Learning Vector Quantization

This paper presents the results of handwritten digit recognition on well-known image databases using state-of-the-art feature extraction and classification techniques. The tested databases are CENPARMI, CEDAR, and MNIST. On the test data set of each database, 80 recognition accuracies are given by combining eight classifiers with ten feature vectors. The features include chaincode feature, gradient feature, profile structure feature, and peripheral direction contributivity. The gradient feature is extracted from either binary image or gray-scale image. The classifiers include the k-nearest neighbor classifier, three neural classifiers, a learning vector quantization classifier, a discriminative learning quadratic discriminant function (DLQDF) classifier, and two support vector classifiers (SVCs). All the classifiers and feature vectors give high recognition accuracies. Relatively, the chaincode feature and the gradient feature show advantage over other features, and the profile structure feature shows efficiency as a complementary feature. The SVC with RBF kernel (SVC-rbf) gives the highest accuracy in most cases but is extremely expensive in storage and computation. Among the non-SV classifiers, the polynomial classifier and DLQDF give the highest accuracies. The results of non-SV classifiers are competitive to the best ones previously reported on the same databases.

La clasificación automática de las imágenes del Diagnóstico de Esparcimiento Thomson constituye uno de los ejes de la automatización del funcionamiento del TJ – II. En consecuencia el sistema de adquisición de datos del Diagnóstico fue sincronizado con el funcionamiento del TJ – II y un clasificador automático de imágenes desarrollado, funcionando este hace algunos años. Este trabajo se centra en el diseño y entrenamiento de nuevos clasificadores para las
imágenes del Diagnóstico de Esparcimiento Thomson. En el mismo se presenta un conjunto de clasificadores desarrollados para la clasificación de las imágenes del Diagnóstico. Los referidos clasificadores fueron diseñados para el funcionamiento en dos etapas, una de preprocesamiento de las imágenes, en la cual se utilizan las
transformadas wavelet, y otra de clasificación, en la cual se recurre a distintas técnicas de aprendizaje, las redes neuronales Learning Vector Quantization, las Máquinas de Vectores Soporte y template matching. Se presenta también una serie de experimentos realizados con los referidos clasificadores y con base en los resultados obtenidos se efectúa un análisis comparativo en cuanto a las potencialidades de los
mismos. Por otro lado, se presenta en el trabajo un servidor de clasificación desarrollado con base en uno de los clasificadores aquí presentados. El referido servidor se integra al sistema de adquisición de datos del Diagnóstico y permite efectuar la clasificación de las imágenes, a petición de un programa cliente, de forma sincronizada con el funcionamiento del TJ – II.
Palabras Claves: redes neuronales, Learning Vector Quantization, Máquina de Vectores Soporte, template matching, clasificador
_____________________________________________________________________
Automatic classification of Thomson Scattering Diagnostic’s images is among the central ideas about the TJ – II automation. Consequently, the Diagnostic’s data acquisition program has been synchronized with the TJ – II operation and an automatic image classifier implemented, and it has been operating for years. This work aims to design and train a set of new classifiers for the Thomson Scattering Diagnostic’s images. In this work a set of classifiers developed for the Diagnostic’s images are shown. They were designed to operate in two stages, firstly the preprocessing stage, in which the wavelet transform is used, and then the classification
stage, in which several learning techniques are used, as Learning Vector Quantization neural networks, Support Vector Machines and template matching. Also, a set of experiments carried out with the designed classifiers are shown and based on the tests performances a comparative analysis is made. On the other hand, in the work a classification server, which were developed using one of the showed classifiers, is presented. The server is integrated in the Diagnostic’s data
acquisition system and allows image classification, at request of a client program, in a synchronized way with TJ – II operation.
Keywords: neural networks, Learning Vector Quantization, Support vector Machines, template matching, classifier

In this work, we propose a two-stage classifier based on the analysis of the heart sound's complexity for murmur identification and classification. The first stage of the classifier verifies if the heart sound (HS) exhibits murmurs. To this end, the chaotic nature of the signal is assessed using the Lyapunov exponents (LEs). The second stage of the method is devoted to the classification of the type of murmur. In opposition to current state of the art methods for murmur classification, a reduced set of features is proposed. This set includes both well-known as well as new features designed to capture the morphological and the chaotic nature of murmurs. The classification scheme is evaluated with three classification methods: Learning Vector Quantization, Gaussian Mixture Models and Support Vector Machines. The achieved results are comparable to reported results in literature, while relying on a significant smaller set of features.

