular composite neural networks (HRCNNs) for selecting templates, therefore, considerable memory is alleviated. In addition, it greatly reduces substantial computation in the matching process because it obviates time alignment. Two databases consisted of 51 spatio-temporal hand gestures were utilized for verifying its performance. An encouraging experimental result confirmed the effectiveness of the proposed method View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Control of a static nonlinear plant using a neural network linearization

    Publication Year: 1998 , Page(s): 2136 - 2141 vol.3
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (460 KB)  

    One possibility to control a static plant is the design of a controller based on the inverse of an identified model. For nonlinear plants, determining or identifying the plant model may be a difficult task. When a state space model of the plant is not explicitly needed, it is possible to consider the plant as a black box and approximate the plant using neural networks. In this paper a control strategy is presented, based on the combination of classical linear control methods with a neural network that inverses the plants nonlinear characteristics. A proof is given that the plant can be positioned with an arbitrary small positioning error. The method is experimentally illustrated on the positioning control of a flexible robot arm. The results of the neural network based control are compared with a PI controller View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Stability analysis of neurocontrol systems using a describing function

    Publication Year: 1998 , Page(s): 2126 - 2130 vol.3
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (232 KB)  

    This paper presents the stability analysis of closed loop systems with a linear plant and a neural network as controller (neurocontroller). The describing function of the neural network is used to determine the bounds for the network weights in order to predict the limit cycles and stable system response View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Designing ANN forecasting architectures from data conflict plots

    Publication Year: 1998 , Page(s): 2519 - 2524 vol.3
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (628 KB)  

    One of the main issues in the analysis of a time series is its forecasting. Many questions arise in the design of a neural network that aims to capture the dynamics of a temporal sequence in order to predict it. In a reproducible way we want to find decision strategies for the preprocessing and the architecture of the network. In this paper we introduce a novel technique to extract important data features, called the data conflict plot. The conflict plot is used to design a modified architecture for the prediction of signals with distinct periodic components. Instead of a single delay line, this architecture is preceded by several incompletely connected delay lines View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A modified current mode Hamming neural network for totally unconstrained handwritten numeral recognition

    Publication Year: 1998 , Page(s): 1857 - 1860 vol.3
    Cited by:  Papers (3)  |  Patents (4)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (368 KB)  

    A compact smart current mode Hamming neural network for classifying complex patterns such as totally unconstrained handwritten digits is presented. It is based on multi-threshold template matching, multistage matching and k-WTB (k-winner-taker-all). The neural classifier consists of two kinds of templates: one is a binary template and the other is a multi-value programmable template, each of them has its own threshold and realized in MOS current mirrors, the current mode k-WTA which is reconfigurable is put forward. The second stage matching templates are programmable from outside the chip. This mixed analog-digital Hamming neural classifier can be fabricated in a standard digital CMOS technology View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Robust learning and identification of patterns in statistical process control charts using a hybrid RBF fuzzy ARTMAP neural network

    Publication Year: 1998 , Page(s): 1694 - 1699 vol.3
    Cited by:  Papers (3)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (596 KB)  

    The quality control of the manufacturing process in FMS is a critical factor, requiring flexible and intelligent quality control systems that are capable of autonomous pattern identification. Due to its learning and generalization capabilities, neural networks have good perspectives for this task. One of the most important difficulties in pattern identification with neural networks is the sensibility to the presentation order of the training patterns. This paper presents a hybrid network, RBF fuzzy-ARTMAP, which is capable of online incremental learning, 98% less sensible to the presentation order of training patterns than the fuzzy-ARTMAP network. Also, this work compares the performance of the RBF fuzzy-ARTMAP network with the fuzzy-ARTMAP network in the identification of six different “patterns” in SPC qualify control charts View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Batch self-organizing maps on a unit sphere

    Publication Year: 1998 , Page(s): 2273 - 2276 vol.3
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (292 KB)  

    Kohonen's batch map over data on a unit sphere is modified. An energy function is proposed and the convergence of the algorithm is proven. It is shown that this deterministic, batch-mode self-organizing algorithm is efficient and performs well View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Face recognition using curvilinear component analysis

