Dissertations / Theses on the topic 'Cellular neural networks'

Les xarxes neuronals cel·lulars altrament anomenades CNNs, són un tipus de sistema dinàmic que relaciona diferents elements que s'anomenen neurones via unes plantilles de paràmetres. Aquest sistema queda completament determinat coneixent quines són les entrades a la xarxa, les sortides i els paràmetres o pesos. En aquest treball fem un estudi exhaustiu sobre aquest tipus de xarxa en el cas més senzill on només hi intervenen dues neurones. Tot i la simplicitat del sistema, veurem que pot tenir una dinàmica molt rica.

Primer de tot, revisem l'estabilitat d'aquest sistema des de dos punts de vista diferents. Usant la teoria de Lyapunov, trobem el rang de paràmetres en el que hem de treballar per aconseguir la convergència de la xarxa cap a un punt fix. Aquest mètode ens obre les portes per abordar els diferents tipus de problemes que es poden resoldre usant una xarxa neuronal cel·lular de dues neurones. D'altra banda, el comportament dinàmic de la CNN està determinat per la funció lineal a trossos que defineix les sortides del sistema. Això ens permet estudiar els diferents sistemes que apareixen en cada una de les regions on el sistema és lineal, aconseguint un estudi complet de l'estabilitat de la xarxa en funció de les posicions locals dels diferents punts d'equilibri del sistema. D'aquí obtenim bàsicament dos tipus de convergència, cap a un punt fix o bé cap a un cicle límit. Aquests resultats ens permeten organitzar aquest estudi bàsicament en aquests dos tipus de convergència. Entendre el sistema d'equacions diferencials que defineixen la CNN en dimensió 1 usant només dues neurones, ens permet trobar les dificultats intrínseques de les xarxes neuronals cel·lulars així com els possibles usos que els hi podem donar. A més, ens donarà les claus per a poder entendre el cas general.

Un dels primers problemes que abordem és la dependència de les sortides del sistema respecte les condicions inicials. La funció de Lyapunov que usem en l'estudi de l'estabilitat es pot veure com una quàdrica si la pensem com a funció de les sortides. La posició i la geometria d'aquesta forma quadràtica ens permeten trobar condicions sobre els paràmetres que descriuen el sistema dinàmic. Treballant en aquestes regions aconseguim abolir el problema de la dependència. A partir d'aquí ja comencem a estudiar les diferents aplicacions de les CNN treballant en un rang de paràmetres on el sistema convergeix a un punt fix. Una primera aplicació la trobem usant aquest tipus de xarxa per a reproduir distribucions de probabilitat tipus Bernoulli usant altre cop la funció de Lyapunov emprada en l'estudi de l'estabilitat. Una altra aplicació apareix quan ens centrem a treballar dins del quadrat unitat. En aquest cas, el sistema és capaç de reproduir funcions lineals.

L'existència de la funció de Lyapunov permet també de construir unes gràfiques que depenen dels paràmetres de la CNN que ens indiquen la relació que hi ha entre les entrades de la CNN i les sortides. Aquestes gràfiques ens donen un algoritme per a dissenyar plantilles de paràmetres reproduint aquestes relacions. També ens obren la porta a un nou problema: com composar diferents plantilles per aconseguir una determinada relació entrada¬sortida. Tot aquest estudi ens porta a pensar en buscar una relació funcional entre les entrades externes a la xarxa i les sortides. Com que les possibles sortides és un conjunt discret d'elements gràcies a la funció lineal a trossos, la correspondència entrada¬sortida es pot pensar com un problema de classificació on cada una de les classes està definida per les diferent possibles sortides. Pensant¬ho d'aquesta manera, estudiem quins problemes de classificació es poden resoldre usant una CNN de dues neurones i trobem quina relació hi ha entre els paràmetres de la CNN, les entrades i les sortides. Això ens permet trobar un mètode per a dissenyar plantilles per a cada problema concret de classificació. A més, els resultats obtinguts d'aquest estudi ens porten cap al problema de reproduir funcions Booleanes usant CNNs i ens mostren alguns dels límits que tenen les xarxes neuronals cel·lulars tot intentant reproduir el capçal de la màquina universal de Turing descoberta per Marvin Minsky l'any 1962.

A partir d'aquí comencem a estudiar la xarxa neuronal cel·lular quan convergeix cap a un cicle límit. Basat en un exemple particular extret del llibre de L.O Chua, estudiem primer com trobar cicles límit en el cas que els paràmetres de la CNN que connecten les diferents neurones siguin antisimètrics. D'aquesta manera trobem en quin rang de paràmetres hem de treballar per assegurar que l'estat final de la xarxa sigui una corba tancada. A més ens dona la base per poder abordar el problema en el cas general. El comportament periòdic d'aquestes corbes ens incita primer a calcular aquest període per cada cicle i després a pensar en possibles aplicacions com ara usar les CNNs per a generar senyals de rellotge.

Finalment, un cop estudiats els diferents tipus de comportament dinàmics i les seves possibles aplicacions, fem un estudi comparatiu de la xarxa neuronal cel·lular quan la sortida està definida per la funció lineal a trossos i quan està definida per la tangent hiperbòlica ja que moltes vegades en la literatura s'usa l'una en comptes de l'altra aprofitant la seva diferenciabilitat. Aquest estudi ens indica que no sempre es pot usar la tangent hiperbòlica en comptes de la funció lineal a trossos ja que la convergència del sistema és diferent en un segons com es defineixin les sortides de la CNN.
Les redes neuronales celulares o CNNs, son un tipo de sistema dinámico que relaciona diferentes elementos llamados neuronas a partir de unas plantillas de parámetros. Este sistema queda completamente determinado conociendo las entradas de la red, las salidas y los parámetros o pesos. En este trabajo hacemos un estudio exhaustivo de estos tipos de red en el caso más sencillo donde sólo intervienen dos neuronas. Este es un sistema muy sencillo que puede llegar a tener una dinámica muy rica.