In this research, an iris recognition system was suggested based on five Artificial Neural Network (ANN) models separately: feed forward (FFBPNN), cascade forward (CFBPNN), function fitting (FitNet), pattern recognition (PatternNet) and learning vector quantization (LVQNet). For each ANN model, two architectures were constructed separately; 4 layers and 7 layers, each with different numbers of hidden layer units (5, 10 and 15). Ten different ANN optimization training algorithms (LM, BFG, BR, CGF, GD, GDM, GDA, GDX, OSS and RP) were used to train each model separately. Many experiments were conducted for each one of the five models. Each model used two different architectures, a different number of hidden layer neurons and ten different training algorithms. The performance results of the models were compared according to mean square error to identify the best ANN model. The results showed that the PatternNet model was the best model used. Finally, comparisons between the ten training...

This paper presents the architecture and VHDL design of a Two Dimensional Discrete Cosine Transform (2D-DCT) with Quantization and zigzag arrangement. This architecture is used as the core and path in JPEG image compression hardware. The 2D-DCT calculation is made using the 2D-DCT Separability property, such that the whole architecture is divided into two 1D-DCT calculations by using a transpose buffer. Architecture for Quantization and zigzag process is also described in this paper. The quantization process is done using division operation. This design aimed to be implemented in Spartan-3E XC3S500 FPGA. The 2D-DCT architecture uses 1891 Slices, 51I/O pins, and 8 multipliers of one Xilinx Spartan-3E XC3S500E FPGA reaches an operating frequency of 101.35 MHz One input block with 8 x 8 elements of 8 bits each is processed in 6604 ns and pipeline latency is 140 clock cycles .

El presente artículo introduce una aproximación al problema de “Anomaly Intrusion Detection” basada en una combinación de algoritmos de “Machine Learning” (ML) supervisados y no supervisados. Los objetivos que se persiguen son: el modelar en forma efectiva el tráfico de una organización y el reducir en forma substancial el porcentaje de Falsos Positivos mientras se mantiene un nivel razonable de detección de anomalías. Se presenta una arquitectura basada en un conjunto de “Self-Organizing Maps” (SOM) para el modelado del tráfico y en el uso de “Linear Vector Quantization” (LVQ) para la clasificación definitiva de los paquetes de tráfico. Los algoritmos desarrollados usan Snort para el preprocesamiento del tráfico de red, y están pensados para ser un complemento de esta herramienta. Los resultados alcanzados hasta el momento muestran que se pueden lograr niveles aceptables de acierto en comparación con otras técnicas. Al final se plantean las conclusiones extraídas del trabajo y dire...

... Whitehead and Choate [48] proposed an evolutionary learning algorithm for training RBF networks. ... According to this approach, the development of LVQ models and reformulated RBF ... and essentially established a link between clustering and learning vector quantization [13], [25 ...

A new Learning Vector Quantization classifier is proposed. The algorithm relies on a new training scheme for labeled sample vectors in feature space. Since weight or prototype vectors are conditioned to a well-known sliding-mode approach with use of a cost function to be minimized in terms of weight updates, new algorithm is called Optimized Generalized Learning Vector Quantization (OGLVQ). Consequently, weights are then associated to the proximity measure employed by conventional Generalized Learning Vector Quantization. New algorithm and some well-known predecessors are designed and tested for comparison with synthetic and publicly available datasets. From the experimental results, it is observed that the new classifier achieves faster training and is more successful and robust in generalizing labeled test samples picked from datasets studied than the counterparts it is compared to.

We studied the potential benefit of using artificial neural networks (ANNs) for the diagnosis of thyroid function. We examined two types of ANN architecture and assessed their robustness in the face of diagnostic noise. The thyroid function data we used had previously been studied by multivariate statistical methods and a variety of pattern-recognition techniques. The total data set comprised 392 cases that had been classified according to both thyroid function and 19 clinical categories. All cases had a complete set of results of six laboratory tests (total thyroxine, free thyroxine, triiodothyronine, triiodothyronine uptake test, thyrotropin, and thyroxine-binding globulin). This data set was divided into subsets used for training the networks and for testing their performance; the test subsets contained various proportions of cases with diagnostic noise to mimic real-life diagnostic situations. The networks studied were a multilayer perceptron trained by back-propagation, and a l...

