By Topic

Neural Networks, 1991., IJCNN-91-Seattle International Joint Conference on

Date 8-12 July 1991

Go

Filter Results

Displaying Results 1 - 25 of 383
  • Iterative methods for neural network design

    Publication Year: 1991
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (31 KB)

    Summary form only given, as follows. Three iterative algorithms for designing Hopfield neural networks were investigated: the relaxation method, the Cimino method, and the primal-dual method. All three algorithms arise from iterative methods for solving systems of linear inequalities. Convergence results were obtained for all three algorithms. In addition, this approach to neural network design is... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Novel preprocessing techniques as an aid to hand-printed character recognition

    Publication Year: 1991
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (48 KB)

    Summary form only given. Two preprocessing techniques designed to greatly reduce the burden of classification on an artificial neural network have been developed. The first is a transform which is invariant to rotation, size, and breaks in characters. The second is a low-level feature extractor, in which the features have been statistically selected. The resulting output yields a 45% reduction in ... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Learning to generate subgoals for action sequences

    Publication Year: 1991
    Cited by:  Papers (4)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (28 KB)

    Summary form only given. None of the existing learning algorithms for neural networks in time-varying environments addresses the problems of learning to `divide and conquer'. Algorithms based on pure gradient descent or on adaptive critic methods are not suitable for dynamic control problems with long time lags between actions and consequences, and that there is a need for algorithms that perform ... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Self-organizing analog fields (SOAF)

    Publication Year: 1991
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (64 KB)

    Summary form only given. A neural network paradigm called self-organizing analog fields (SOAF) learns many-to-many, analog (as opposed to winner-take-all, WTA) spatiotemporal mappings. The learning mechanisms employed are unsupervised learning of features and self-organization of connectivity. Homeostasis is defined, providing a necessary extra degree of freedom to solve the credit assignment prob... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Approximate representation of a continuous function by a neural network with scaled or unscaled sigmoid units

    Publication Year: 1991
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (52 KB)

    Summary form only given. The author investigates the capability of three-layered neural networks with a scaled or unscaled sigmoid activation function of the hidden layer units in uniformly approximate representation of continuous functions. For the approximation on a compact set, any sigmoid function can be the activation function without scaling. Even for the approximation of continuous function... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Optimal dimensioning of counterpropagation neural networks

    Publication Year: 1991, Page(s):455 - 459 vol.2
    Cited by:  Papers (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (400 KB)

    The absence of automated tools in the area of automated neural network design can be explained by the corresponding paucity of rigorous neural network composition techniques. The author suggests a hybrid architecture as the basis for a computer-aided neural network engineering tool. Such a tool is expected to complete automatically the minute yet important neural network architecture details. The ... View full abstract»

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

    Publication Year: 1991
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (44 KB)

    Summary form only given, as follows. A structural viewpoint for programming computers is discussed that constitutes a common point between humans and computers. A structural description of an intended activity is closer to the processes that are implemented in computers than a corresponding description in phrases. The activity is described in the form of abstract networks. Derivations from the mec... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A logical architecture for supervised learning

    Publication Year: 1991
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (32 KB)

    Summary form only given, as follows. The author discusses a neural network architecture for supervised learning with inherent stability properties. The architecture employs two ART 1 (adaptive resonance theory) unsupervised systems with supervision through interconnects. The supervised learning system, trained in a particular manner, responds properly to the training set of patterns and responds t... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A design principle of the analog neural network for combinatorial optimization: findings from a study on the placement problem

    Publication Year: 1991
    Cited by:  Papers (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (52 KB)

    Summary form only given. A design principle of a Hopfield model to be used for combinatorial optimization is discussed through its application to a circuit block placement problem. The authors propose several means to improve the energy function of the Hopfield model, and a tuning rule of its parameters. Computer simulations show that the Hopfield model yields quite satisfactory solutions to the h... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Fuzzy rule extraction from a multilayered neural network

    Publication Year: 1991, Page(s):461 - 465 vol.2
    Cited by:  Papers (21)  |  Patents (8)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (284 KB)

    A fuzzy rule extraction method from a multilayered neural network is proposed to realize the multiplicative merits of neural and fuzzy theories. The proposed method uses a fuzzy-neuron network structure which includes input and output layers that convert input signals to membership values. An evaluation index, called the `casual index' can evaluate the weights which are acquired by learning and ca... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Computational model of a biologically plausible cognitive map

    Publication Year: 1991
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (52 KB)

    Summary form only given, as follows. Humans develop spatial cognitive maps using knowledge gathered while repeatedly observing their environment. This knowledge may come from landmarks present in the environment, the relative position(s) of these landmarks, or available paths along which an observer can reach these landmarks. Recent neurophysiological findings in rats suggest that there are specia... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A superior error function for training neural networks

    Publication Year: 1991, Page(s):49 - 52 vol.2
    Cited by:  Papers (3)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (224 KB)

    The authors present an error function `kerr' which does not have the objectionable properties exhibited by the error function `ferr' usually used to train neural networks. When combined with the conjugate gradient method, `kerr' shows dramatic speedups over `ferr' and back-propagation. Ferr is a sum of squares; in kerr each square is divided by a factor View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Pre-segmented handwritten digit recognition using neural networks

    Publication Year: 1991
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (52 KB)