Primero, revisamos la estabilidad de este sistema desde dos puntos de vista diferentes. Usando la teoría de Lyapunov, encontramos el rango de parámetros en el que hemos de trabajar para conseguir que la red converja hacia un punto fijo. Este método nos abre las puertas parar poder abordar los diferentes tipos de problemas que se pueden resolver usando una red neuronal celular de dos neuronas. Por otro lado, el comportamiento dinámico de la CNN está determinado por la función lineal a tramos que define las salidas del sistema. Esto nos permite estudiar los diferentes sistemas que aparecen en cada una de las regiones donde el sistema es lineal, consiguiendo un estudio completo de la estabilidad de la red en función de las posiciones locales de los diferentes puntos de equilibrio del sistema. Obtenemos básicamente dos tipos de convergencia, hacia a un punto fijo o hacia un ciclo límite. Estos resultados nos permiten organizar este estudio básicamente en estos dos tipos de convergencia. Entender el sistema de ecuaciones diferenciales que definen la CNN en dimensión 1 usando solamente dos neuronas, nos permite encontrar las dificultades intrínsecas de las redes neuronales celulares así como sus posibles usos. Además, nos va a dar los puntos clave para poder entender el caso general. Uno de los primeros problemas que abordamos es la dependencia de las salidas del sistema respecto de las condiciones iniciales. La función de Lyapunov que usamos en el estudio de la estabilidad es una cuadrica si la pensamos como función de las salidas. La posición y la geometría de esta forma cuadrática nos permiten encontrar condiciones sobre los parámetros que describen el sistema dinámico. Trabajando en estas regiones logramos resolver el problema de la dependencia. A partir de aquí ya podemos empezar a estudiar las diferentes aplicaciones de las CNNs trabajando en un rango de parámetros donde el sistema converge a un punto fijo. Una primera aplicación la encontramos usando este tipo de red para reproducir distribuciones de probabilidad tipo Bernoulli usando otra vez la función de Lyapunov usada en el estudio de la estabilidad. Otra aplicación aparece cuando nos centramos en trabajar dentro del cuadrado unidad. En este caso, el sistema es capaz de reproducir funciones lineales.

La existencia de la función de Lyapuno v permite también construir unas graficas que dependen de los parámetros de la CNN que nos indican la relación que hay entre las entradas de la CNN y las salidas. Estas graficas nos dan un algoritmo para diseñar plantillas de parámetros reproduciendo estas relaciones. También nos abren la puerta hacia un nuevo problema: como componer diferentes plantillas para conseguir una determinada relación entrada¬salida. Todo este estudio nos lleva a pensar en buscar una relación funcional entre las entradas externas a la red y las salidas. Teniendo en cuenta que las posibles salidas es un conjunto discreto de elementos gracias a la función lineal a tramos, la correspondencia entrada¬salida se puede pensar como un problema de clasificación donde cada una de las clases está definida por las diferentes posibles salidas. Pensándolo de esta forma, estudiamos qué problemas de clasificación se pueden resolver usando una CNN de dos neuronas y encontramos la relación que hay entre los parámetros de la CNN, las entradas y las salidas. Esto nos permite encontrar un método de diseño de plantillas para cada problema concreto de clasificación. Además, los resultados obtenidos en este estudio nos conducen hacia el problema de reproducir funciones Booleanas usando CNNs y nos muestran algunos de los límites que tienen las redes neuronales celulares al intentar reproducir el cabezal (la cabeza) de la máquina universal de Turing descubierta por Marvin Minsky el año 1962.

A partir de aquí empezamos a estudiar la red neuronal celular cuando ésta converge hacia un ciclo límite. Basándonos en un ejemplo particular sacado del libro de L.O Chua, estudiamos primero como encontrar ciclos límite en el caso que los parámetros de la CNN que conectan las diferentes neuronas sean anti¬simétricos. De esta forma encontramos el rango de parámetros en el cuál hemos de trabajar para asegurar que el estado final de la red sea una curva cerrada. Además nos da la base para poder abordar el problema en el caso general. El comportamiento periódico de estas curvas incita primero a calcular su periodo para cada ciclo y luego a pensar en posibles aplicaciones como por ejemplo usar las CNNs para generar señales de reloj.

Finalmente, estudiados ya los diferentes tipos de comportamiento dinámico y sus posibles aplicaciones, hacemos un estudio comparativo de la red neuronal celular cuando la salida está definida por la función lineal a trozos y cuando está definida por la tangente hiperbólica ya que muchas veces en la literatura se usa una en vez de la otra intentado aprovechar su diferenciabilidad. Este estudio nos indica que no siempre se puede intercambiar dichas funciones ya que la convergencia del sistema es distinta según como se definan las salidas de la CNN.
In this dissertation we review the two neuron cellular neural network stability using the Lyapunov theory, and using the different local dynamic behavior derived from the piecewise linear function use. We study then a geometrical way to understand the system dynamics. The Lyapunov stability, gives us the key point to tackle the different convergence problems that can be studied when the CNN system converges to a fixed¬point. The geometric stability shed light on the convergence to limit cycles. This work is basically organized based on these two convergence classes.

We try to make an exhaustive study about Cellular Neural Networks in order to find the intrinsic difficulties, and the possible uses of a CNN. Understanding the CNN system in a lower dimension, give us some of the main keys in order to understand the general case. That's why we will focus our study in the one dimensional CNN case with only two neurons.

From the results obtained using the Lyapunov function, we propose some methods to avoid the dependence on initial conditions problem. Its intrinsic characteristics as a quadratic form of the output values gives us the key points to find parameters where the final outputs do not depend on initial conditions. At this point, we are able to study different CNN applications for parameter range where the system converges to a fixed¬point. We start by using CNNs to reproduce Bernoulli probability distributions, based on the Lyapunov function geometry. Secondly, we reproduce linear functions while working inside the unit square.

The existence of the Lyapunov function allows us to construct a map, called convergence map, depending on the CNN parameters, which relates the CNN inputs with the final outputs. This map gives us a recipe to design templates performing some desired input¬output associations. The results obtained drive us into the template composition problem. We study the way different templates can be applied in sequence. From the results obtained in the template design problem, we may think on finding a functional relation between the external inputs and the final outputs. Because the set of final states is discrete, thanks to the piecewise linear function, this correspondence can be thought as a classification problem. Each one of the different classes is defined by the different final states which, will depend on the CNN parameters.

Next, we study which classifications problems can be solved by a two neuron CNN, and relate them with weight parameters. In this case, we also find a recipe to design templates performing these classification problems. The results obtained allow us to tackle the problem to realize Boolean functions using CNNs, and show us some CNN limits trying to reproduce the header of a universal Turing machine.

Based on a particular limit cycle example extracted from Chua's book, we start this study with anti symmetric connections between cells. The results obtained can be generalized for CNNs with opposite sign parameters. We have seen in the stability study that limit cycles have the possibility to exist for this parameter range. Periodic behavior of these curves is computed in a particular case. The limit cycle period can be expressed as a function of the CNN parameters, and can be used to generate clock signals.

Finally, we compare the CNN dynamic behavior using different output functions, hyperbolic tangent and piecewise linear function. Many times in the literature, hyperbolic tangent is used instead of piecewise linear function because of its differentiability along the plane. Nevertheless, in some particular regions in the parameter space, they exhibit a different number of equilibrium points. Then, for theoretical results, hyperbolic tangent should not be used instead of piecewise linear function.