In this module, Learning Vector Quantization LVQ neural network is first time introduced as a classifier for Arabic handwriting. Classification has been performed in two different strategies, in first strategy, we use one classifier for all 53 Arabic Character Basic Shapes CBSs in training and testing phases, in second strategy we use three classifiers and three subsets of 53 Arabic CBSs, the three subsets of Arabic CBSs are; ascending CBSs, descending CBSs and embedded CBSs. Three training algorithms; OLVQ1, LVQ2 and LVQ3 were examined and OLVQ1 found as the best learning algorithm.

A new algorithm based on learning vector quantisation classifier is presented based on a modified proximity-measure, which enforces a predetermined correct classification level in training while using sliding-mode approach for stable variation in weight updates towards convergence. The proposed algorithm and some well-known counterparts are implemented by using Python libraries and compared in a task of text classification for document categorisation. Results reveal that the new classifier is a successful contender to those algorithms in terms of testing and training performances.

We present a new Generalized Learning Vector Quantization classi-fier called Optimally Generalized Learning Vector Quantization based on a novel weight-update rule for learning labeled samples. The algorithm attains stable pro-totype/weight vector dynamics in terms of estimated current and previous weights and their updates. Resulting weight update term is then related to the proximity measure used by Generalized Learning Vector Quantization classifiers. New algorithm and some major counterparts are tested and compared for synthetic and publicly available datasets. For both the datasets studied, it is seen that the new classifier outperforms its counterparts in training and testing with accuracy above 80% its counterparts and in robustness against model parameter varition.

This article describes the application of Multi-Layer Perceptron (MLP), Probabilistic Neural Network and Kohonen's Learning Vector Quantization to the problem of diagnosing Multiple Sclerosis. The classification information is obtained from brainstem trigeminal evoked potential. The performance of the neural networks based classifiers is compared with that of the human experts and the Bayes classifier. The ability of the MLP classifier to generalize is far better than that of the Bayes classifier. The efficiency of the neural network based classifiers in conjunction with several types of well-known evoked potential features, such as Fourier transform space, latency and temporal wave, is examined. Although a large clinical data base would be necessary, before this approach can be fully validated, the initial results are promising.

Proses simulasi yang dilakukan untuk menghasilkan sebuah sistem pengenalan
meliputi beberapa tahap, yaitu tahap pengolahan citra dan tahap pelatihan dan
pengenalan. Tahap pengolahan citra dimulai dari Gray Scale, Thresholding, segmentasi,
dan normalisasi. Pengolahan citra diperlukan untuk memudahkan pengolahan data
gambar sebelum masuk dalam tahap pelatihan. Kemudian tahap kedua adalah pelatihan
dan pengenalan. Pada tahap ini, metode LVQ mulai digunakan dan menentukan bobot,
target error, maxepoch, dan laju pelatihan (Learning rate). Data yang dijadikan sebagai
input adalah citra huruf yang dinormalisasi sehingga berukuran 20x20 dan berekstensi
bitmap (.bmp). Simulasi pengenalan tulisan ini dilakukan dalam beberapa tahapan
sehingga bisa dengan membuat pengenalan menjadi lebih tajam. Tolak ukur keberhasilan
sistem pengenalan tulisan tangan ini adalah dengan menghitung nilai Termination Error
Rate dan tingkat keakuratan dalam pengenalan tanda tangan. Dari simulasi ini diperoleh
struktur JST dengan jumlah nilai learning rate 0,003 nilai target error 0,00001 dan jumlah
epoch sebesar 10.000 karena dalam rentang epoch 1000 sampai 10000 perubahan epoch
tidak mempengaruhi kinerja sistem. Sistem yang terbentuk mampu mengenali citra yang
berisi huruf yang digunakan sebagai bobot dengan nilai keakuratan rata – rata sebesar
61,07% dan rata – rata keakuratan hasil pengenalan terhadap citra yang belum dilakukan
pembelajaran sebesar 48,17%.