By Topic

Neural Networks for Signal Processing X, 2000. Proceedings of the 2000 IEEE Signal Processing Society Workshop

Date 11-13 Dec. 2000

Go

Filter Results

Displaying Results 1 - 25 of 51
  • Neural Networks for Signal Processing X [front matter]

    Publication Year: 2000 , Page(s): i - iv
    Save to Project icon | Request Permissions | PDF file iconPDF (144 KB)  
    Freely Available from IEEE
  • Index

    Publication Year: 2000 , Page(s): 945 - 947
    Save to Project icon | Request Permissions | PDF file iconPDF (95 KB)  
    Freely Available from IEEE
  • An empirical investigation of the user-parameters and performance of continuous PBIL algorithms [population-based incremental learning]

    Publication Year: 2000 , Page(s): 702 - 710 vol.2
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (428 KB)  

    Evolutionary algorithms (EAs) are powerful methods for solving optimization problems, inspired by natural systems and incorporating population-based searching. Although the implementation of EAs is in many cases quite straightforward, it almost always involves making choices which can be viewed as assumptions regarding the nature of the problem to be solved. In this paper, one such choice is examined: the setting of user-defined parameters in three simple algorithms for solving unconstrained continuous optimization problems. Thre results agree with the notion that these algorithms are often robust to parameter settings, but also reveal interesting relationships between the parameters View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A novel method for CFAR data fusion

    Publication Year: 2000 , Page(s): 711 - 720 vol.2
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (484 KB)  

    Detection systems with distributed sensors and data fusion are increasingly being used by surveillance systems. There has been a great deal of theoretical study into decentralized detection networks that are composed of similar independent sensors. To solve the resulting nonlinear system, an exhaustive search and some approximation methods are usually adopted. However, these often either cause the system to be insensitive to some parameters or they lead to suboptimal results. In this paper, a genetic algorithm is investigated in order to obtain optimal results on constant false alarm rate (CFAR) data fusion View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Neural computation approach for the maximum-likelihood sequence estimation of communications signal

    Publication Year: 2000 , Page(s): 721 - 728 vol.2
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (336 KB)  

    A novel detection approach for signals in digital communications is proposed in this paper by using the NNTCTG (neural network with transient chaos and time-varying gain) developed by the author (1997, 1998). The maximum-likelihood signal detection problem can be always described as a complex optimization problem with so many local optima that conventional Hopfield-type neural networks cannot be applied. To amend the drawbacks of Hopfield-type networks, the NNTCTG is used to search for globally optimal or near-optimal solutions of the optimization problems with lots of local optima, since it has richer and more flexible dynamics than conventional networks with only point attractors. We established a neuro-based detection model for digital communication signals and analyzed its working procedure in detail. Two simulation experiments were conducted to illustrate the validity and effectiveness of the proposed approach View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Optimal feature selection using information maximisation: case of biomedical data

    Publication Year: 2000 , Page(s): 841 - 850 vol.2
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (360 KB)  

    The hybrid information maximisation (HIM) algorithm is derived. This algorithm is based on maximising the mutual information (MI) between the input and output of a network using the infomax principle, and between outputs of different network modules using the Imax algorithm. These two folds enable reducing the redundancy in output units in addition to selecting higher order features from input units. We analyse the proposed algorithm and generalise the learning procedure of the Imax algorithm. We show that the proposed HIM algorithm provides a better representation of input compared to the original two algorithms when used separately. An example showing the power of the HIM algorithm in the analysis of EEG data is discussed View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Adaptive multidimensional spline neural network for digital equalization

    Publication Year: 2000 , Page(s): 729 - 735 vol.2
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (268 KB)  

    Presents a new neural architecture that is suitable for digital signal processing applications. The architecture, which is based on adaptable multidimensional activation functions, allows one to collect information from the previous network layer in aggregate form. In other words, the number of network connections (the structural complexity) can be very low with respect to the problem complexity. This fact, as experimentally demonstrated in this paper, improves the network's generalization capabilities and speeds up the convergence of the learning process. A specific learning algorithm is derived, and experimental results on channel equalization demonstrate the effectiveness of the proposed architecture View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Face recognition using a new distance metric

    Publication Year: 2000 , Page(s): 584 - 593 vol.2
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (468 KB)  