Artificial neural networks are widely used for parallel processing of data analysis and visual information. The most prominent example of artificial neural networks is a cellular neural network (CNN), composed from two-dimensional arrays of simple first-order dynamical systems (“cells”) that are interconnected by wires. The information, to be processed by a CNN, represents the initial state of the network, and the parallel information processing is performed by converging to one of the stable spatial equilibrium states of the multi-stable CNN. This thesis studies a specific type of CNNs designed to perform the winner-take-all function of finding the largest among the n numbers, using the network dynamics. In a wider context, this amounts to automatically detecting a target spot in the given visual picture. The research, reported in this thesis, demonstrates that the addition of fast on-off switching (blinking) connections significantly improves the functionality of winner-take-all CNNs. Numerical calculations are performed to reveal the dependence of the probability, that the CNN correctly classifies the largest number, on the switching frequency.

La thèse propose une approche de l'apprentissage temporel par des mécanismes d'auto-organisation à grain fin. Le manuscrit situe dans un premier temps le travail dans la perspective de contribuer à promouvoir une informatique cellulaire. Il s'agit d'une informatique où les calculs se répartissent en un grand nombre de calculs élémentaires, exécutés en parallèle, échangeant de l'information entre eux. Le caractère cellulaire tient à ce qu'en plus d’être à grain fin, une telle architecture assure que les connexions entre calculateurs respectent une topologie spatiale, en accord avec les contraintes des évolutions technologiques futures des matériels. Dans le manuscrit, la plupart des architectures informatiques distribuées sont examinées suivant cette perspective, pour conclure que peu d'entre elles relèvent strictement du paradigme cellulaire.Nous nous sommes intéressé à la capacité d'apprentissage de ces architectures, du fait de l'importance de cette notion dans le domaine connexe des réseaux de neurones par exemple, sans oublier toutefois que les systèmes cellulaires sont par construction des systèmes complexes dynamiques. Cette composante dynamique incontournable a motivé notre focalisation sur l'apprentissage temporel, dont nous avons passé en revue les déclinaisons dans les domaines des réseaux de neurones supervisés et des cartes auto-organisatrices.Nous avons finalement proposé une architecture qui contribue à la promotion du calcul cellulaire en ce sens qu'elle exhibe des propriétés d'auto-organisation pour l'extraction de la représentation des états du système dynamique qui lui fournit ses entrées, même si ces dernières sont ambiguës et ne reflètent que partiellement cet état. Du fait de la présence d'un cluster pour nos simulations, nous avons pu mettre en œuvre une architecture complexe, et voir émerger des phénomènes nouveaux. Sur la base de ces résultats, nous développons une critique qui ouvre des perspectives sur la suite à donner à nos travaux
The thesis proposes a sequence learning approach that uses the mechanism of fine grain self-organization. The manuscript initially starts by situating this effort in the perspective of contributing to the promotion of cellular computing paradigm in computer science. Computation within this paradigm is divided into a large number of elementary calculations carried out in parallel by computing cells, with information exchange between them.In addition to their fine grain nature, the cellular nature of such architectures lies in the spatial topology of the connections between cells that complies with to the constraints of the technological evolution of hardware in the future. In the manuscript, most of the distributed architecture known in computer science are examined following this perspective, to find that very few of them fall within the cellular paradigm.We are interested in the learning capacity of these architectures, because of the importance of this notion in the related domain of neural networks for example, without forgetting, however, that cellular systems are complex dynamical systems by construction.This inevitable dynamical component has motivated our focus on the learning of temporal sequences, for which we reviewed the different models in the domains of neural networks and self-organization maps.At the end, we proposed an architecture that contributes to the promotion of cellular computing in the sense that it exhibits self-organization properties employed in the extraction of a representation of a dynamical system states that provides the architecture with its entries, even if the latter are ambiguous such that they partially reflect the system state. We profited from an existing supercomputer to simulate complex architecture, that indeed exhibited a new emergent behavior. Based on these results we pursued a critical study that sets the perspective for future work

Internal Symmetry Networks are a recently developed class of Cellular Neural Network inspired by the phenomenon of internal symmetry in quantum physics. Their hidden unit activations are acted on non-trivially by the dihedral group of symmetries of the square. Here, we extend Internal Symmetry Networks to include recurrent connections, and train them by backpropagation to perform a variety of image processing tasks, smoothing, sharpening, edge detection, synthetic image segmentation, texture segmentation and object recognition. By a large number of experiments, we find some guidelines to construct appropriate configurations of the net for different tasks.

Discovering cancer at an early stage is an effective way to increase the chance of survival. However, since most screening processes are done manually it is time inefficient and thus costly. One way of automizing the screening process could be to classify cells using Convolutional Neural Networks. Convolutional Neural Networks have been proven to produce high accuracy for image classification tasks. This thesis investigates if Convolutional Neural Networks can be used as a tool to detect cellular changes due to malignancy in the oral cavity and uterine cervix. Two datasets containing oral cells and two datasets containing cervical cells were used. The cells were divided into normal and abnormal cells for a binary classification. The performance was evaluated for two different network architectures, ResNet and VGG. For the oral datasets the accuracy varied between 78-82% correctly classified cells depending on the dataset and network. For the cervical datasets the accuracy varied between 84-86% correctly classified cells depending on the dataset and network. These results indicates a high potential for classifying abnormalities for oral and cervical cells. ResNet was shown to be the preferable network, with a higher accuracy and a smaller standard deviation.

Evolving from neuro-biological insights, neural network technology gives a computer system an amazing capacity to actually generate decisions dynamically. However, as the amount of data to be processed increases, there is a demand for developing new types of networks such as Cellular Neural Networks (CNN), to ease the computational burden without compromising the outcomes. The objective of this thesis is to research the capability of Shunting Inhibitory Cellular Neural Networks (SICNN) to solve the clarity problems in ultrasound imaging. In this thesis, we begin by reviewing a number of traditional enhancement techniques and measures. Since the entire work of this thesis is based upon a particular model of the CNN, we present a brief review of CNN theory and its applications. The SICNN biological inspiration, derivation and stability issues are reviewed with a view to understand its working principle. We then probe a general study of the feed forward and recurrent SICNN systems. Here, the essential response properties of both SICNN Systems are investigated in depth. The enhancing properties of the recurrent SICNN and its advantages compared to more traditional techniques are also studied. After a thorough investigation into the SICNN response properties, we introduce its application for enhancement in Ultrasound Imaging (UI) modality.

