By Topic

Neural Networks, 1998. Proceedings. Vth Brazilian Symposium on

Date 9-11 Dec. 1998

Filter Results

Displaying Results 1 - 25 of 48
  • Proceedings 5th Brazilian Symposium on Neural Networks (Cat. No.98EX209)

    Publication Year: 1998
    Save to Project icon | Request Permissions | PDF file iconPDF (219 KB)  
    Freely Available from IEEE
  • Index of authors

    Publication Year: 1998 , Page(s): 261
    Save to Project icon | Request Permissions | PDF file iconPDF (150 KB)  
    Freely Available from IEEE
  • Extracting rules from feedforward Boolean neural networks

    Publication Year: 1998 , Page(s): 61 - 66
    Cited by:  Papers (1)  |  Patents (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (180 KB)  

    A method to extract rules from a feedforward Boolean neural networks (BNN) is described. We argue that rule extraction in BNN is more feasible and more natural than in other artificial neural networks models. This technique allows the understanding of how the neural networks reach a solution of a problem. A straight forward application of rule extraction from neural networks is the design of expert systems View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • RBF neural networks and MTI for text independent speaker identification

    Publication Year: 1998 , Page(s): 124 - 129
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (224 KB)  

    Artificial neural networks applied to speaker recognition tasks have being addressed by several researchers. This paper presents an investigation of the use of radial basis function (RBF) neural networks as classifiers applied to speaker identification tasks. A novel way to organize the speech frames in order to represent the speakers-the minimal temporal information (MTI)-is introduced and a comparison with the traditional multilayer perceptron (MLP) is presented. The results obtained indicate that the use of RBF neural networks are promising in speaker recognition and the MTI strategy to organize the speech frames are able to improve the RBF recognition rate View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A self-organizing map model for analysis of musical time series

    Publication Year: 1998 , Page(s): 140 - 145
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (264 KB)  

    Proposes a representation for unvoiced musical sequences, and tests experimentally our hierarchical artificial neural model on a musical time series-the third voice of the sixteenth four-part fugue in G minor of the Well-Tempered Clavier (vol. I) of J.S. Bach. The results obtained suggest that the model can perform efficiently on both recognition and discrimination of real musical sequences. It could recognize instances of a referential sequence-the theme of the fugue-in the presence of noise, and could also discriminate those instances out from the entire music View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Integration of hierarchical censored production rule (HCPR)-based system and neural networks

    Publication Year: 1998 , Page(s): 73 - 78
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (48 KB)  

    Over the past few years, researchers have successfully developed a number of systems that combine the strength of the symbolic and connectionist approaches to artificial intelligence. Most of the efforts have employed standard production rules, IF⟨condition⟩ THEN ⟨action⟩ as underlying symbolic representation. This paper is an attempt towards integrating hierarchical censored production rule based system and neural networks. A HCPR has the form: decision (if, precondition) (unless, censor conditions) (generality, general information) (specificity, specific information) which can be made to exhibit variable precision in the reasoning such that both certainty of belief in a conclusion and its specificity may be controlled by the reasoning process. The proposed hybrid system would have numerous applications where decision must be taken in real time and with uncertain information View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A neural-network based approach for recognition of pose and motion gestures on a mobile robot

    Publication Year: 1998 , Page(s): 79 - 84
    Cited by:  Papers (1)  |  Patents (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (268 KB)  

    Since a variety of changes in both robotic hardware and software suggests that service robots will soon become possible, to find “natural” ways of communication between human and robots is of fundamental importance for the robotic field. The paper describes a gesture-based interface for human-robot interaction, which enables people to instruct robots through easy-to-perform arm gestures. Such gestures might be static pose gestures, which involve only a specific configuration of the person's arm, or they might be dynamic motion gestures, that is, they involve motion (such as waving). Gestures are recognized in real-time at approximate frame rate, using neural networks. A fast, color-based tracking algorithm enables the robot to track and follow a person reliably through office environments with drastically changing lighting conditions. Results are reported in the context of an interactive clean-up task, where a person guides the robot to specific locations that need to be cleaned, and the robot picks up trash which it then delivers to the nearest trash-bin View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • An adaptive neural fuzzy network model for seasonal stream flow forecasting