    Many classification techniques use a distance metric as a measure of the similarity between patterns, and their generalisation performance is often strongly related to the effectiveness of the measure. This paper introduces a distance metric based on the Mahalanobis distance function, which is statistically more reliable than some metrics but does not discard discriminating information, often regarded as “noise”. In addition, it may be computed quickly. This paper develops this metric and experimentally shows that it may be used in a classifier to give the lowest error rate (2.63%) as well as the best training and classification times for a face recognition task View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Recognition of defects in high voltage transmission lines using the acoustic signal of corona effect

    Publication Year: 2000 , Page(s): 869 - 875 vol.2
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (368 KB)  

    The paper deals with the analysis of the possible application of neural networks to the recognition of typical damage of UHV transmission lines. The acoustic signal generated as a result of corona effects is used as a damage symptom, as its intensity is usually increased after damage occurrence or after contamination of the surface of a conductor or an insulator string. The primary problem in the diagnostic process is the distinguishing between signals generated as results of damage and contamination. The problem is not solved by methods based on the RF signal interference or by the classical methods of acoustic signal analysis. The construction and verification of the assumed diagnostic model have been carried out by experimental studies in laboratory conditions, where typical damage and contamination of the transmission line elements have been simulated View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Networks for networks: Internet analysis using graphical statistical models

    Publication Year: 2000 , Page(s): 755 - 764 vol.2
    Cited by:  Papers (4)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (452 KB)  

    A novel graphical framework for statistical modeling of distributed computer networks is presented in this paper. The framework enables the inference of packet losses across internal links in the network based solely on external (end-to-end) measurements, which can be easily made at end systems without network cooperation. This inference problem is commonly referred to as network tomography. Our modeling and inference framework is based on probabilistic factor graphs (or Bayesian networks). A computationally efficient probability propagation (message passing) algorithm is developed for network inference that is capable of producing exact marginal distributions (as well as point estimates) of link-level network parameters. Simulation experiments demonstrate the potential of our new framework View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Arithmetic-unit and processor design for neural networks

    Publication Year: 2000 , Page(s): 935 - 944 vol.2
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (572 KB)  

    The last decade saw a proliferation of research into the design of neurocomputers, many of which did not get beyond the prototype stage. We argue that, on the whole, neurocomputers are no longer viable; like, say, database computers before them, their time has passed before they became a common reality. We consider the implementation of hardware neural networks, from the level of arithmetic to complete individual processors and parallel processors and show that currents trends in computer architecture are not supportive of a case for custom neurocomputers. We argue that in the future, neural network processing ought to be mostly restricted to general-purpose processors or to processors that have been designed for other applications. There are just one or two two exceptions to this View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Off-line signature verification using HMMs and cross-validation

    Publication Year: 2000 , Page(s): 859 - 868 vol.2
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (508 KB)  

    We propose an HMM-based approach for off-line signature verification. One of the novelty aspects of our method lies in the ability to dynamically and automatically derive the various author-dependent parameters, required to set an optimal decision rule for the verification process. In this context, the cross-validation principle is used to derive not only the best HMM models, but also an optimal acceptation/rejection decision threshold for each author. This leads to a high discrimination between actual authors and impostors in the context of random forgeries. To quantitatively evaluate the generalization capabilities of our approach, we considered two conceptually different experimental tests carried out on two sets of 40 and 60 authors respectively, each author providing 40 signatures. The results obtained on these two sets show the robustness of our approach View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Support vector machines for speaker verification and identification

    Publication Year: 2000 , Page(s): 775 - 784 vol.2
    Cited by:  Papers (34)  |  Patents (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (460 KB)  

    The performance of the support vector machine (SVM) on a speaker verification task is assessed. Since speaker verification requires binary decisions, support vector machines seem to be a promising candidate to perform the task. A new technique for normalising the polynomial kernel is developed and used to achieve performance comparable to other classifiers on the YOHO database. We also present results on a speaker identification task View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Automatic evaluation of adaptive algorithms over the Internet

    Publication Year: 2000 , Page(s): 886 - 895 vol.2
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (716 KB)  

    We describe a system for the automatic evaluation of adaptive algorithms over the Internet. We motivate our discussion by setting up a system to evaluate region of interest finding algorithms. This is a signal processing/pattern recognition problem that occurs in many different application domains, and for which neural network solutions have been proposed as highly flexible and trainable solutions. However, we are not aware of any comprehensive empirical studies identifying the strengths and weaknesses of each method. The system we describe is able to automate such a study, and has the potential to be of great benefit to algorithm developers and the research community as a whole View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Alertness monitor using neural networks for EEG analysis

