By Topic

Neural Networks, 2004. Proceedings. 2004 IEEE International Joint Conference on

Date 25-29 July 2004

Go

Filter Results

Displaying Results 1 - 25 of 184
  • An FPGA implementation of 1,024-neuron system for PAPR reduction of OFDM signal

    Page(s): 2625 - 2629 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (634 KB) |  | HTML iconHTML  

    The aim of this paper is to reduce computational complexity of the neural network for PAPR reduction of OFDM signal, and to implement the neural network including 1,024 neurons by FPGA for practical OFDM transmitter of the terrestrial digital broadcast. A couple of IDFTs reduce computational complexity of the neuron updating from O(N2) to O(N log N). This neural network is designed using VHDL for Xilinx FPGA device, XC2V6000, and 1,024-neuron system is implemented by less than 30% of resources of the device. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • RTD-based compact programmable gates

    Page(s): 2637 - 2640 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (606 KB) |  | HTML iconHTML  

    This work presents novel and extremely compact implementations of programmable gates on the basis of the multi-threshold threshold gate concept. The circuit consists of resonant tunnelling diodes (RTDs) and heterostructure field effect transistors (HFETs) and its operating principle is based on the controlled quenching of clocked series-connected RTDs. The proposed generic circuit topology is presented and the methodology to design specific programmable gates is introduced. A number of programmable gates are shown and their operation is validated. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Self organising neural place codes for vision based robot navigation

    Page(s): 2501 - 2506 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (774 KB) |  | HTML iconHTML  

    Autonomous robots must be able to navigate independently within an environment. In the animal brain, so-called place cells respond to the environment where the animal is. We present a model of place cells based on self-organising maps. The aim of this paper is to show how image invariance can improve the performance of the neural place codes and make the model more robust to noise. The paper also demonstrates that localisation can be learned without having a pre-defined map given to the robot by humans and that after training, a robot can localise itself within a learned environment. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Homomorphic processing system and ratio rule for color image enhancement

    Page(s): 2507 - 2511 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (692 KB) |  | HTML iconHTML  

    Homomorphic filter is an illumination-reflectance model that can be used to develop a frequency domain procedure for improving the appearance of an image by simultaneous gray-level range compression and contrast enhancement. Many previously reported methods on homomorphic filter for color images shows that the homomorphic filter consistently provides excellent dynamic range compression but is lacking final color rendition. We present a novel color image enhancement process to overcome this limitation. The color image enhancement process involved using a neural network algorithm, namely ratio rule, to pre-process and post-process the color image in the homomorphic system. This method improves the appearance of images as perceived by the human eye, and/or to render these images more suitable for computer analysis. That is, both color rendition and dynamic range compression are achieved using this method. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Daily peak temperature forecasting with Elman neural networks

    Page(s): 2765 - 2769 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (652 KB) |  | HTML iconHTML  

    This work presents a forecaster based on an Elman artificial neural network trained with resilient backpropagation algorithm for predicting the daily peak temperatures one day ahead. The available time series was recorded at Petrosino (TP), in the west coast of Sicily, Italy and it is composed by temperature (min and max values), the humidity (min and max values) and the rainfall value between January 1st, 1995 and May 14th, 2003. Performances and reliabilities of the proposed model were evaluated by a number of measures, comparing different neural models. Experimental results show very good prediction performances. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Clustering using neural networks and Kullback-Leibler divergency

    Page(s): 2813 - 2817 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (627 KB) |  | HTML iconHTML  

    In this work we develop a clustering algorithm based on Kullback-Leibler divergence as the dissimilarity measurement. That measure is used with an algorithm that uses the classical vector quantization with competitive neural networks to perform the clustering of spatially complex data sets. The algorithm is also presented as an alternative tool to obtain a model based on Gaussian mixture of complex data sets. The clustering algorithm is tested with several data sets generated artificially. All sets in the data set is also modelled with a Gaussian mixture using the proposed algorithm. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Applications of Clifford support vector machines and Clifford moments for classification

    Page(s): 3003 - 3008 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (663 KB) |  | HTML iconHTML  

    This paper introduces the Clifford support vector machines as a generalization of the real- and complex- valued support vector machines using the Clifford geometric algebra. In this framework we handle the design of kernels involving the Clifford or geometric product for linear and nonlinear classification. We present an interesting application to classify mechanical tools using the concept of Clifford moments which is invariant under affine transformations. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A SVCA model for the competition on artificial time series (CATS) benchmark