A significant fraction of data traffic is transmitted via cellular networks. When introducing fifth-generation (5G) radio access technology, the maximum bitrate of the radio link increases significantly, and the delay is lowered. Network congestion occurs when the sender attempts to send data at a higher rate than the network link or nodes can handle. In order to improve the performance of the mobile networks, many congestion control techniques and approaches have been developed over the years. Varying radio conditions in mobile networks make it challenging to indicate the occurrence of the congestion using packet loss as congestion indicator. This master thesis develops a congestion control algorithm based on Artificial Intelligence (AI) technologies, evaluates and compares it with existing state-of-the-art congestion control algorithms that are used with TCP today.In this study, we use the abundant readable physical layer information exchanged between the base stations and the user equipment to predict the available bandwidth. Two neural network models, Multi-Layer Perceptron (MLP) and Long Short-Term Memory (LSTM), are introduced as congestion control algorithms based on cross-layer information in order to improve user throughput and utilize the available capacity as much as possible.Evaluation in a Long-Term Evolution (LTE) network system simulator confirms that the estimation of LSTM model is able to track the varying link capacity, while MLP is less accurate and induces higher delay. The sender uses the estimated link capacity to adjust its packet sending behavior. Our evaluation reveals that for large flows, the LSTM model can attain higher throughput than state-of-the-art congestion control algorithms, which are the Google Bottleneck Bandwidth and Round-trip propagation time (BBR) algorithm and the Data Center TCP (DCTCP) algorithm. However, it has higher latency than that of these two algorithms. The MLP based model provides unstable performance compared to LSTM; its prediction is not accurate enough and has the highest latency among the algorithms.In conclusion, the LSTM does not underperform the state-of-the-art congestion control algorithms. However, it does not provide additional performance gains in current settings. The MLP model underperforms BBR and DCTCP with L4S and it is not stable enough to be used as a congestion control algorithms.
En betydande del av datatrafiken överförs via mobilnät. Vid introduktion av femte generationens (5G) radioåtkomstteknik ökar den maximala bithastigheten för radiolänken betydligt och förseningen sänks. Nätstockning uppstår när avsändaren försöker skicka data med högre hastighet än nätverkslänken eller noderna kan hantera. För att förbättra prestandan i mobilnät har många tekniker för trängselkontroll utvecklats under åren. Varierande radioförhållanden i mobilnätet gör det utmanande att indikera förekomsten av trängseln med hjälp av paketförlust som trängselindikator. Detta examensarbete utvecklar en trängselkontrollalgoritm baserad på AI-teknik (Artificial Intelligence), utvärderar och jämför den med befintliga toppmoderna trängselkontrollalgoritmer som används med TCP idag.I denna studie använder vi den rikliga läsbara informationen om fysiskt lager som utbyts mellan basstationerna och användarutrustningen för att förutsäga den tillgängliga bandbredden. Två neurala nätverksmodeller, Multi-Layer Perceptron (MLP) och Long Short-Term Memory (LSTM), introduceras som trängselkontrollalgoritmer baserade på tvärskiktsinformation för att förbättra användarens genomströmning och utnyttja den tillgängliga kapaciteten så mycket som möjligt.Utvärdering i en LTE-nätverkssystemsimulator (Long Term Evolution) bekräftar att uppskattningen av LSTM-modellen kan spåra den varierande länkkapaciteten, medan MLP är mindre exakt och inducerar högre fördröjning. Avsändaren använder den uppskattade länkkapaciteten för att justera sitt paketets sändningsbeteende. Vår utvärdering avslöjar att för stora flöden kan LSTM-modellen uppnå högre genomströmning än modernaste trängselkontrollalgoritmer, som är Google Bottleneck Bandbredd och BBR-algoritm och Data Center TCP (DCTCP) ) algoritm. Men det har högre latens än för dessa två algoritmer. Den MLP-baserade modellen ger instabil prestanda jämfört med LSTM; dess förutsägelse är inte nog noggrann och har den högsta latensen bland algoritmerna.Sammanfattningsvis underpresterar LSTM inte de senaste toppkontrollalgoritmerna. Det ger emellertid inte ytterligare prestationsvinster i de aktuella inställningarna. MLP-modellen underpresterar BBR och DCTCP med L4S och den är inte tillräckligt stabil för att användas som en överbelastningskontrollalgoritm.

This thesis proposes the use of Machine Learning (ML) to two very distinct, yet compelling, applications – predicting cardiac arrhythmia episodes and predicting base station association in 5G networks comprising of virtual cells. In the first scenario, Support Vector Machines (SVMs) are used to classify features extracted from electrocardiogram (EKG) signals. The second problem requires a different formulation departing from traditional ML classification where the objective is to partition feature space into constituent class regions. Instead, the intention here is to identify temporal patterns in unequal-length sequences. Using Recurrent Neural Networks (RNNs), it is demonstrated that accurate predictions can be made as to the base station most likely to provide connectivity for a mobile device as it moves. Atrial Fibrillation (AF) is a common cardiac arrhythmia affecting several million people in the United States. It is a condition in which the upper chambers of the heart are unable to contract effectively leading to inhibited blood flow to the ventricles. The stagnation of blood is one of the major risk factors for stroke. The Computers in Cardiology Challenge 2001 was organized to further research into the prediction of episodes of AF. This research revisits the problem with some modifications. Patient-specific classifiers are developed for AF prediction using a different dataset and employing shorter EKG signal epochs. SVM classification yielded an average accuracy of just above 95% in identifying EKG epochs appearing just prior to fibrillatory rhythms. 5G cellular networks were envisaged to provide enhanced data rates for mobile broadband, support low-latency communication, and enable the Internet of Things (IoT). Handovers contribute to latency as mobile devices are switched between base stations due to movements. Given that customers may not be willing to continuously share their exact locations due to privacy concerns and the establishment of a mobile network architecture with dynamically created virtual cells, this research presents a solution for proactive mobility management using RNNs. A RNN is trained to identify patterns in variable-length sequences of Received Signal Strength (RSS) values, where a mobile device is permitted to connect to more than a single base station at a time. A classification accuracy of over 98% was achieved in a simulation model that was set up emulating an urban environment.

Due to the vast success of bioengineering techniques, a series of large-scale analysis tools has been developed to discover the functional organization of cells. Among them, cDNA microarray has emerged as a powerful technology that enables biologists to cDNA microarray technology has enabled biologists to study thousands of genes simultaneously within an entire organism, and thus obtain a better understanding of the gene interaction and regulation mechanisms involved. Although microarray technology has been developed so as to offer high tolerances, there exists high signal irregularity through the surface of the microarray image. The imperfection in the microarray image generation process causes noises of many types, which contaminate the resulting image. These errors and noises will propagate down through, and can significantly affect, all subsequent processing and analysis. Therefore, to realize the potential of such technology it is crucial to obtain high quality image data that would indeed reflect the underlying biology in the samples. One of the key steps in extracting information from a microarray image is segmentation: identifying which pixels within an image represent which gene. This area of spotted microarray image analysis has received relatively little attention relative to the advances in proceeding analysis stages. But, the lack of advanced image analysis, including the segmentation, results in sub-optimal data being used in all downstream analysis methods. Although there is recently much research on microarray image analysis with many methods have been proposed, some methods produce better results than others. In general, the most effective approaches require considerable run time (processing) power to process an entire image. Furthermore, there has been little progress on developing sufficiently fast yet efficient and effective algorithms the segmentation of the microarray image by using a highly sophisticated framework such as Cellular Neural Networks (CNNs). It is, therefore, the aim of this thesis to investigate and develop novel methods processing microarray images. The goal is to produce results that outperform the currently available approaches in terms of PSNR, k-means and ICC measurements.