    Publication Year: 1998 , Page(s): 215 - 219
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (128 KB)  

    This paper presents an adaptive neural fuzzy network model for seasonal stream flow forecasting. The model is based on a constructive learning method that adds neurons to the network structure whenever new knowledge is necessary so that it learns the fuzzy rules and membership functions essential for modeling a fuzzy system. The model was implemented to forecast monthly average inflow on an one-step-ahead basis. It was tested on three hydroelectric plants located in different river basins in Brazil. When the results were compared with those of a multilayer feedforward neural network model, the present model revealed at least a 50% decrease in the forecasting error View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Target recognition using evolutionary neural networks

    Publication Year: 1998 , Page(s): 226 - 231
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (160 KB)  

    The work presented here is part of the SAPRI Project-System for Acquisition, Processing and Recognition of Images-which is being developed for the Brazilian Naval Force. This project aims the treatment of radar images, supplying the specification of an environment, which can provide safety sailing and traffic control. This system involves areas like neural networks, image processing and pattern recognition. This paper presents some preliminary results achieved by using an evolutionary approach to optimize MLP network architectures for the target recognition task, using different learning algorithms View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Improving reinforcement learning control via online bilinear action interpolation

    Publication Year: 1998 , Page(s): 102 - 105
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (80 KB)  

    Reinforcement learning has been used as a reasonably successful method for the problem of model-free learning of action policies for some control problems. However, it is usually assumed that the process to be controlled is either open loop stable or of slow dynamics, when frequency of failures before acceptable performance or input-output processing time are not issues of primary importance. We consider the problem of model-free regulation for an unstable plant. As in many cases the need for state quantisation is an algorithmic storage requirement rather than a sensor limitation, we propose a modification of a standard reinforcement learning method that uses as additional information the distance between sampled and represented states, embedded in actions that are a result of a distance-wise local interpolation scheme. We obtained faster learning under minimal disturbance of the original learning scheme, and the modification is computationally modest enough to allow for real-time implementation View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A realistic computer simulation of primary somatosensory cortex replicating static properties of topographic organization

    Publication Year: 1998 , Page(s): 169 - 173
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (308 KB)  

    A model of the somatosensory system has been constructed with the neuro-simulator GENESIS as a network of 1,024 pyramidal cells and 512 baskets cells connected to 512 tactile receptors representing the hand surface reproducing processes of formation and maintenance of somatotopic maps. The model presents results such as variability in the shapes and sizes of the areas of cortical representation, values of cortical magnifications in agreement with experimental findings and linear decay of receptive fields overlap as a function of cortical distance between recording sites in normal conditions View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Feedforward neural network initialization: an evolutionary approach

    Publication Year: 1998 , Page(s): 43 - 48
    Cited by:  Papers (2)  |  Patents (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (76 KB)  

    The initial set of weights to be used in supervised learning for multilayer neural networks has a strong influence in the learning speed and in the quality of the solution obtained after convergence. An inadequate initial choice of the weight values may cause the training process to get stuck in a poor local minimum or to face abnormal numerical problems. There are several proposed techniques that try to avoid both local minima and numerical instability, only by means of a proper definition of the initial set of weights. This paper focuses on the application of genetic algorithms (GA) as a tool to analyze the space of weights, in order to achieve good initial conditions for supervised learning. GAs almost-global sampling compliments connectionist local search techniques well, and allows one to find some very important characteristics in the initial set of weights for multilayer networks. The results presented are compared, for a set of benchmarks, with that produced by other approaches found in the literature View full abstract»

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

    Publication Year: 1998 , Page(s): 130 - 134
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (72 KB)  