    Page(s): 2777 - 2782 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (733 KB) |  | HTML iconHTML  

    This paper predicts the 100 missing values in CATS Benchmark. The SVCA model is an autoregressive model in which the coefficients vary smoothly with time. The model is fitted to the first differences of the data by minimising the residual sum of squared, subject certain restrictions that enable the gaps left by the missing observations to be bridged. The path of each time-varying coefficient is described by a combination of a sine and cosine function. The latter are specified via their amplitudes, phases and periods. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Iris recognition by a rotation spreading neural network

    Page(s): 2589 - 2594 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (698 KB) |  | HTML iconHTML  

    We previously proposed a rotation spreading neural network (R-SAN net). This neural net can recognize the orientation of an object irrespective of its shape, and its shape irrespective of its orientation. The characteristic of the R-SAN net is to spread the information of axis orientation of the object on surrounding space by spreading weight that has similar tuning characteristics to axis orientation neurons in the parietal cortex. The R-SAN net is suitable for the shape and orientation recognition of a concentric circular pattern, because it uses polar conversion. Previously, the validity of this neural net for learning and recollection of human faces had been demonstrated by off-line processing. In the present study, we used iris images instead of face images for the input patterns of the neural network. We attempted real-time individual identification by the iris patterns. In the recognition experiment, the R-SAN net was able to simultaneously recognize the orientation and shape of iris images of learned persons. We suggest application of the R-SAN net as an iris recognition system. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Design and optimization of Amari neural fields for early auditory-visual integration

    Page(s): 2523 - 2528 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (824 KB) |  | HTML iconHTML  

    We introduce a computational model of sensor fusion based on the topographic representations of a "two-microphone and one camera" configuration. Our aim is to perform a robust multimodal attention-mechanism in artificial systems. In our approach, we consider neurophysiological findings to discuss the biological plausibility of the coding and extraction of spatial features, but also meet the demands and constraints of applications in the field of human-robot interaction. In contrast to the common technique of processing different modalities separately and finally combine multiple localization hypotheses, we integrate auditory and visual data on an early level. This can be considered as focusing the attention or controlling the gaze onto salient objects. Our computational model is inspired by findings about the inferior colliculus in the auditory pathway and the visual and multimodal sections of the superior colliculus. Accordingly it includes: a) an auditory map, based on interaural time delays, b) a visual map, based on spatio-temporal intensity difference and c) a bimodal map where multisensory response enhancement is performed and motor-commands can be derived. After introducing a modified Amari-neural field architecture in the bimodal model, we place emphasis on a novel method of evaluation and parameter-optimization based on biology-inspired specifications and real-world experiments. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Texture classification by support vector machines with kernels for higher-order Gabor filtering

    Page(s): 3009 - 3014 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (783 KB) |  | HTML iconHTML  

    A support vector machine (SVM), which employs a kernel corresponding to feature extraction of local higher order moment spectra (LHOMS) of an image, is introduced. In order to overcome the curse of dimensionality when utilizing LHOMS image features in conventional multi channel filtering, an inner product kernel of LHOMS is derived. In the experiments, the SVM with LHOMS kernel is applied to image texture classification. It is shown that it can efficiently utilize the higher order features, and that the classification ratio is improved due to the introduction of the Gaussian window function for a stable local feature extraction. Further, it is discussed that the kernels for higher-order moment spectra and higher-order moments in the same orders becomes identical, indicating the equivalence of the two types of features in the kernel-function level. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Image enlargement as an edge estimation

    Page(s): 2577 - 2581 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (1051 KB) |  | HTML iconHTML  

    A robust image enlargement algorithm is presented in this paper. We formulate the image enlargement process as an edge information estimation process. In order to achieve a higher resolution, we first perform pixel duplication [W.K. Pratt, 1991] on the target image to form an initial high resolution image. Then the edge details of the enlarged image are estimated by using a novel neural network called "agent swarm regression network ASRN", which is trained by a set of low resolution (LR) / high resolution (HR) image patch pairs. Two benchmark images were used to verify the performance of the proposed algorithm. The results show that the enlarged images by the proposed algorithm are sharper than those by the conventional methods. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Hippocampal theta phase coding for instantaneous acquisition of experienced events

    Page(s): 3053 - 3058 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (690 KB) |  | HTML iconHTML  