The entire thesis text is included in the research.pdf file; the official abstract appears in the short.pdf file; a non-technical public abstract appears in the public.pdf file. Title from PDF of title page (University of Missouri--Columbia, viewed on November 13, 2009). Thesis advisor: Dr. Guilherme DeSouza. Includes bibliographical references.

Esta tesis doctoral se enmarca en la temática de Gestion de Recursos Radioelectricos en los sistemas de comunicaciones móviles, 3G, B3G y 4G. Como es ampliamente conocido los recursos radios son escasos, particularmente en los sistemas de comunicaciones móviles celulares, y en consecuencia es de uso obligado una gestión eficiente de los mismos. Desde un punto de vista práctico si bien estos sistemas se basan en el uso de tecnologías de acceso radio estandarizada, no es este el caso para los algoritmos subyacentes a la gestión de los recursos radio, de manera que siempre son posibles nuevas realizaciones de los mismos que resulten más convenientes en el marco del estándar en cuestión. Es en esta dirección hacia donde apunta la realización de esta tesis doctoral, y entiendo que lo consigue con éxito al introducir nuevas estrategias de Gestion de los Recursos Radio en el marco de las estrategias de múltiple acceso CDMA en los sistemas 3G, TDMA/CDMA en los sistemas B3G y OFDMA en los sistemas 4G. Esta tesis, tras identificar los gestores más usuales de gestión de recursos radio y una breve descripción de los mismos, introduce una descripción básicamente autocontenida de los aspectos más relevantes de los sistemas de acceso múltiple WCDMA y OFDMA. En este sentido se detallan mecanismos de su funcionamiento que con posterioridad serán utilizados en la definición y especificación de los algoritmos de gestión de recursos propiamente dichos. Con posterioridad se hace un breve recorrido sobre lo que son las redes neuronales , para finalizar en una exposición más detallada de las Redes Neuronales de Hopfield que serán el hilo conductor de los trabajos de esta tesis. En particular se describen las ecuaciones que caracterizan estas redes como sistemas dinámicos y se establecen sus condiciones de convergencia a través de los teoremas de estabilidad Lyapunov y la definición de la función Energía.De la conjunción de las particularidades de los sistemas de acceso WCDMA, TDMA y OFDMA y de las redes neuronales de Hopfield se van desarrollando una serie de algoritmos que operan en escenarios unicelulares y que entiendo novedosos y que a continuación enumeran brevemente.Admisión en un sistema WCDMA , enlace ascendente, mediante una gestión optimizada de las distintas velocidades de transmisión asignadas a los usuarios que comparten el acceso y que se les permite distintos perfiles. Aspectos relativos a la robustez del algoritmo, y en particular a su convergencia son también detallados. Se suponen restricciones de carga de la red máxima, repartición del espectro justa y potencia máxima disponible en los terminales móviles. Se suponen un servicio en tiempo real con velocidades variables. La probabilidad de bloqueo se usa para exhibir las prestaciones del algoritmo.Gestión de las velocidades de los usuarios ya admitidos en un sistema WCDMA,enlace ascendente, con objeto de garantizarles una definida probabilidad de satisfacción superior a un determinado valor y que está basada en las velocidades reales de transmisión asignadas. Se supone también un servicio en tiempo real con velocidades variables y las mismas restricciones que en Admisión. Gestión de las velocidades de los usuarios ya admitidos en un sistema WCDMA, enlace descendente, con objeto de garantizarles un máximo retardo en la entrega de paquetes. Se suponen restricciones de repartición del espectro justa y potencia máxima disponible en la estación de base. Se supone un servicio interactivo basado en un modelo de tráfico para servicios www. Se introduce también un algoritmo de referencia a efectos comparativos. La probabilidad de pérdida es el parámetro usado para valorar las prestaciones del algoritmo.Gestión combinada de servicios en tiempo real e interactivos en sistemas WCDMA, enlace descendente. Incorpora parte de los algoritmos anteriormente enunciados y se mantienen los mismos modelos de tráfico y las mismas restricciones. Se han usado en esta caso las probabilidades de satisfacción y de pérdida para capturar el la velocidad de transmisión agregada y retardo respectivamente Algoritmo de Gestión común de recursos radio para un escenario B3G donde un usuario puede ser servido por más de un acceso. En este caso se han usado WCDMA y TDMA. Algoritmos de Gestión de las velocidades de los usuarios ya admitidos en un sistema OFDMA, enlace descendente, con objeto de garantizarles un máximo retardo en la entrega de paquetes.La tesis apunta también hacia prometedoras futuras líneas de investigación que pretenden explotar la base de la metodología desarrollada en esta tesis y que consisten en escenarios celulares centralizadas para pasar después a distribuidas en entornos multicelulares y en particular para los sistemas OFDMA , base de los accesos en 4G.

Thesis (M. Phil.)--Hong Kong University of Science and Technology, 2003.
Includes bibliographical references (leaves 50-51). Also available in electronic version. Access restricted to campus users.

Cellular level learning is vital to almost all brain function, and extensive homeostatic plasticity is required to maintain brain functionality. While much has been learned about cellular level plasticity in vivo, how these mechanisms affect higher level functionality is not readily apparent. The cellular level circuitry of most networks that process information is unknown. A variety of models were developed to better understand plasticity in both learning and homeostasis.

Spike time dependent plasticity (STDP) and reward-modulated plasticity may be the primary methods through which neurons record information. We implemented rewarded STDP to model foraging behavior in a virtual environment. When appropriate homeostatic mechanisms were in place, the network of spiking neurons developed the capability of producing highly successful decision-making.

The networks used in the foraging model used a very simple initial configuration to avoid assumptions about network organization. More realistic network configurations can help to show how plasticity interacts with genetically determined network. We developed three network models of synaptic mechanisms of FM sweep processing based on published experimental data. One of these, the 'inhibitory sideband' model, used frequency selective inputs to a network of excitatory and inhibitory cells. The strength and asymmetry of these connections resulted in neurons responsive to sweeps in a single direction and of sufficient rate. STDP was shown to be capable of causing to become selective for sweeps in the same direction as a repeatedly presented training sweep.