    Publication Year: 1998 , Page(s): 1778 - 1783 vol.3
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (428 KB)  

    Automated face recognition can be applied in a number of situations including personal identification, mug shot matching, store security, and crowd surveillance. A large number of techniques based on linear methods of dimensionality reduction, such as principal component analysis (PCA), have recently been proposed. Motivated by the possibility of increased performance, we pursue in this paper a face recognition paradigm based on nonlinear methods of dimensionality reduction. More specifically, we use the recently proposed curvilinear component analysis (CCA) to obtain a reduced dimension representation of face images. Two types of classifiers, a k-NN classifier and a pseudo-inverse rule based classifier are used for assigning class labels to sample vectors in the reduced dimension space. The algorithm is found to be much faster and has better performance than a linear PCA based approach View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Modeling time dependencies in the mixture of experts

    Publication Year: 1998 , Page(s): 2324 - 2327 vol.3
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (324 KB)  

    The mixture of experts, as it was originally formulated, is a static algorithm in the sense that the output of the network, and parameter updates during training, are completely independent from one time step to the next. This independence creates difficulties when the model is applied to time series prediction. We address this by adding memory to the mixture of experts. A Gaussian assumption on each expert's error is replaced by a chi-square distribution on the local (in time) root mean square error. We derive new gradient descent equations, and present a simulation that demonstrates an improvement in the segmentation of a time series over the classical algorithm View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Identifying cutting sound characteristics in machine tool industry with a neural network

    Publication Year: 1998 , Page(s): 2459 - 2464 vol.3
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (396 KB)  

    This paper presents a method for identifying cutting sound characteristics for machine tool industry based on a robust time-variant sound recognition system. The sound signal is compressed using linear prediction analysis method and then recognized by an artificial neural network. The procedure taken here is based on the following: (1) extraction of time-variant spectral features (i.e., raw data of sound), (2) characterization of each sample by observing the autocorrelation coefficients and reflection coefficients of the sampled data, and (3) training of an artificial neural network to identify extracted sound samples. The proposed technique is shown to be very effective, accurate, and powerful in performing sound data identification View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Optimal tailoring of trajectories, growing training sets and recurrent networks for spoken word recognition

    Publication Year: 1998 , Page(s): 2169 - 2174 vol.3
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (476 KB)  

    A novel system that efficiently integrates two types of neural networks for reliably performing isolated word recognition is described. The recognition system comprises of a feature extractor that includes a self organizing map for an optimal tailoring of trajectory representations of words in reduced dimension feature spaces. Experimental results indicate that such lower dimensional trajectories can provide a reliable representation of spoken words, while reducing the training complexity for the recognition of the trajectory. A recurrent neural network is employed for performing trajectory recognition and a method that allows us to progressively grow the training set is utilized for network training. The optimal tailoring of trajectories and growing training sets are two innovations that result in a superior training of the recurrent neural network, which in turn delivers a robust word recognition performance tolerating wide variations in the speech signal View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A neural network based technique to locate and classify microcalcifications in digital mammograms

    Publication Year: 1998 , Page(s): 1790 - 1793 vol.3
    Cited by:  Papers (4)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (400 KB)  

    This paper proposes a technique that extracts suspicious areas containing microcalcifications in digital mammograms and classifies them into two categories whether they contain benign or malignant clusters. The centroids and radiuses provided by expert radiologist are being used to locate and extract suspicious areas. Neural network's generalisation abilities are used to classify them into benign or malignant. The technique has been implemented in C++ on the SP2 supercomputer. The database from the Department of Radiology at the University of Nijmegen and Lawrence Livermore National Laboratory has been used for the experiments. The preliminary results are very promising. Some of them are presented in this paper View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A fast exact parallel implementation of the k-nearest neighbour pattern classifier

    Publication Year: 1998 , Page(s): 1867 - 1872 vol.3
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (496 KB)  