    This paper presents a system for calibration of the involved parameters in the classification and characterization process of fuzzy concepts that use t-norm differentiable and related membership families for the definition of characterizing functions. An empirical evaluation using the data set iris shows the comparability with the classical methods of classification. We also make a conjecture that relate any continuous membership function with the defined families View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • An isolated word speech recognition system based on Kohonen neural network

    Publication Year: 1998 , Page(s): 151 - 156
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (92 KB)  

    Describes a non-uniform segmentation algorithm based on a Kohonen neural network applied to an isolated word recognition system. This procedure realizes a temporal normalization and the resulting fixed number of acoustic vectors is then submitted to a multilayer perceptron network in order to recognize the spoken words View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A neo-fuzzy-neuron with real time training applied to flux observer for an induction motor

    Publication Year: 1998 , Page(s): 67 - 72
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (328 KB)  

    Presents an alternative algorithm for induction machines rotor flux observation. The novel procedure is based on a neo-fuzzy-neuron (NFN) with real time training. The main characteristics of this novel observer are: quick and accurate convergence and adaptability to system dynamics, requiring only the stator current measurements. The fuzzy-neural network employed here does not require previous training. The NFN is described, as well as its application to a rotor flux observer of a three-phase induction machine. Network training and observer performance are assessed by simulations and experimental results View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Evaluation of neural classifiers using statistic methods for identification of laryngeal pathologies

    Publication Year: 1998 , Page(s): 220 - 225
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (56 KB)  

    The use of statistical elements, like nonparametric tests and principal components analysis, allows the evaluation of the behavior and the performance of artificial neural networks when acoustical measurements are used to identify larynx diseases from which patterns are naturally overlapped. In this work, techniques to improve the results of neural network through the manipulation of the training patterns and convergence control will be discussed View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Improving backpropagation with sliding mode control

    Publication Year: 1998 , Page(s): 8 - 13
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (184 KB)  

    Sliding mode control is applied as a procedure to adapt weights of a multilayer perceptron. Standard backpropagation weight update equations are used for providing error estimates for the output and hidden layers, similarly to the classical algorithm. The sliding mode procedures are then introduced to adapt weights taking into consideration the standard backpropagation errors. As demonstrated throughout this paper, the introduction of sliding mode has resulted in a much faster version of the standard backpropagation. The speed-up achieved is around two times the standard version View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Design of radial basis function network as classifier in face recognition using eigenfaces

    Publication Year: 1998 , Page(s): 118 - 123
    Cited by:  Papers (5)  |  Patents (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (76 KB)  

    In this paper we investigate alternative designs of a radial basis function network acting as classifier in a face recognition system. The inputs to the RBF network are the projections of a face image over the principal components. A database of 250 facial images of 25 persons is used for training and evaluation. Two RBF designs are studied: the forward selection and the Gaussian mixture model. Both designs are also compared to the conventional Euclidean and Mahalanobis classifiers. A set of experiments evaluates the recognition rate of each method as a function of the number of principal components used to characterize the image samples. The results of the experiments indicate that the Gaussian mixture model RBF achieves the best performance while allowing less neurons in the hidden layer. The Gaussian mixture model approach shows also to be less sensitive to the choice of the training set View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Self-organizing modeling in forecasting daily river flows

    Publication Year: 1998 , Page(s): 210 - 214
    Cited by:  Papers (5)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (100 KB)  

    In a new approach, which corresponds in a better way to the actions of human nervous system, the connections between several neurons are not fixed but change in dependence on the neurons themselves. This article presents a GMDH (group method of data handling) algorithm with active neurons. These neurons are able, during the learning or self-organizing process, to estimate which inputs are important to minimize the given objective function of the neuron. The nonlinear GMDH model approach is shown to provide better representation of the daily average water inflow forecasting, than the models based on Box-Jenkins method, currently in use in the Brazilian Electrical Sector View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Artificial neural network applied to power system protection