The experience dependent plasticity, occurs primarily during the waking state, however, sleep is essential for learning. Slow wave sleep activity may be essential for memory consolidation and homeostasis. We developed a model of slow wave sleep that included methods to calculate the electrical field in the space around the network. We show here that a network model of up and down states displays this CSD profile only if a frequency-filtering extracellular medium is assumed. While initiation of the active cortical states during sleep slow oscillation has been intensively studied, the it's termination remains poorly understood. We explored the impact of intrinsic and synaptic inhibition on the state transition. We found that synaptic inhibition controls the duration and the synchrony of active state termination.

Together these models set the stage for a model network that can learn through input driven processes in a waking state then explore the consolidation of memory in a sleeping state. This will allow us to explore in greater detail how plasticity on the level of a single cell contributes to learning and stability on the level of the whole brain.

Huge amounts of data are being collected as a result of the increased use of mobile telecommunications. Insight into information and knowledge derived from these databases can give operators a competitive edge in terms of customer care and retention,
marketing and fraud detection. One of the strategies for fraud detection checks for signs of questionable changes in user behavior. Although the intentions of the mobile phone users cannot be observed, their intentions are reflected in the call data which define usage patterns. Over a period of time, an individual phone generates a large pattern of use. While call data are recorded for subscribers for billing purposes, we are making no prior assumptions about the data indicative of fraudulent call patterns, i.e. the calls made for billing purpose are unlabeled. Further analysis is thus, required to be able to isolate fraudulent usage. An unsupervised learning algorithm can analyse and cluster call patterns for each subscriber in order to facilitate the fraud detection process.

This research investigates the unsupervised learning potentials of two neural networks for the profiling of calls made by users over a period of time in a mobile telecommunication network. Our study provides a comparative analysis and application of Self-Organizing Maps (SOM) and Long Short-Term Memory (LSTM) recurrent neural networks algorithms to user call data records in order to conduct a descriptive data mining on users call patterns.

Our investigation shows the learning ability of both techniques to discriminate user call patterns
the LSTM recurrent neural network algorithm providing a better discrimination than the SOM algorithm in terms of long time series modelling. LSTM discriminates different types of temporal sequences and groups them according to a variety of features. The ordered features can later be interpreted and labeled according to specific requirements of the mobile service provider. Thus, suspicious call behaviours are isolated within the mobile telecommunication network and can be used to to identify fraudulent call patterns. We give results using masked call data
from a real mobile telecommunication network.

Nessa dissertação, é investigado o armazenamento e a recuperação de padrões de forma biologicamente inspirada no cérebro. Os modelos estudados consistiram de redes neurais realimentadas, que tentam modelar certos aspectos dinâmicos do funcionamento do cérebro. Em particular, atenção especial foi dada às Redes Neurais Celulares, que constituem uma versão localmente acoplada do já clássico modelo de Hopfield. Além da análise de estabilidade das redes consideradas, foi realizado um teste com o intuito de avaliar o desempenho de diversos métodos de construção de memórias endereçáveis por conteúdo (memórias associativas) em Redes Neurais Celulares.
In this dissertation we investigate biologically inspired models of pattern storage and retrieval, by means of feedback neural networks. These networks try to model some of the dynamical aspects of brain functioning. The study concentrated in Cellular Neural Networks, a local coupled version of the classical Hopfield model. The research comprised stability analysis of the referred networks, as well as performance tests of various methods for content-addressable (associative) memory design in Cellular Neural Networks.

To support the continuously increasing number of mobile telephone users around the world, mobile communication systems have become more advanced and sophisticated in their designs. As a result of the great success with the second generation mobile radio networks, deployment of the third and development of fourth generations, the demand for higher data rates to support available services, such as internet connection, video telephony and personal navigation systems, is ever growing. To be able to meet the requirements regarding bandwidth and number of users, enhancements of existing systems and introductions of conceptually new technologies and techniques have been researched and developed. Although new proposed technologies in theory provide increased network capacity, the backbone of a successful roll-out of a mobile telephone system is inevitably the planning of the network’s cellular structure. Hence, the fundamental aspect to a reliable cellular planning is the knowledge about the physical radio channel for wide sets of different propagation scenarios. Therefore, to study radio wave propagation in typical Australian environments, the Commonwealth Scientific and Industrial Research Organisation (CSIRO) and the Australian Telecommunications Cooperative Research Centre (ATcrc) in collaboration developed a cellular code division multiple access (CDMA) pilot scanner. The pilot scanner measurement equipment enables for radio wave propagation measurements in available commercial CDMA mobile radio networks, which in Australia are usually deployed for extensive rural areas. Over time, the collected measurement data has been used to characterise many different types of mobile radio environments and some of the results are presented in this thesis. The thesis is divided into an introduction section and four parts based on peer-reviewed international research publications. The introduction section presents the reader with some relevant background on channel and propagation modelling. Also, the CDMA scanner measurement system that was developed in parallel with the research results founding this thesis is presented. The first part presents work on the evaluation and development of the different revisions of the Recommendation ITU-R P.1546 point-to-area radio wave propagation prediction model. In particular, the modified application of the terrain clearance angle (TCA) and the calculation method of the effective antenna height are scrutinized. In the second part, the correlation between the smallscale fading characteristics, described by the Ricean K-factor, and the vegetation density in the vicinity of the mobile receiving antenna is investigated. The third part presents an artificial neural network (ANN) based technique incorporated to predict path loss in rural macrocell environments. Obtained results, such as prediction accuracy and training time, are presented for different sized ANNs and different training approaches. Finally, the fourth part proposes an extension of the path loss ANN enabling the model to also predict small-scale fading characteristics.