    Summary form only given. Results of current research suggest that multilayer neural networks with local `receptive fields' and shared weights can be applied successfully to presegmented handwritten digit recognition. It was demonstrated that handwritten digit recognition without segmentation problem is actually quite simple; even traditional techniques such as the k nearest neighbor (KNN)... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Neural recognition of cursive handwriting

    Publication Year: 1991
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (52 KB)

    Summary form only given, as follows. A project which aims at the recognition of dynamically acquired cursive handwriting by means of concepts of neural computation is being carried out. Preliminary recognition results were obtained with a computational architecture (Scriptnet) that combines self-organization with optimization View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Genetic generation of both the weights and architecture for a neural network

    Publication Year: 1991, Page(s):397 - 404 vol.2
    Cited by:  Papers (48)  |  Patents (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (568 KB)

    Shows how to find both the weights and architecture for a neural network, including the number of layers, the number of processing elements per layer, and the connectivity between processing elements. This is accomplished by using a recently developed extension to the genetic algorithm which genetically breeds a population of LISP symbolic expressions of varying size and shape until the desired pe... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A rigorous result about the off-line learning approximation

    Publication Year: 1991
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (52 KB)

    Summary form only given, as follows. The authors consider a mathematical justification of the offline approximation in continuous-time neural networks. In real-time models, a network behavior is characterized by two distinct dynamics evolving according to different time scales, the weight dynamics which is the `slow' dynamics and the activation dynamics which is the `fast' dynamics. The offline ap... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Synchronized oscillations during cooperative feature linking in visual cortex

    Publication Year: 1991, Page(s):249 - 254 vol.2
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (432 KB)

    A neural network model of synchronized oscillations in the visual cortex is presented to account for recent neurophysiological findings that such synchronization may reflect global properties of the stimulus. In these experiments, synchronization of oscillatory firing responses to moving bar stimuli occurred not only for nearby neurons, but also occurred between neurons separated by several cortic... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Recurrent and feedforward backpropagation for time independent pattern recognition

    Publication Year: 1991, Page(s):591 - 596 vol.2
    Cited by:  Papers (3)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (348 KB)

    Based on a unified description of neural algorithms for time-independent pattern recognition, the authors discuss the generalization ability of a three-layer perceptron for recurrent backpropagation depending on the number of learning epochs and the steepness of the neurons' threshold function. Moreover, the suitability of recurrent and feedforward backpropagation learning algorithms for implement... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • An autonomous digital neural network architecture for segmenting hand-printed characters into visually pleasing and low variability sub-classes

    Publication Year: 1991
    Cited by:  Papers (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (76 KB)

    Summary form only given. A novel self-organizing neural network architecture has been developed which segments examples of hand-printed characters into subclusters of limited variability. This reduction in variability aids the recognition task of the network, and results obtained not only show a promising improvement over existing recognition techniques, but also indicate better performance than f... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Learning of category boundaries based on inverse recall by multilayer neural network

    Publication Year: 1991, Page(s):7 - 12 vol.2
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (428 KB)

    The author describes a method for inverse recall by a multilayer neural network with constraints. He also describes a query learning algorithm using the inverse recall for learning category boundaries in a pattern space. The method considered finds patterns near category boundaries and automatically produces a graded set of patterns between the adjacent categories. Seed patterns are selected based... View full abstract»

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

    Publication Year: 1991, Page(s):77 - 82 vol.2
    Cited by:  Papers (6)  |  Patents (1)
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (424 KB)

    It is noted that many aspects of the problem of improving generalization in feedforward neural networks have not been studied in any depth. The authors address the importance of this problem and propose two techniques to improve generalizations; proper selection of the training ensemble, and a partitioned learning strategy. These techniques are applied to a complex 2D classification problem. They ... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Associative memory networks, fault-tolerance and coding theory

    Publication Year: 1991, Page(s):437 - 440 vol.2
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (220 KB)

    The binary-couplings neural network used as an associative memory is studied under the effects of synaptic damage. The model of damage is to flip synapses (Jij→-Jij) randomly with a probability λ. A simple model based on a network with constant stabilities is proposed that allows the calculation of the storage properties of the damaged networks. ... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A connectionist simulation of human word associations

    Publication Year: 1991
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (48 KB)

    Summary form only given, as follows. On the basis of the frequencies and co-occurrences of words in large bodies of text, autoassociative lexical nets which describe the connective weights between 371 words were computed. These nets were used to predict the free associative responses of the word association norms collected by Russel and Jenkins from college students. Different assumptions about th... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • An unsupervised hyperspheric multilayer feedforward neural network classifier

    Publication Year: 1991, Page(s):59 - 64 vol.2
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (348 KB)

    A neural network model intended for the classification of patterns into distinct categories is introduced. Arbitrarily accurate category formation in a predefined feature space is asymptotically achieved by means of an unsupervised learning algorithm. Two model variants, one under a category separability assumption and the other under a category probability density unimodality (and nonseparability... View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Speech synthesis of phonemic triplets through a neural network-controlled formant synthesizer

    Publication Year: 1991
    Request permission for commercial reuse | Click to expandAbstract | PDF file iconPDF (80 KB)

    Summary form only given. The problems associated with automatic speech synthesis are related, to a large extent, to the methods of controlling the mathematical models of the human vocal tract and its properties as they change with time during discourse. In formant synthesis, which is the most effective synthesis method, rules are applied to relate the incoming phonemic information to values of the... View full abstract»

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