    Publication Year: 2000 , Page(s): 814 - 820 vol.2
    Cited by:  Papers (6)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (300 KB)  

    The goal is to detect the instance at which a person has lost the level of alertness necessary to assure safe operation of a vehicle or display vigilance. A neural network is proposed to detect the onset of this signal characteristic. The input to this neural network system is a modified feature vector composed of the associated wavelet representations at different scales. The output of the neural network is a binary decision as to whether the EEG represents an alert state or a drowsy state View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Face detection and eye localization by neural network based color segmentation

    Publication Year: 2000 , Page(s): 507 - 516 vol.2
    Cited by:  Papers (1)  |  Patents (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (632 KB)  

    This paper presents a neural network based scheme for human face detection and eye localization in color images under an unconstrained scene. A self-growing probabilistic decision-based neural network (SPDNN) is used to learn the conditional distribution for each color classes. Pixels of a color image are first classified into facial or non-facial regions, then pixels in the facial region are followed by eye region segmentation. The class of each pixel is determined by using the conditional distribution of the chrominance components of pixels belonging to each class. The paper demonstrates a successful application of SPDNN to face detection and eye localization on a database of 755 images from 151 persons. Regarding the performance, experimental results are elaborated. As to the processing speed, the face detection and eye localization processes consume approximately 560 ms on a Pentium-II personal computer View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Using statistical and neural network methods to explore the relationship between systematic risk and firm's long term investment activities

    Publication Year: 2000 , Page(s): 851 - 858 vol.2
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (328 KB)  

    The paper presents a neural network method to explore the relationship between the systematic risk and long term investing activities for Taiwan's companies in the fiber industry and electronics industry. In general, diversification from long-term investment may reduce a firm's systematic risk, but the empirical results in some literature are controversial. For years regression methods have been used to analyze the possible impact of long-term investment activities on systematic risk. Since their relationship may not be linear, we thus propose a neural network based sensitivity analysis for the possible non-linear relationships. We adapt five years of data (1994-1998) from the TEJ financial database to conduct the proposed analysis. The results show that the systematic risk is reduced with investment activities for the fiber industry. But for the electronics industry, the systematic risk is higher as firms increase the long-term investment ratio. The difference is even more significant when we only consider the companies with a higher portion of long term investment in assets. This diverse effect between industries may be one of the reasons why the empirical results are inconsistent. Our study can clarify the controversy between the systematic risk and long-term investment activities, and offer some possible explanations View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • The use of problem knowledge to improve the robustness of a fuzzy neural network

    Publication Year: 2000 , Page(s): 682 - 691 vol.2
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (536 KB)  

    Neural networks generally take a long time to train. This is because the network is initialized using random values for the weights. These random values have no relationship to the problem to be solved. The network is also more likely to converge to a non-optimal solution when initialized with random weights. This paper discusses how a fuzzy neural network can be initialized using problem knowledge. This initialization method improves the network robustness when training using uncertain data. It is shown that the use of problem knowledge-based rules can compensate for the uncertainty in the training data View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Nonlinear active noise control using Lyapunov theory and RBF network

    Publication Year: 2000 , Page(s): 916 - 925 vol.2
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (416 KB)  

    A new approach to design an efficient algorithm for the ANC system is proposed. The transversal filter-based controllers (FIR and IIR) are first considered. A Lyapunov function of the error is defined and filter coefficients are then adaptively adjusted based on Lyapunov stability theory so that the error converges to zero asymptotically. The design is independent of the statistical properties of signals and its computational complexity is comparable to FXLMS. It has fast error convergence properties and the stability is guaranteed by Lyapunov stability theory. This scheme can be further extended to an efficient nonlinear ANC using an RBF network for excellent performance. Simulation examples are demonstrated to show the degree of noise cancellation this scheme can achieve View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A recursive approach to joint image restoration and compensated blur identification

    Publication Year: 2000 , Page(s): 567 - 575 vol.2
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (380 KB)  