L'objectiu principal d'aquesta tesi consisteix a estudiar la factibilitat d'implementar un sensor càmera CNN amb plena funcionalitat basat en FPGA de baix cost adequat per a aplicacions en robots mòbils. L'estudi dels fonaments de les xarxes cel•lulars no lineals (CNNs) i la seva aplicació eficaç en matrius de portes programables (FPGAs) s'ha complementat, d'una banda amb el paral•lelisme que s'estableix entre arquitectura multi-nucli de les CNNs i els eixams de robots mòbils, i per l'altre banda amb la correlació dinàmica de CNNs i arquitectures memristive. A més, els memristors es consideren els substituts dels futurs dispositius de memòria flash per la seva capacitat d'integració d'alta densitat i el seu consum d'energia prop de zero. En el nostre cas, hem estat interessats en el desenvolupament d’FPGAs que han deixat de ser simples dispositius per a la creació ràpida de prototips ASIC per esdevenir complets dispositius reconfigurables amb integració de la memòria i els elements de processament general. En particular, s'han explorat com les arquitectures implementades CNN en FPGAs poden ser optimitzades en termes d’àrea ocupada en el dispositiu i el seu consum de potència. El nostre objectiu final ens ah portat a implementar de manera eficient una CNN-UM amb complet funcionament a un baix cost i baix consum sobre una FPGA amb tecnología flash. Per tant, futurs estudis sobre l’arquitectura eficient de la CNN sobre la FPGA i la interconnexió amb els robots comercials disponibles és un dels objectius d'aquesta tesi que se seguiran en les línies de futur exposades en aquest treball.
El objetivo principal de esta tesis consiste en estudiar la factibilidad de implementar un sensor cámara CNN con plena funcionalidad basado en FPGA de bajo coste adecuado para aplicaciones en robots móviles. El estudio de los fundamentos de las redes celulares no lineales (CNNs) y su aplicación eficaz en matrices de puertas programables (FPGAs) se ha complementado, por un lado con el paralelismo que se establece entre arquitectura multi -núcleo de las CNNs y los enjambres de robots móviles, y por el otro lado con la correlación dinámica de CNNs y arquitecturas memristive. Además, los memristors se consideran los sustitutos de los futuros dispositivos de memoria flash por su capacidad de integración de alta densidad y su consumo de energía cerca de cero. En nuestro caso, hemos estado interesados en el desarrollo de FPGAs que han dejado de ser simples dispositivos para la creación rápida de prototipos ASIC para convertirse en completos dispositivos reconfigurables con integración de la memoria y los elementos de procesamiento general. En particular, se han explorado como las arquitecturas implementadas CNN en FPGAs pueden ser optimizadas en términos de área ocupada en el dispositivo y su consumo de potencia. Nuestro objetivo final nos ah llevado a implementar de manera eficiente una CNN-UM con completo funcionamiento a un bajo coste y bajo consumo sobre una FPGA con tecnología flash. Por lo tanto, futuros estudios sobre la arquitectura eficiente de la CNN sobre la FPGA y la interconexión con los robots comerciales disponibles es uno de los objetivos de esta tesis que se seguirán en las líneas de futuro expuestas en este trabajo.
The main goal of this thesis consists in studying the feasibility to implement a full-functionality CNN camera sensor based on low-cost FPGA device suitable for mobile robotic applications. The study of Cellular Nonlinear Networks (CNNs) fundamentals and its efficient implementation on Field Programmable Gate Arrays (FPGAs) has been complemented, on one side with the parallelism established between multi-core CNN architecture and swarm of mobile robots, and on the other side with the dynamics correlation of CNNs and memristive architectures. Furthermore, memristors are considered the future substitutes of flash memory devices because of its capability of high density integration and its close to zero power consumption. In our case, we have been interested in the development of FPGAs that have ceased to be simple devices for ASIC fast prototyping to become complete reconfigurable devices embedding memory and processing elements. In particular, we have explored how the CNN architectures implemented on FPGAs can be optimized in terms of area occupied on the device or power consumption. Our final accomplishment has been implementing efficiently a fully functional reconfigurable CNN-UM on a low-cost low-power FPGA based on flash technology. Therefore, further studies on an efficient CNN architecture on FPGA and interfacing it with commercially-available robots is one of the objectives of this thesis that will be followed in the future directions exposed in this work.

Le néocortex possède un état activé dans lequel l'activité corticalemanifeste un comportement complexe. Au niveau cellulaire, l'activitéest caractérisée par de fortes fluctuations sous-liminaires dupotential membranaire et une décharge irrégulière à bassefréquence. Au niveau du réseau, l'activité est marquée par un faibleniveau de synchronie et une dynamique chaotique. Néanmoins, c'est dansce régime que l'information est traitée de manière fiable par lesréseaux neuronaux. Ce régime est donc crucial pour le traitement del'information par le cortex. Dans cette thèse, nous contribuons à sacompréhension en examinant comment les propriétés biophysiques auniveau cellulaire combinées avec les propriétés d'architecture desréseaux façonnent cette dynamique asynchrone.Cette thèse repose sur les modèles de dynamique de réseaux appelésmodèles de champ moyen, un formalisme théorique qui décrit ladynamique de population grâce à une approche auto-consistante. Aucoeur de ce formalisme se trouve la fonction de transfertneuronale : la fonction entrée-sortie d'un neurone. La première partiede cette thèse s'attache à dériver des fonctions de transfertbiologiquement réalistes en incorporant des caractérisationsexpérimentales.Dans un premier temps, nous avons examiné in vitro comment lesneurones néocorticaux pyramidaux de la couche V du cortex visuelrépondent à des fluctuations du potentiel membranaire. Nous avonsobservé que les neurones individuels ne diffèrent pas seulement entermes d'excitabilité, mais qu'ils diffèrent aussi par leurssensibilités aux paramètres des fluctuations. Dans un deuxième temps,nous avons étudié de manière théorique comment l'intégrationdendritique dans des structures arborescentes façonne les fluctuationsau soma. Nous avons observé que, en fonction des propriétés del'activité présynaptique, différentes comodulations des paramètres desfluctuations pouvaient être obtenues. En combinant cette observationavec nos mesures expérimentales, nous avons observé que cela induisaitdes couplages différents entre activité synaptique et déchargeneuronale pour chaque neurone. Nous proposons donc que, puisque cemécanisme offre un moyen d'activer spécifiquement certains neurones enfonction des propriétés de l'entrée, l'hétérogénéité biophysiquepourrait contribuer à l'encodage de propriétés des stimuli dans lestraitements de l'information sensorielle.La deuxième partie de cette thèse examine comment les propriétésd'architecture des réseaux neuronaux se combinent avec les propriétésbiophysiques et affectent les réponses sensorielles via des effets dedynamiques de populations.Nous avons tout d'abord examiné de manière théorique comment un hautniveau d'activité spontanée impactait les réponses post-synaptiquesdans le cortex. Nous avons observé que la compétition entre lerecrutement dans le réseau cortical activé et les effets deconductances associés prédisaient une relation non-triviale entrel'intensité des stimuli et l'amplitude des réponses. Cette prédictionfut observée dans des enregistrements de réponses post-synaptiquesdans le cortex auditif du rat in vivo en réponse à des stimulicorticaux, thalamiques et auditifs.Pour finir, en tirant avantage des approches de champ moyen, nousavons construit un modèle grande échelle du réseau des couches II-IIIincluant le réseau des fibres horizontales. Nous avons examiné lespropriétés intégratives spatio-temporelles du modèle et nous les avonscomparées avec des mesures par imagerie optique de l'activitécérébrale chez le singe éveillé. En particulier, nous avonsreconstruit une expérience typique du traitement sensoriel: lemouvement apparent. Le modèle prédit un fort signal suppressif dont leprofil spatio-temporel correspond quantitativement à celui observé invivo
The neocortex of awake animals displays an activated state in whichcortical activity manifests highly complex, seemingly noisybehavior. At the level of single neurons the activity is characterizedby strong subthreshold fluctuations and irregular firing at lowrate. At the network level, the activity is weakly synchronized andexhibits a chaotic dynamics. Yet, it is within this regime thatinformation is processed reliably through neural networks. This regimeis thus crucial to neural computation. In this thesis, we contributeto its understanding by investigating how the biophysical propertiesat the cellular level combined with the properties of the networkarchitecture shapes this asynchronous dynamics.This thesis builds up on the so-called mean-field models of networkdynamics, a theoretical formalism that describes population dynamicsvia a self-consistency approach. At the core of this formalism lie theneuronal transfer function: the input-output description of individualneurons. The first part of this thesis focuses on derivingbiologically-realistic neuronal transfer functions. We firstformulate a two step procedure to incorporate biological details (suchas an extended dendritic structure and the effect of various ionicchannels) into this transfer function based on experimentalcharacterizations.First, we investigated in vitro how layer V pyramidal neocorticalneurons respond to membrane potential fluctuations on a cell-by-cellbasis. We found that, not only individual neurons strongly differ interms of their excitability, but also, and unexpectedly, in theirsensitivities to fluctuations. In addition, using theoreticalmodeling, we attempted to reproduce these results. The model predictsthat heterogeneous levels of biophysical properties such as sodiuminactivation, sharpness of sodium activation and spike frequencyadaptation account for the observed diversity of firing rateresponses.Then, we studied theoretically how dendritic integration in branchedstructures shape the membrane potential fluctuations at the soma. Wefound that, depending on the type of presynaptic activity, variouscomodulations of the membrane potential fluctuations could beachieved. We showed that, when combining this observation with theheterogeneous firing responses found experimentally, individual neuronsdifferentially responded to the different types of presynapticactivities. We thus propose that, because this mechanism offers a wayto produce specific activation as a function of the input properties,biophysical heterogeneity might contribute to the encoding of the stimulusproperties during sensory processing in neural networks.The second part of this thesis investigates how circuit properties,such as recurrent connectivity and lateral connectivity, combine withbiophysical properties to impact sensory responses through effectsmediated by population dynamics.We first investigated what was the effect of a high level of ongoingdynamics (the Up-state compared to the Down-state) on the scaling ofpost-synaptic responses. We found that the competition between therecruitment within the active recurrent network (in favor of highresponses in the Up-state) and the increased conductance level due tobackground activity (in favor of reduced responses in the Up-state)predicted a non trivial stimulus-response relationship as a functionof the intensity of the stimulation. This prediction was shown toaccurately capture measurements of post-synaptic membrane potentialresponses in response to cortical, thalamic or auditory stimulation inrat auditory cortex in vivo.Finally, by taking advantage of the mean-field approach, weconstructed a tractable large-scale model of the layer II-III networkincluding the horizontal fiber network. We investigate thespatio-temporal properties of this large-scale model and we compareits predictions with voltage sensitive dye imaging in awake fixatingmonkey