    Theta rhythm dependent activity of rat hippocampal cells "theta phase precession" was elucidated based on the hypothesis that theta phase coding enables instantaneous acquisition of experienced events. By using a neural network model of theta phase coding we demonstrate high acquisition abilities of spatial and temporal events. A theoretical prerequisite for this computational power predicts a hippocampal-entorhinal network mechanism to regulate theta phase coding. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Supervisory level neural network identifier for a small power system with a STATCOM and a generator

    Page(s): 2901 - 2906 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (650 KB) |  | HTML iconHTML  

    A neural network based identifier is designed for effective control of a small power system. The power network in this work is considered from an external point of view, i.e., from a. supervisory level. Such a neuroidentifier can serve as a general model of such a plant, and then used for different neural network based control schemes. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Learning multiple correct classifications from incomplete data using weakened implicit negatives

    Page(s): 2953 - 2957 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (683 KB) |  | HTML iconHTML  

    Classification problems with output class overlap create problems for standard neural network approaches. We present a modification of a simple feedforward neural network that is capable of learning problems with output overlap, including problems exhibiting hierarchical class structures in the output. Our method of applying weakened implicit negatives to address overlap and ambiguity allows the algorithm to learn a large portion of the hierarchical structure from very incomplete data. Our results show an improvement of approximately 58% over a standard backpropagation network on the hierarchical problem. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Self-organizing map hardware accelerator system and its application to realtime image enlargement

    Page(s): 2683 - 2687 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (653 KB) |  | HTML iconHTML  

    We propose a new fast learning algorithm for SOM and its digital hardware design based on the massively parallel architecture. When this proposed algorithm is realized by using Xilinx XC2V6000-6 FPGA, a maximum performance of 17500 MCUPS is achieved and up to 256 competing units (16 × 16 map) can be implemented. Each competing unit have a weight vector which is represented by 128 elements of 16 bits accuracy. Furthermore, we applied the proposed hardware to a realtime digital image enlargement system. In the case of full color (24 bits) image enlargement from QQVGA (160 × 120 pixel) to QVGA (320 × 240 pixel), a proposed hardware requires only 0.12 second per image, while the personal computer (Intel XEON, 2.8 GHz Dual) requires more than 5 seconds per image. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Greedy network-growing by Minkowski distance functions

    Page(s): 2837 - 2842 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (747 KB) |  | HTML iconHTML  

    We propose a new network-growing method to accelerate learning and to extract explicit features in complex input patterns. We have so far proposed a new type of network-growing algorithm called greedy network growing algorithm (Kamimura, R, et al., 2002), (Kamimura, R, 2003). By this algorithm, a network can grow gradually by maximizing information on input patterns. In the algorithm, the inverse of the square of the ordinary Euclidean distance between input patterns and connection weights is used to produce competitive unit outputs. When applied to some problems, the method has shown slow learning, and sometimes the method cannot produce a state where information is large enough to produce explicit internal representations. To remedy this shortcoming, we introduce here Minkowski distance between input patterns and connection weights used to produce competitive unit outputs. When the parameter for Minkowski distance is larger, some detailed parts in input patterns can be eliminated, which enables networks to converge faster and to extract main parts of input patterns. We applied our new method to the analysis of some economic data. In the experiment, results confirm that a new method with Minkowski distance can significantly accelerate learning, and clearer features can be extracted. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Adaptive task decomposition and modular multilayer perceptrons for letter recognition

    Page(s): 2937 - 2942 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (771 KB) |  | HTML iconHTML  

    This paper proposes a task decomposition method, which divides a large-scale learning problem into multiple limited-scale pairs of training subsets and cross validation (CV) subsets. Correspondingly, modular multilayer perceptrons are set up. At first, one training subset only consists of its own class and several most neighboring categories, and then some classes in the CV subset are moved into it according to the generalization error of the module. This work presents an empirical formula for selecting the initial number of hidden nodes, and a method for determining the optimal number of hidden units with the help of singular value decomposition. The result for letter recognition shows that the above methods are quite effective. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Accuracy of joint entropy and mutual information estimates

    Page(s): 2843 - 2846 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (567 KB) |  | HTML iconHTML  

    In practice, researchers often face the problem of being able to collect only one, possibly large, dataset, and they are forced to make inferences from a single sample. Based on the results of the polarisation operator technique of Bowman et al (1969), we computed the dependence of joint entropy and mutual information estimates on the sample size in terms of asymptotic series. These expressions enabled us to control the bias of the estimates caused by finite sample sizes and obtain an expression for the accuracies. The result is important in data mining when joint entropy and mutual information are used to find interdependences within large data sets with unknown underlying structures. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Effect of noise on the performance of the temporally-sequenced intelligent block-matching and motion-segmentation algorithm

