The rapid-eye-movement (REM) sleep is one of the most important parts in overnight sleep. In this study, an automatic REM sleep staging rule is introduced. Compared with the traditional REM detection method, a distinct feature of this method is that it only requires two EOG signals and thus reduces the number of input signal channels significantly. We calculate the correlation coefficient series between two EOG signals. By representing such a series with a VQ coding method, several techniques are proposed to improve the classification rate. Experimental results are given to demonstrate the effectiveness of the proposed approach.

A Microsoft Windows program is designed to implement a Multi-Language Identification system based on formants estimation and vector quantization classifier with n-Gram and HMM. LPC is used here as an effective method for formants feature extraction of the speakers, and a new method for distance measure of VQ is also proposed.

A new ensemble method for cluster analysis is introduced, which can be interpreted in two different ways: As complexity-reducing preprocessing stage for hierarchical clustering and as combination procedure for several partitioning results. The basic idea is to locate and combine structurally stable cluster centers and/or prototypes. Random effects of the training set are reduced by repeatedly training on resampled sets (bootstrap samples). We discuss the algorithm both from a more theoretical and an applied point of view and demonstrate it on several data sets. (author's abstract) / Series: Working Papers SFB "Adaptive Information Systems and Modelling in Economics and Management Science"

Joint source-channel coding (JSCC) has emerged to be a major field of research recently. Channel optimized vector quantization (COVQ) is a simple feasible JSCC scheme introduced for communication over practical channels. In this work, we propose an iterative design algorithm, referred to as the iterative maximum a posteriori (MAP) decoded (IMD) algorithm, to improve COVQ systems. Based on this algorithm, we design a COVQ based on symbol MAP hard-decision demodulation that exploits the non-uniformity of the quantization indices probability distribution. The IMD design algorithm consists of a loop which starts by designing a COVQ, obtaining the index source distribution, updating the discrete memoryless channel (DMC) according to the achieved index distribution, and redesigning the COVQ. This loop stops when the point-to-point distortion is minimized. We consider memoryless Gaussian and Gauss-Markov sources transmitted over binary phase-shift keying modulated additive white Gaussian noise (AWGN) and Rayleigh fading channels. Our scheme, which is shown to have less encoding complexity than conventional COVQ and less encoding complexity and storage requirements than soft-decision demodulated (SDD) COVQ systems, is also shown to provide a notable signal-to-distortion ratio (SDR) gain over the conventional COVQ designed for hard-decision demodulated channels while sometimes matching or exceeding the SDD COVQ performance, especially for higher quantization dimensions and/or rates. In addition to our main result, we also propose another iterative algorithm to design SDD COVQ based on the notion of the JSCC error exponent. This system is shown to have some gain over classical SDD COVQ both in terms of the SDR and the exponent itself. / Thesis (Master, Mathematics & Statistics) -- Queen's University, 2008-08-29 17:58:52.329

No description available.

La presente tesis, tuvo como objetivo mostrar una estrategia a través de redes neuronales, para los vehículos del Cuerpo General de Bomberos Voluntarios del Perú (CGBVP) durante una emergencia en el distrito de Surco, contribuyendo a la fluidez vehicular de las unidades en situaciones de emergencia. A nivel mundial se puede apreciar que se han desarrollado diferentes estrategias o sistemas que apoyan a las unidades de emergencia. El desarrollo del sistema propuesto consiste en preparar los semáforos con anticipación al paso de una unidad. Para ello se consideraron dos tipos de datos, ubicación y dirección, con el fin de activar los semáforos tiempo antes que el vehículo llegue a la intersección. El presente estudio analizó la red Neuronal LVQ (Learning Vector Quantization) y 2 tipos de red Backpropagation con el fin de determinar cuál de ellas es la más adecuada para el caso propuesto. Finalmente a través de simulaciones se determinó la red Backpropagation [100 85 10] obtuvo mejores resultados, siendo el de regresión igual a 0.99 y presentando valores de error en un rango de 10^-5 o menores. El algoritmo por Backpropagation [100 85 10] demostró durante sus 3 simulaciones responder correctamente a los 3 escenarios planteados. Demostrando únicamente variaciones pequeñas durante las simulaciones pero ninguna superando valores aceptables de 0 o 1 lógico. The following thesis had as objective to show a strategy using neural networks to help vehicles of the fire fighter brigade in Peru (CGBVP) during emergencies on the district of Surco, helping with the response times of the unit on emergency situations. Worldwide can be seen that strategies or systems are being used to help lower the problems of traffic. The development of the proposed system consist on preparing the traffic lights previous the arrival of the unit to the intersection. For this 2 type of data is being considered, location and direction, in order to activate the lights time before the vehicle arrives to the intersection. The present study analyzed the LVQ (Learning Vector Quantization) and 2 types of backpropagation networks in order to determine which of them is the most fitting for the situation to handle. Finally, going through the simulations it was determined that the [100 85 10] backpropagation network had the best response, being the regression 0.99 and showing error on the range of 10^-5 or lowers. The algorithm by backpropagation [100 85 10] showed during the 3 simulations that works property on all 3 situations. It showed small variations on some of the simulations but nothing out of the acceptable values of a logic 1 or 0.

