By Topic

Neural Networks for Signal Processing II Proceedings of the 1992 IEEE Workshop

Aug. 31 1992-Sept. 2 1992

Filter Results

Displaying Results 1 - 25 of 64
  • An electronic parallel neural CAM for decoding

    Publication Year: 1992, Page(s):581 - 587
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (288 KB)

    The authors report measurements taken on an electronic neural system configured for content addressable memory (CAM) using a high-capacity architecture. It is shown that Boltzmann and mean-field learning networks can be implemented in a parallel, analog VLSI system. This system was used to perform experiments with mean-field CAM. The hardware settles on a stored codeword in about 10 mu s roughly i... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Globally trained neural network architecture for image compression

    Publication Year: 1992, Page(s):289 - 295
    Cited by:  Papers (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (321 KB)

    The authors discuss the development of a coding system for image transmission based on block-transform coding and vector quantization. Moreover, a classification of the image blocks is performed in the spatial domain. An architecture incorporating both multilayered perceptron and self-organizing feature map neural networks and a block classification is considered to realize the image coding scheme... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Spectral representations for speech recognition by neural networks - A tutorial

    Publication Year: 1992, Page(s):214 - 222
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (301 KB)

    Spectrum-based speech representations are discussed. Spectral representations, in order to be useful for speech recognition, need to be justified from both the computational (analytical) and the perceptual viewpoints. The authors' discussion of spectral representations, therefore, includes both the computational model and the associated measures of similarity that are appropriate for neural networ... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Text-independent talker identification system combining connectionist and conventional models

    Publication Year: 1992, Page(s):131 - 138
    Cited by:  Papers (2)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (391 KB)

    Several techniques have been used for speaker identification which have different characteristics and capabilities. The respective merits of three different systems respectively employing neural networks, hidden Markov models, and multivariate autoregressive models are compared. A novel text-independent speaker identification system based on the cooperation of these different techniques is present... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Empirical risk optimisation: neural networks and dynamic programming

    Publication Year: 1992, Page(s):121 - 130
    Cited by:  Papers (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (430 KB)

    The authors propose a novel system for speech recognition which makes a multilayer perceptron and a dynamic programming module cooperate. It is trained through a cost function inspired by learning vector quantization which approximates the empirical average risk of misclassification. All the modules of the system are trained simultaneously through gradient backpropagation; this ensures the optimal... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Neural Networks for Signal Processing II. Proceedings of the IEEE-SP Workshop (Cat. No.92TH0430-9)

    Publication Year: 1992
    Request permission for commercial reuse | PDF file iconPDF (32 KB)
    Freely Available from IEEE
  • A partial analysis of stochastic convergence in a generalized two-layer perceptron with backpropagation learning

    Publication Year: 1992, Page(s):522 - 530
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (320 KB)

    The authors study the stationary points of a two-layer perceptron which attempts to identify the parameters of a specific stochastic nonlinear system. The training sequence is modeled as the output of the nonlinear system, with an input comprising an independent sequence of zero mean Gaussian vectors with independent components. The training rule is a limiting case of backpropagation (to simplify ... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Maximum mutual information training of a neural predictive-based HMM speech recognition system

    Publication Year: 1992, Page(s):164 - 173
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (328 KB)

    A corrective training scheme based on the maximum mutual information (MMI) criterion is developed for training a neural predictive-based HMM (hidden Markov model) speech recognition system. The performance of the system on speech recognition tasks when trained with this technique is compared to its performance when trained using the maximum likelihood (ML) criterion. Preliminary results obtained i... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A recurrent neural network for nonlinear time series prediction-a comparative study

    Publication Year: 1992, Page(s):531 - 539
    Cited by:  Papers (4)  |  Patents (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (292 KB)

    The performance of recurrent neural networks (RNNs) is compared with those of conventional nonlinear prediction schemes, such as a Kalman predictor (KP) based on a state-dependent model and a second-order Volterra filter. Simulation results on some typical nonlinear time series data indicate that the neural network can predict with accuracies on a par with the KP. It is noted that a higher-order e... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Training continuous density hidden Markov models in association with self-organizing maps and LVQ

    Publication Year: 1992, Page(s):174 - 183
    Cited by:  Papers (2)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (436 KB)

    The authors propose a novel initialization method for continuous observation density hidden Markov models (CDHMMs) that is based on self-organizing maps (SOMs) and learning vector quantization (LVQ). The framework is to transcribe speech into phoneme sequences using CDHMMs as phoneme models. When numerous mixtures of, for example, Gaussian density functions are used to model the observation distri... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Dispersive networks for nonlinear adaptive filters

    Publication Year: 1992, Page(s):540 - 549
    Cited by:  Papers (4)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (364 KB)

    The authors describe a dispersive network architecture that can be used for nonlinear adaptive channel equalization and signal prediction. Dispersive networks contain internal delay elements that spread out features in the input signal over time and space, so that they influence the output at multiple points in the future. When used for equalization, these networks can compensate for nonlinear cha... View full abstract»

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

    Publication Year: 1992, Page(s):184 - 193
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (404 KB)

    The authors first introduce a novel approach for unsupervised sequence classification, the competitive sequence learning (CSL) system. The CSL system consists of several extended Kohonen feature maps which are ordered in a hierarchy. The CSL maps develop a representation for subsequences during the training procedure, with an increasing abstraction on the higher maps. The authors apply their appro... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • An adaptive neural network model for distinguishing line- and edge detection from texture segregation

    Publication Year: 1992, Page(s):391 - 400
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (468 KB)