    Page(s): 2595 - 2600 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (764 KB) |  | HTML iconHTML  

    Most algorithms for motion-based segmentation depend on the system's ability to estimate optic flow from successive image frames. Block-matching is often used for this, but it faces the problems of noise-sensitivity and texture-insufficiency. Recently, we proposed a two-pathway approach based on locally coupled neural networks to address this issue. The system uses a pixel-level (P) pathway to perform robust block-matching in regions with sufficient texture, and a region-level (R) pathway to estimate motion from feature matching in low-texture regions. The fused optic-flow from the P and R pathways is then segmented by a pulse-coupled neural network (PCNN). The algorithm has produced very good results on synthetic and natural images. We show that its performance shows significant robustness to additive noise in the images. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Kernel-based canonical coordinate decomposition of two-channel nonlinear maps

    Page(s): 3019 - 3024 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (720 KB) |  | HTML iconHTML  

    A kernel-based formulation for decomposing nonlinear maps of two data channels into their canonical coordinates is derived. Each data channel is implicitly mapped to a high dimensional feature space defined by a nonlinear kernel. The canonical coordinates of the nonlinear maps are then found by transforming the kernel maps with the eigenvector matrices of a coupled asymmetric generalized eigenvalue problem. This generalized eigenvalue problem is constructed in the explicit space of kernel maps. The measures of linear dependence and coherence between the nonlinear maps of the channels are also presented. These measures may be determined in the kernel domain, without explicit computation of the nonlinear mappings. A numerical example is also presented. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Prototype optoelectronic Hamming neural network

    Page(s): 2659 - 2663 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (663 KB) |  | HTML iconHTML  

    We describe the hardware implementation of a Hamming classifier using an optoelectronic architecture. It is composed of two layers, the first layer is an optoelectronic matrix-vector multiplier based on the optical broadcast architecture; it is a novel architecture composed of a set of electronic neurons that receive the input sequentially by means of an optical broadcast interconnection. The second layer is an electronic winner take all circuit. The main characteristic of the system is that it is readily scalable in speed and size to large numbers of pixel neurons. We will describe the optoelectronic architecture, the hardware implementation of a prototype and evaluation of its performance characteristics. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Sparse coding and NMF

    Page(s): 2529 - 2533 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (906 KB) |  | HTML iconHTML  

    Non-negative matrix factorization (NMF) is a very efficient parameter-free method for decomposing multivariate data into strictly positive activations and basis vectors. However, the method is not suited for overcomplete representations, where usually sparse coding paradigms apply. We show how to merge the concepts of non-negative factorization with sparsity conditions. The result is a multiplicative algorithm that is comparable in efficiency to standard NMF, but that can be used to gain sensible solutions in the overcomplete cases. This is of interest e.g. for the case of learning and modeling of arrays of receptive fields arranged in a visual processing map, where an overcomplete representation is unavoidable. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • On the need for on-line learning in brain-computer interfaces

    Page(s): 2877 - 2882 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (757 KB) |  | HTML iconHTML  

    We motivate the need for on-line learning in brain-computer interfaces (BCI) and illustrate its benefits with the simplest method, namely fixed learning rates. However, the use of this method is supported by the risk of hampering the user to acquire suitable control of the BCI if the embedded classifier changes too rapidly. We report the results with 3 beginner subjects in a series of consecutive recordings, where the classifiers are iteratively trained with the data of a given session and tested on the next session. Interestingly, performance improved over sessions significantly for 2 of the subjects. These results show that on-line learning improves systematically the performance of the subjects. Moreover, performance with online learning is statistically similar to that obtained training the classifier off-line with the same amount of data. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A new accelerated EM based learning of the image parameters and restoration

    Page(s): 2513 - 2518 vol.4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (632 KB) |  | HTML iconHTML  

    We propose a new method based on the accelerated expectation maximization (EM) algorithm to learn the unknown image parameters and restoration. Acceleration is provided using fisher scoring (FS) optimization in the M step. Only a small number FS iteration is required for each M step. Our proposed algorithm reaches to the local minima in few steps whereas conventional EM needs more iteration. We also estimate the regularization parameter in the same single structure. Thanks to the FS optimization, it is possible to avoid complicated second derivative of the log-likelihood function by using only the gradient values. View full abstract»

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