Submitted by Daniella Sodre (daniella.sodre@ufpe.br) on 2015-03-09T14:01:45Z No. of bitstreams: 2 Dissertacao Telmo Filho_DEFINITIVA.pdf: 781380 bytes, checksum: fb398deff6f8aa856428277eb3236020 (MD5) license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) / Made available in DSpace on 2015-03-09T14:01:45Z (GMT). No. of bitstreams: 2 Dissertacao Telmo Filho_DEFINITIVA.pdf: 781380 bytes, checksum: fb398deff6f8aa856428277eb3236020 (MD5) license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Previous issue date: 2013-02-27 / A Análise de Dados Simbólicos lida com tipos de dados complexos, capazes de modelar a variabilidade interna dos dados e dados imprecisos. Dados simbólicos intervalares surgem naturalmente de valores como variação de temperatura diária, pressão sanguínea, entre outros. Esta dissertação introduz um algoritmo de Learning Vector Quantization para dados simbólicos intervalares, que usa uma distância Euclidiana intervalar ponderada e generalizada para medir a distância entre instâncias de dados e protótipos. A distância proposta tem quatro casos especiais. O primeiro caso é a distância Euclidiana intervalar e tende a modelar classes e clusters com formas esféricas. O segundo caso é uma distância intervalar baseada em protótipos que modela subregiões não-esféricas e de tamanhos similares dentro das classes. O terceiro caso permite à distância lidar com subregiões não-esféricas e de tamanhos variados dentro das classes. O último caso permite à distância modelar classes desbalanceadas, compostas de subregiões de várias formas e tamanhos. Experimentos são feitos para avaliar os desempenhos do Learning Vector Quantization intervalar proposto, usando todos os quatro casos da distância proposta. Três conjuntos de dados intervalares sintéticos e um conjunto de dados intervalares reais são usados nesses experimentos e seus resultados mostram a utilidade de uma distância localmente ponderada.

Submitted by João Arthur Martins (joao.arthur@ufpe.br) on 2015-03-12T17:06:49Z No. of bitstreams: 2 Dissertacao Telmo Silva Filho.pdf: 781380 bytes, checksum: fb398deff6f8aa856428277eb3236020 (MD5) license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) / Approved for entry into archive by Daniella Sodre (daniella.sodre@ufpe.br) on 2015-03-13T13:23:59Z (GMT) No. of bitstreams: 2 Dissertacao Telmo Silva Filho.pdf: 781380 bytes, checksum: fb398deff6f8aa856428277eb3236020 (MD5) license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) / Made available in DSpace on 2015-03-13T13:23:59Z (GMT). No. of bitstreams: 2 Dissertacao Telmo Silva Filho.pdf: 781380 bytes, checksum: fb398deff6f8aa856428277eb3236020 (MD5) license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Previous issue date: 2013-02-27 / A Análise de Dados Simbólicos lida com tipos de dados complexos, capazes de modelar a variabilidade interna dos dados e dados imprecisos. Dados simbólicos intervalares surgem naturalmente de valores como variação de temperatura diária, pressão sanguínea, entre outros. Esta dissertação introduz um algoritmo de Learning Vector Quantization para dados simbólicos intervalares, que usa uma distância Euclidiana intervalar ponderada e generalizada para medir a distância entre instâncias de dados e protótipos. A distância proposta tem quatro casos especiais. O primeiro caso é a distância Euclidiana intervalar e tende a modelar classes e clusters com formas esféricas. O segundo caso é uma distância intervalar baseada em protótipos que modela subregiões não-esféricas e de tamanhos similares dentro das classes. O terceiro caso permite à distância lidar com subregiões não-esféricas e de tamanhos variados dentro das classes. O último caso permite à distância modelar classes desbalanceadas, compostas de subregiões de várias formas e tamanhos. Experimentos são feitos para avaliar os desempenhos do Learning Vector Quantization intervalar proposto, usando todos os quatro casos da distância proposta. Três conjuntos de dados intervalares sintéticos e um conjunto de dados intervalares reais são usados nesses experimentos e seus resultados mostram a utilidade de uma distância localmente ponderada.

Deep learning techniques have achieved profound success in many challenging real-world applications, including image recognition, speech recognition, and machine translation. This success has increased the demand for developing deep neural networks and more effective learning approaches. The aim of this thesis is to consider the problem of learning a neural network classifier and to propose a novel approach to solve this problem under the Information Bottleneck (IB) principle. Based on the IB principle, we associate with the classification problem a representation learning problem, which we call ``IB learning". A careful investigation shows there is an unconventional quantization problem that is closely related to IB learning. We formulate this problem and call it ``IB quantization". We show that IB learning is, in fact, equivalent to the IB quantization problem. The classical results in rate-distortion theory then suggest that IB learning can benefit from a vector quantization approach, namely, simultaneously learning the representations of multiple input objects. Such an approach assisted with some variational techniques, result in a novel learning framework that we call ``Aggregated Learning (AgrLearn)", for classification with neural network models. In this framework, several objects are jointly classified by a single neural network. In other words, AgrLearn can simultaneously optimize against multiple data samples which is different from standard neural networks. In this learning framework, two classes are introduced, ``deterministic AgrLearn (dAgrLearn)" and ``probabilistic AgrLearn (pAgrLearn)". We verify the effectiveness of this framework through extensive experiments on standard image recognition tasks. We show the performance of this framework over a real world natural language processing (NLP) task, sentiment analysis. We also compare the effectiveness of this framework with other available frameworks for the IB learning problem.

Purpose: This article provides an empirical identification of groups of marketing scholars who share common beliefs about the role of science and the logic of scientific discovery. Design: We use Topology Representing Network quantization to empirically identify classes of marketing researchers within a representative sample of marketing professors. Findings: We find six distinct classes of marketing scholars. They differ with regard to popularity (size) and productivity (levels of publication output). Comparing the sub-samples of German-speaking and US respondents shows cross-cultural differences. Value: The study enhances our understanding of the current scientific orientation(s) of marketing. It may help to motivate marketing scholars to ponder on their own positions and assist them in judging where they may belong. Future comparisons over time would give us indication about the future of the academic discipline of marketing.(author's abstract)