    The authors consider an important paradigm in vision: distinguishing object contours or edges (and lines) from object surface textures. To accomplish this, an artificial neural network model, called the EDANN model, is used for both texture segregation and line and edge detection starting from a common bank of spatial filters. The model provides different representations of a retinal image in such... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Adaptive training of feedback neural networks for non-linear filtering

    Publication Year: 1992, Page(s):550 - 559
    Cited by:  Papers (6)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (348 KB)

    The authors propose a general framework which encompasses the training of neural networks and the adaptation of filters. It is shown that neural networks can be considered as general nonlinear filters which can be trained adaptively, i.e., which can undergo continual training. A unified view of gradient-based training algorithms for feedback networks is proposed, which gives rise to new algorithms... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Hierarchical perceptron (HiPer) networks for signal/image classifications

    Publication Year: 1992, Page(s):267 - 278
    Cited by:  Papers (4)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (496 KB)

    A new class of decision-based neural networks (DBNNs) is introduced. These networks combine the perceptron-like learning rule with a hierarchical nonlinear network structure and are called HiPer nets. Two HiPer net structures are proposed: hidden-node and subcluster structures. The authors explore several variants of HiPer nets based on the different hierarchical structures and basis functions and... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A two-layer Kohonen neural network using a cochlear model as a front-end processor for a speech recognition system

    Publication Year: 1992, Page(s):139 - 148
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (364 KB)

    The authors describe a two-layer neural network speech recognition system based on Kohonen's algorithm. A cochlear model is used as a front-end processor for the system. The basilar membrane is represented by a cascade of 128 digital filters, of which 90 filters fall within the speech bandwidth of 250 Hz to 4 kHz. The outputs of these 90 filters are presented as the input vector to the first layer... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Unsupervised multi-level segmentation of multispectral images

    Publication Year: 1992, Page(s):363 - 372
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (520 KB)

    The authors describe a scheme that performs multilevel segmentation of an image at many scales using a multiresolution texture representation. Each level uses anisotropic diffusion to segment a multispectral image at successively lower resolutions. Texture and statistical similarities between and within levels guides the diffusion process. The restriction of coarse-to-fine segmentation is removed,... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Ensemble methods for handwritten digit recognition

    Publication Year: 1992, Page(s):333 - 342
    Cited by:  Papers (26)  |  Patents (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (404 KB)

    Neural network ensembles are applied to handwritten digit recognition. The individual networks of the ensemble are combinations of sparse look-up tables (LUTs) with random receptive fields. It is shown that the consensus of a group of networks outperforms the best individual of the ensemble. It is further shown that it is possible to estimate the ensemble performance as well as the learning curve ... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A mathematical model for speech processing

    Publication Year: 1992, Page(s):194 - 203
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (408 KB)

    The authors develop a mathematical model of the mechanisms that the auditory apparatus uses for signal processing. They have studied the model of the peripheral auditory apparatus described by S. Seneff (1985, 1988). They complete it by adding new features such as a pitch detector and a neural synchrony detector module, by modifying some filter parameters, and by integrating it with the variations... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Adaptive segmentation of textured images using linear prediction and neural networks

    Publication Year: 1992, Page(s):401 - 410
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (624 KB)

    An adaptive technique for classifying and segmenting textured images is presented. This technique uses an efficient least squares algorithm for recursive estimation of two-dimensional autoregressive texture models and neural networks for recursive classification of the models. A network with fixed, but space-varying, interconnection weights is used to optimally select a small representative set of... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Pattern classification with a codebook-excited neural network

    Publication Year: 1992, Page(s):223 - 232
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (396 KB)

    A codebook-excited neural network (CENN) is formed by a multi-layer perceptron excited by a set of code vectors. The authors study its discriminant performance and compare it with other models. The performance improvement with the CENN is demonstrated in a number of cases. The CENN has been developed for classification. The multilayer codebook-excited feedforward neural network enhances the separa... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Constructing neural networks for contact tracking

    Publication Year: 1992, Page(s):560 - 569
    Cited by:  Papers (3)  |  Patents (2)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (432 KB)

    A neural network approach for contact state estimation is presented. This neural network, NICE (neurally inspired contact estimation), has been constructed to directly embody the major problem domain constraint of uniform contact velocity and heading. NICE networks are constructed, not trained, to estimate contact position and motion from angle-of-arrival (AOA) measurements. The major advantages o... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Classification of simulated radar imagery using lateral inhibition neural networks

    Publication Year: 1992, Page(s):279 - 288
    Cited by:  Papers (4)  |  Patents (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (412 KB)

    The use of neural networks for the classification of simulated inverse synthetic aperture radar imagery is investigated. Symmetries of the artificial imagery make the use of localized moments a convenient preprocessing tool for the inputs to a neural network. A database of simulated targets was obtained by warping dynamical models to representative angles and generating images with differing targe... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A neural feedforward network with a polynomial nonlinearity

    Publication Year: 1992, Page(s):49 - 58
    Cited by:  Patents (5)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (308 KB)

    A novel neural network based on the Wiener model is proposed. The network is composed of a hidden layer of preprocessing neurons followed by a polynomial nonlinearity and a linear output neuron. The author tries to solve the problem of finding an appropriate preprocessing method by using a modified backpropagation algorithm. It is shown by the use of calculation trees that the proposed approach is... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Self-structuring hidden control neural models

    Publication Year: 1992, Page(s):149 - 156
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (292 KB)

    The authors propose a self-structuring hidden control (SHC) neural model for pattern recognition which establishes a near-optimal architecture during training. A significant network architecture reduction in terms of the number of hidden processing elements (PEs) is typically achieved. The SHC model combines self-structuring architecture generation with nonlinear prediction and hidden Markov model... View full abstract»

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