    Presents a new recursive scheme to blind image deconvolution based on joint image restoration and compensated blur identification. The technique projects a novel cost function into the image and blur subspaces, and optimizes them recursively using alternating minimization. A hierarchical neural network is employed to provide an adaptive, perception-based restoration. The sparse connections of the network are instrumental in reducing the computational cost of the restoration. On the other hand, conjugate gradient optimization is adopted to identify the blur due to its computational efficiency. A compensation scheme is developed to address the issue of ambiguous blur identification arising from the edge and hairy texture regions. Experimental results show that the new approach is effective and robust in restoring the degraded images as well as in identifying the blurs View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Multiuser demodulators using adaptive polynomial perceptrons in CDMA systems

    Publication Year: 2000 , Page(s): 746 - 754 vol.2
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (328 KB)  

    The polynomial perceptron multiuser demodulator (PPMUD) and the bilinear recursive polynomial perceptron multiuser demodulator with decision feedback (BRPMUD) are applied to a digital communication system using spread spectrum. The proposed multiuser demodulators are compared with the conventional receiver, the multilayer perceptron multiuser demodulator (MLPMUD) and the radial basis function multiuser demodulator (RBFMUD) in terms of bit-error rate (BER). In order to obtain a satisfactory BER, the structure of the PPMUD is complex and needs long training periods. On the other hand, the BRPMUD shows good performance and has a much simpler structure than the other multiuser demodulators using neural networks View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Support vector machine-based text detection in digital video

    Publication Year: 2000 , Page(s): 634 - 641 vol.2
    Cited by:  Papers (4)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (280 KB)  

    Textual data within video frames are very useful for describing the contents of the video frames, as they enable both keyword and free-text-based searching. In this paper, we pose the problem of text location in digital video as an example of supervised texture classification and use a support vector machine (SVM) as the texture classifier. Unlike other text detection methods, we do not incorporate any explicit texture feature extraction scheme. Instead, the gray-level values of the raw pixels are directly fed to the classifier. This is based on the observation that a SVM has the capability of learning in a high-dimensional space and of incorporating a feature extraction scheme in its own architecture. In comparison with a neural network-based text detection method, the SVM classifier illustrates the excellence of the proposed method View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Lateral inhibition mechanism in computational auditory model and its application in robust speech recognition

    Publication Year: 2000 , Page(s): 785 - 794 vol.2
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (580 KB)  

    In the auditory neural system, the lateral inhibition mechanism is very common, such as in the cochlear nucleus, auditory cortex, etc. The function of this lateral inhibition is to sharpen the contrast of the temporal and spatial structures, thus prominent features of stimulation in spatial and temporal domains can be enhanced. In this paper, a new mathematical model based on lateral inhibition is proposed. Traditional feature MFCC (Mel Frequency Cepstral Coefficient) and auditory feature AFCC (Auditory Frequency Cepstral Coefficient) are processed by this model, new features can be obtained such as MFCCI and AFCCI. Experiments with the new features with a HMM show that they can improve the robustness of the recognition rate in noisy conditions View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Echocardiographic image sequence segmentation using self-organizing maps

    Publication Year: 2000 , Page(s): 594 - 603 vol.2
    Cited by:  Patents (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (528 KB)  

    Presents a new approach for echocardiographic image sequence segmentation. The proposed method uses a self-organizing map to approximate the probability density function of the image patterns. The map is post-processed by the k-means clustering algorithm, in order to detect groups of neurons whose weights are similar. Each segmented image of the sequence is generated by correlation of its pixels and clusters found in the map. The best number of clusters is dependent on the application. To validate the segmentation procedure, we used a segmented sequence to successfully measure the variation of the interventricular septum width View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Effect of parallel ensembles to self-generating neural networks for chaotic time series prediction

    Publication Year: 2000 , Page(s): 896 - 905 vol.2
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (380 KB)  

    Self-generating neural networks (SGNNs) have the features of simplicity of network design and fast processing by automatically constructing a self-generating neural tree (SGNT) from a given training data set. Though the prediction accuracy of SGNNs for chaotic time series prediction is improved by adopting the ensemble averaging method, the computation time increases in proportion to the number of SGNNs in an ensemble. We investigate the improving capability of the prediction accuracy and the parallel efficiency of ensemble SGNNs (ESGNNs) for three chaotic time series prediction problems on a MIMD parallel computer. We allocate each SGNN to each processor. Our results show that the more the number of processors increases, the more the improvement of the prediction accuracy is obtained for all problems View full abstract»

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