Neural Networks for Signal Processing III - Proceedings of the 1993 IEEE-SP Workshop

6-9 Sept. 1993

Filter Results

Displaying Results 1 - 25 of 63
  • Multisensor image classification by structured neural networks

    Publication Year: 1993, Page(s):311 - 320
    Cited by:  Papers (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (605 KB)

    The application of structured neural networks to the supervised classification of multisensor images is discussed. The purpose is to give a criterion for network architecture definition and to allow the interpretation of the network behavior. The latter result can be used to understand the importance of sensors and related channels to the classification task. The networks' architecture is configur... 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 III - Proceedings of the 1993 IEEE-SP Workshop

    Publication Year: 1993
    Request permission for commercial reuse | PDF file iconPDF (37 KB)
    Freely Available from IEEE
  • Fast VLSI implementations for MRF and ANN applications

    Publication Year: 1993, Page(s):460 - 469
    Cited by:  Patents (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (460 KB)

    A VLSI architecture for real-time Markov random field optimization is presented. This architecture contains a simple and fast hardware module implementing the sigmoid function using look-up tables and piecewise linear interpolation. Error bounds are given for computing with this module, and it is shown that it can also used for improving the performance of a systolic architecture for artificial ne... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Quantized, piecewise linear filter network

    Publication Year: 1993, Page(s):470 - 474
    Cited by:  Papers (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (180 KB)

    A quantization based piecewise linear filter network is defined. A method for the training of this network based on local approximation in the input space is devised. The training is carried out by repeatedly alternating between vector quantization of the training set into quantization classes and equalization of the quantization classes linear filter mean square training errors. The equalization ... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Backpropagation through time with fixed memory size requirements

    Publication Year: 1993, Page(s):207 - 215
    Cited by:  Papers (2)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (416 KB)

    A generalization of the backpropagation through time (BPTT) algorithm is presented, which, under reasonable assumptions, can lead to fixed memory size requirements. The idea is to model BPTT as the storage of activations and errors in tapped delay lines, and then generalize the tap delay line to a gamma memory. Since the depth of the gamma filter is T=K/μ, where K is the filter order and μ a... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A feedforward neural network for the wavelet decomposition of discrete time signals

    Publication Year: 1993, Page(s):475 - 484
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (348 KB)

    A feedforward neural network with sigmoidal activation functions is proposed to perform the wavelet decomposition of a discrete time signals. The proposed network is made of two parts, the main network and the auxiliary network. The learning of the auxiliary network is achieved off-line, in a prior phase, in order to identify the desired wavelet. This identification is possible due to the properti... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Hierarchical recurrent networks for learning musical structure

    Publication Year: 1993, Page(s):216 - 225
    Cited by:  Papers (1)  |  Patents (3)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (500 KB)

    Layered neural networks employing feedback links have been proposed for certain sequential pattern tasks in automatic music composition. A hierarchical version of this type of network is studied. The use of such a hierarchical neural network for modeling coarse and fine temporal structure in music is investigated. This network is trained on two classical waltzes and then used to generate novel wal... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Phoneme recognition based on multi-resolution and non-causal context

    Publication Year: 1993, Page(s):343 - 352
    Cited by:  Papers (2)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (480 KB)

    An alternative view of neural-network-based phoneme recognition using multiresolution ideas and noncausal context is suggested. Some suggestions are made regarding target and error weight functions to improve performance and simplify training. Based on these observations, a simple network with self recurrent links of different delays is proposed and tested on the task of speaker- independent recog... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Evaluation of character recognition systems

    Publication Year: 1993, Page(s):485 - 496
    Cited by:  Papers (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (536 KB)

    Eleven different Census Optical Character Recognition Systems systems are evaluated using correlations between the answers of different systems, comparing the decrease in error rate as a function of confidence of recognition, and comparing the writer dependence of recognition. This comparison shows that methods that use different algorithms for feature extraction and recognition perform with very ... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Characterization of network responses to known, unknown, and ambiguous inputs

    Publication Year: 1993, Page(s):226 - 231
    Cited by:  Papers (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (328 KB)

    Neural networks typically classify patterns by mapping the input feature space to the corners of the m-dimensional unit hypercube where m is the number of output classes. When classifier networks of graded threshold neurons are presented with patterns that are strong, ambiguous, or unknown, characteristic responses are emitted. A second tier network can be used to characterize the decision of the ... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • LS-based training algorithm for neural networks

    Publication Year: 1993, Page(s):22 - 29
    Cited by:  Papers (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (580 KB)

    A new training algorithm is presented as a faster alternative to the backpropagation (BP) method. The new approach is based on the solution of a linear system at each step of the learning phase. The squared error at the output of each layer before the nonlinearity is minimized on the entire set of the learning patterns by a block least squares (LS) algorithm. The optimal weights for each layer are... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Text-Dependent speaker verification using recurrent time delay neural networks for feature extraction

    Publication Year: 1993, Page(s):353 - 361
    Cited by:  Papers (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (304 KB)

    The possible application of time delay neural network (TDNN) to the text-dependent speaker verification problem is described and evaluated. Each person to be verified has a personalized neural network, which is trained to extract representative feature vector of the speaker by a particular utterance. A novel model called recurrent time delay neural networks is investigated. The training is carried... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Competitive learning and winning-weighted competition for optimal vector quantizer design

    Publication Year: 1993, Page(s):50 - 59
    Cited by:  Papers (2)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (404 KB)