    A neural network architecture is presented that precisely implements the k-nearest-neighbour (k-NN) pattern classification rule. Given n exemplars, the size of the architecture grows O(n) and the time taken per classification grows O(log n). This offers perhaps the most useful neural implementation of the k-NN classifier compared to previous implementations, which suffer either from worst-case exponential training time, excessively large networks, unpredictable classification times, or inexact implementations of the classification rule View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • BKYY dimension reduction and determination

    Publication Year: 1998 , Page(s): 1822 - 1827 vol.3
    Cited by:  Papers (3)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (536 KB)  

    A new theory is proposed for dimension reduction and determination (DRD), called the Bayesian Kullback Ying-Yang (BKYY) learning theory, which is a special case BYY learning system. This theory not only includes the conventional factor analysis, principal component analysis (PCA) type linear mapping, and LMSER based nonlinear PCA as special cases, but also provides a unified general framework with a stochastic implementing procedure for developing various linear and nonlinear DRD techniques together with a new theory for determining the dimension k of the reduced subspace. As examples, we provide: 1) a new batch and adaptive algorithm for factor analysis, 2) criteria for determining the number of factors and the dimension of PCA subspace, 3) a procedure for a specific nonlinear BYY DRD based on Gaussian mixtures, and 4) extensions for auto-association and LMSER nonlinear PCA. Some experimental results are demonstrated View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A temporal difference method-based prediction scheme applied to fading power signals

    Publication Year: 1998 , Page(s): 1954 - 1959 vol.3
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (560 KB)  

    We first briefly discuss the operating principle of the temporal difference (TD) method. A TD method-based multi-step ahead prediction scheme using the modified Elman neural network (MENN) is then set up. This prediction approach provides for online adaptation and fast convergence rate. Next, it is applied to the prediction of the occurrence of long term deep fading in mobile communication systems. Simulation experiments reveal that our prediction scheme is capable of predicting the degree of occurrence possibility of deep fading. Based on this prediction result, the power control of cellular phone systems employing the reinforcement learning method will be investigated in the future View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Neural networks for nonlinear mutual prediction of coupled chaotic time series

    Publication Year: 1998 , Page(s): 1937 - 1942 vol.3
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (292 KB)  

    Multilayer perceptrons (MLP) were trained to mutually predict nonlinearly coupled identical Henon systems. Several combinations of input and target time series were presented to networks of different structure during training. After presenting the trained networks with a short segment of Henon data they were able to generate Henon time series of variable duration. This was verified by comparing the attractors of the training set and the generated data. Furthermore, the performance of mutual prediction of data outside the training set was found to be dependent on the strength of coupling among the chaotic time series and on their similarities regarding their generation equations. The method was also applied to univariate and multivariate ECoG data. The motivation of this work is to predict and analyse the development of epileptic seizures by searching for recurring nonlinear dependencies and similarities in multivariate ECoG recordings View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Simple learning algorithm for recurrent networks to realize short-term memories

    Publication Year: 1998 , Page(s): 2367 - 2372 vol.3
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (424 KB)  

    A simple supervised learning algorithm for recurrent neural networks is proposed. It needs only O(n2) memories and O(n 2) calculations, where n is the number of neurons, by limiting the problems to a delayed recognition (short-term memory) problem. Since O(n2) is the same as the order of the number of connections in the neural network, it is suitable for implementation. This learning algorithm is similar to the conventional static backpropagation learning. Connection weights are modified by the products of the propagated error signal and some variables that hold the information about the past pre-synaptic neuron output View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Constructing high order perceptrons with genetic algorithms

    Publication Year: 1998 , Page(s): 1920 - 1925 vol.3
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (620 KB)  