    Publication Year: 1998 , Page(s): 247 - 252
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (88 KB)  

    The main objective of this paper is the implementation of an alternative protection model to transmission lines applying artificial neural networks (ANN). An improvement in performance to the conventional distance relay is expected, once the ANNs can learn the different fault conditions as well as network changes in order to operate in less time correctly. In this work, the relay protection zone (96% of a transmission line length) was determined by forward and reverse single-line-to-ground fault condition. The input data shows the trip/no trip decision of a protection system. The approach used in this paper utilizes the voltage and current post-fault samples as input to a moving data window. The implemented neural network should capture the knowledge for the correct relay operation facing the different network conditions View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Unsupervised neural network learning for blind sources separation

    Publication Year: 1998 , Page(s): 30 - 38
    Cited by:  Papers (1)  |  Patents (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (2336 KB)  

    Review of independent component analyses (ICA) and blind sources separation (BSS) employing in terms of unsupervised neural networks technology are given. For example, imagery features occurring in human visual systems are the continuing reduction of redundancy towards the “sparse edge maps”. When edges are multiplying together as the vector inner product they result in almost zero, namely pseudo-orthogonal ICA. This fact has been derived from the first principle of artificial neural networks using the maximum entropy information-theoretical formalism by Bell and Sejnowski (1996). We explore the blind de-mixing condition for more than two objects using two sensor measurement. We design two smart cameras with short term working memory to do better image de-mixing of more than two objects. We consider channel communication application that we can efficiently mix four images using matrices [A0] and [A1] to send through two channels View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Optimization of the combinatorial neural model

    Publication Year: 1998 , Page(s): 49 - 54
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (60 KB)  

    We present significant optimization of the so-called combinatorial neural model (CNM). CNM is a hybrid (neural/symbolic) model that has been used in areas such as expert system development and data mining. The paper first explains the CNM architecture and then presents CNM optimization together with empiric results. The most important optimization aims at taming combinatorial explosion, which is the main problem inherent to this model View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Some results on activation and scaling of sparse distributed memory

    Publication Year: 1998 , Page(s): 157 - 160
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (160 KB)  

    We consider two aspects on the efficiency of Kanerva's sparse distributed memory (SDM). First, it has been suggested that in certain situations it would make sense to use different activation probabilities for writing and reading in SDM. However, here we model such a situation and find that, at least approximately, it is optimal to use the same probabilities for writing and reading. Second, and more important, we investigate the scaling up of SDM, in connection with some observations made by Sjodin (1997). It is shown that the original SDM (here in Jaeckel's version) does not scale up if the reading address is disturbed, but that this can be remedied by using a kind of SDM with sparse address vectors, showing that SDM could well be used as a clean-up memory in computing with large patterns View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Training linear neural network with early stopped learning and ridge estimation

    Publication Year: 1998 , Page(s): 14 - 19
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (1100 KB)  

    This paper addresses the problem of supervised learning in layered neural network with linear units and includes an analysis of the effect of noise on training algorithms. We survey most of the known results on linear networks. The connections to classical statistical ideas such as ordinary least squares are emphasized View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Context and scale influencing clustering through unsupervised neural networks

    Publication Year: 1998 , Page(s): 235 - 240
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (988 KB)  

    This work aims at proposing a neural network model for clustering in which no information about the desired output is given, and influences due to context and scale are considered. Two models of unsupervised neural networks are described. The first is a winner-take-all (WTA) algorithm with pre-established lateral inhibition, the second is a model with Hebbian and anti-Hebbian learning. Both models have the same architecture but the second one has adaptable lateral inhibitory links. The proposed models are used in two different domains: classification of the iris and classification of animals. In the first, the patterns are formed by continuous inputs, while in the second, the inputs are mainly binary. The proposed models are evaluated according to their capacity of generalization, ability to classify nonlinearly separable patterns and robustness when clustering noisy patterns View full abstract»

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