    It is essential to build a nonparametric model to estimate a probability density function p(x) in the areas of vector quantization, pattern recognition, control, and many others. A generalization of Kohonen learning, the winning-weighted competitive learning (WWCL), is presented for a better approximation of p(x) and fast learning convergence by introducing the principle of maximum information pre... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Classification of biomagnetic field patterns by neural networks

    Publication Year: 1993, Page(s):572 - 581
    Cited by:  Papers (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (484 KB)

    When analyzing biomagnetic fields, three-dimensional source reconstruction plays an important role. A method is presented which facilitates preprocessing for such a reconstruction. A neural net classifier decides whether from a given magnetoencephalographic map a localization using a given source model can be carried through. The performance in practical applications is compared for various types ... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A modified recurrent cascade-correlation network for radar signal pulse detection

    Publication Year: 1993, Page(s):497 - 506
    Cited by:  Patents (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (528 KB)

    A Jordan-style cascade-correlation architecture is developed for radar signal pulse detection. The cascade-correlation learning architecture is modified to facilitate hardware implementation of the network. The network is constructed using only two hidden layers, with nodes added to the layers in a lateral fashion. Comparisons to networks trained using backpropagation and genetic algorithms indica... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A modular neural network architecture for pattern classification

    Publication Year: 1993, Page(s):232 - 238
    Cited by:  Papers (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (248 KB)

    A modular neural network architecture is proposed to classify binary and continuous patterns. This system consists of a supervised feedforward backpropagation network and an unsupervised self-organization map network. The supervised feedforward (basic) network is trained until a saturation error level occurs. Simultaneously, the unsupervised self-organization map (control) network fluids the mappi... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • VLSI Hamming neural net showing digital decoding

    Publication Year: 1993, Page(s):405 - 410
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (196 KB)

    The electrical performance in PSpice for a VLSI Hamming neural net in CMOS 2-micron technology is presented. It features analog parallel processing and digital decoding, keeping the low circuit-interconnection complexity of early schemes View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Neural networks for localized approximation of real functions

    Publication Year: 1993, Page(s):190 - 196
    Cited by:  Papers (2)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (300 KB)

    The problem of constructing universal networks capable of approximating all functions having bounded derivatives is discussed. It is demonstrated that, using standard ideas from the theory of spline approximation, it is possible to construct such networks to provide localized approximation. The networks can be used to implement multivariate analogues of the Chui-Wang wavelets (1990) and also for t... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Map estimation and the multilayer perceptron

    Publication Year: 1993, Page(s):30 - 39
    Cited by:  Papers (7)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (364 KB)

    The utility of linear transformations to compress data efficiently for purposes of estimation via neural net techniques is describing. A method for comparing transforms based upon transform domain error bounds is reviewed, and a method for improving the transform through re-ordering is described. The bounds also give clues as to how many transform coefficients are necessary and when training can b... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A technique for adapting to speech rate

    Publication Year: 1993, Page(s):382 - 391
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (484 KB)

    A technique is proposed for automatically estimating and dynamically adapting to the rate of a speech signal. A recurrent network is first trained to predict the input signal at a normal rate. Once trained, the network essentially becomes a model of the signal. Then, with the weights fixed, the network's time constant is adapted using gradient descent as it receives the same signal at a different ... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A new learning approach based on equidistortion principle for optimal vector quantizer design

    Publication Year: 1993, Page(s):362 - 371
    Cited by:  Papers (3)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (448 KB)

    The authors theoretically derive a basic principle called the equidistortion principle for the design of optimal vector quantizers. This principle can be regarded as a extension of Gersho's theory (1979). A new learning algorithm is presented with a selection mechanism based on this principle. Since no probabilistic model is assumed in deriving the principle, the associated algorithm, unlike conve... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Hierarchical wavelet neural networks

    Publication Year: 1993, Page(s):60 - 67
    Cited by:  Papers (2)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (372 KB)

    Neural networks can be used in nonlinear system modeling and prediction applications. Wavelet decomposition provides a method of examining a signal at multiple scales. The authors draw upon the connection between these two fields. A method is outlined which exploits the localized, hierarchical nature of wavelets in the learning of time series. This is achieved by having a dynamic network-one in wh... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Ordered vector quantization for neural network pattern classification

    Publication Year: 1993, Page(s):141 - 150
    Cited by:  Papers (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (668 KB)

    The accurate classification of time sequences of vectors is a common goal in signal processing. Vector quantization (VQ) has commonly been used to help encode vectors for subsequent classification. The authors depart from this past approach proposing the use of VQ codebook indices, as opposed to codebook vectors. It is shown that one-dimensional ordering of these indices markedly improves the neur... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Designer networks for time series processing

    Publication Year: 1993, Page(s):78 - 87
    Cited by:  Papers (9)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (472 KB)

    The conventional tapped-delay neural net may be analyzed using statistical methods and the results of such analysis can be applied to model optimization. The authors review and extend efforts to demonstrate the power of this strategy within time series processing. They attempt to design compact networks using the so-called optima brain damage (OBD) method. The benefits from compact architectures a... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • From regularization to radial, tensor and additive splines

    Publication Year: 1993, Page(s):3 - 10
    Cited by:  Papers (1)  |  Patents (3)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (316 KB)

    Poggio and Girosi showed that regularization principles lead to approximation schemes which are equivalent to networks with one layer of hidden units, called regularization networks. They summarize their results (1993) that show that regularization networks encompass a much broader range of approximation schemes, including many of the general additive models and some of the neural networks. In par... View full abstract»

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