    Constructive induction, which is defined to be the process of constructing new and useful features from existing ones, has been extensively studied in the literature. Since the number of possible high order features for any given learning problem is exponential in the number of input attributes (where the order of a feature is defined to be the number of attributes of which it is composed), the main problem faced by constructive induction is in selecting which features to use out of this exponentially large set of potential features. For any feature set chosen the desirable characteristics are minimality and generalization performance. The paper uses a combination of genetic algorithms and linear programming techniques to generate feature sets. The genetic algorithm searches for higher order features while at the same time seeking to minimize the size of the feature set in order to produce a feature set with good generalization accuracy. The features chosen are used as inputs to a high order perceptron network which is trained with an interior point linear programming method. Performance on a holdout set is used in conjunction with complexity penalization in order to insure that the final feature set generated by the genetic algorithm does not overfit the training data View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Robust nonlinear control using neural networks

    Publication Year: 1998 , Page(s): 2104 - 2109 vol.3
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (432 KB)  

    In this article, the influence of uncertainty on weights and biases of neural networks on the input/output behavior is investigated. Moreover, a uncertainty description of uncertain neural networks is derived and an appropriate norm bound of the model uncertainty, which is needed for robust control design, is derived. Finally, feedback linearization is used in order to fully incorporate neural networks in standard robust l1 model based control View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Recurrent neural networks for reinforcement learning: architecture, learning algorithms and internal representation

    Publication Year: 1998 , Page(s): 2010 - 2015 vol.3
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (480 KB)  

    Reinforcement learning is a learning scheme for an autonomous agent that allows the agent to find the optimal policy of taking actions which maximize a scalar reinforcement signal in unknown environments. If the agent has access to the whole state of the environment, a reactive policy which maps the sensory input to the action is sufficient. However, if the state of the environment is partially observable, special methods for creating a dynamic policy that utilizes the past observations are necessary. To overcome this problem, the authors have proposed a method using recurrent neural networks with Q-learning, as a learning agent. The paper compares several types of network architecture and learning algorithms for this method through computer simulation. Further, the internal representation in the trained networks is examined using a clustering technique. It shows that the representation of the environmental state is developed well in the networks View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Hybrid and constructive neural networks applied to a prediction problem in agriculture

    Publication Year: 1998 , Page(s): 1932 - 1936 vol.3
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (400 KB)  

    The application of artificial neural networks to the solution of problems in agriculture is rarely seen. Motivated by their great potential for dealing with nonlinear prediction tasks, two neural network architectures are independently used to implement alternative tools with the goal of predicting soya production: a hybrid architecture, based on a composition of a Kohonen self-organizing map and a multilayer perceptron, and a constructive architecture, based on projection pursuit learning. Whenever a low harvest is anticipated by the prediction tool, from a set of data extracted at the beginning of the life cycle of the plant, the ultimate purpose is to employ techniques for the correction of the soil composition, aiming at reversing the scene. The output to be predicted is a nonlinear function of a high number of input variables, which prevents the adoption of conventional prediction strategies. The two prediction tools presented here can be directly applied to all prediction problems of similar complexity in other research areas View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Robust speech recognition with dynamic synapses

    Publication Year: 1998 , Page(s): 2175 - 2179 vol.3
    Cited by:  Papers (2)  |  Patents (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (528 KB)  

    We have developed a speech recognition system employing the concept of dynamic synapses. A dynamic synapse incorporates fundamental features of biological neurons including presynaptic mechanisms influencing the probability of neurotransmitter release from an axon terminal. With these mechanisms, the probability of neurotransmitter release becomes a function of the temporal pattern of action potential occurrence, and hence, transforming a spike train into a sequence of discrete release events. When presynaptic mechanisms vary quantitatively across the axon terminals of a single neuron, an array of spatially distributed temporal patterns can be generated. In other words, information is coded in the spatio-temporal patterns of release events which provides an exponential growth of coding capacity for the output signals of a single neuron. A dynamic learning algorithm is developed in which alterations of the presynaptic mechanisms lead to different pattern transformation functions while changes in the postsynaptic mechanisms determines how the synaptic signals are to be combined. We demonstrate the computational capability of dynamic synapses by performing speech recognition from unprocessed, noisy raw waveforms of words spoken by multiple speakers with a simple neural network consisting of a small number of neurons connected with dynamic synapses. The system is highly robust against noise, and outperformed human listeners under some conditions View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.