Plants are essential for the existence of most living things on this planet. Plants are used for providing food, shelter, and medicine. The ability to identify plants is very important for several applications, including conservation of endangered plant species, rehabilitation of lands after mining activities and differentiating crop plants from weeds. In recent times, many researchers have made attempts to develop automated plant species recognition systems. However, the current computer-based plants recognition systems have limitations as some plants are naturally complex, thus it is difficult to extract and represent their features. Further, natural differences of features within the same plant and similarities between plants of different species cause problems in classification. This thesis developed a novel hybrid intelligent system based on a neuro-genetic model for automatic recognition of plants using leaf image analysis based on novel approach of combining several image descriptors with Cellular Neural Networks (CNN), Genetic Algorithm (GA), and Probabilistic Neural Networks (PNN) to address classification challenges in plant computer-based plant species identification using the images of plant leaves. A GA-based feature selection module was developed to select the best of these leaf features. Particle Swam Optimization (PSO) and Principal Component Analysis (PCA) were also used sideways for comparison and to provide rigorous feature selection and analysis. Statistical analysis using ANOVA and correlation techniques confirmed the effectiveness of the GA-based and PSO-based techniques as there were no redundant features, since the subset of features selected by both techniques correlated well. The number of principal components (PC) from the past were selected by conventional method associated with PCA. However, in this study, GA was used to select a minimum number of PC from the original PC space. This reduced computational cost with respect to time and increased the accuracy of the classifier used. The algebraic nature of the GA’s fitness function ensures good performance of the GA. Furthermore, GA was also used to optimize the parameters of a CNN (CNN for image segmentation) and then uniquely combined with PNN to improve and stabilize the performance of the classification system. The CNN (being an ordinary differential equation (ODE)) was solved using Runge-Kutta 4th order algorithm in order to minimize descritisation errors associated with edge detection. This study involved the extraction of 112 features from the images of plant species found in the Flavia dataset (publically available) using MATLAB programming environment. These features include Zernike Moments (20 ZMs), Fourier Descriptors (21 FDs), Legendre Moments (20 LMs), Hu 7 Moments (7 Hu7Ms), Texture Properties (22 TP) , Geometrical Properties (10 GP), and Colour features (12 CF). With the use of GA, only 14 features were finally selected for optimal accuracy. The PNN was genetically optimized to ensure optimal accuracy since it is not the best practise to fix the tunning parameters for the PNN arbitrarily. Two separate GA algorithms were implemented to optimize the PNN, that is, the GA provided by MATLAB Optimization Toolbox (GA1) and a separately implemented GA (GA2). The best chromosome (PNN spread) for GA1 was 0.035 with associated classification accuracy of 91.3740% while a spread value of 0.06 was obtained from GA2 giving rise to improved classification accuracy of 92.62%. The PNN-based classifier used in this study was benchmarked against other classifiers such as Multi-layer perceptron (MLP), K Nearest Neigbhour (kNN), Naive Bayes Classifier (NBC), Radial Basis Function (RBF), Ensemble classifiers (Adaboost). The best candidate among these classifiers was the genetically optimized PNN. Some computational theoretic properties on PNN are also presented.

Els paradigmes cel·lulars, com les xarxes neuronals cel·lulars (CNN, en anglès) i els autòmats cel·lulars (CA, en anglès), són una eina excel·lent de càlcul, al ser equivalents a una màquina universal de Turing. La introducció de la màquina universal CNN (CNN-UM, en anglès) ha permès desenvolupar hardware, el nucli computacional del qual funciona segons la filosofia cel·lular; aquest hardware ha trobat aplicació en diversos camps al llarg de la darrera dècada. Malgrat això, encara hi ha moltes preguntes a obertes sobre com definir els algoritmes d'una CNN-UM i com estudiar la dinàmica dels autòmats cel·lulars. En aquesta tesis es tracten els dos problemes: primer, es demostra que es possible acotar l'espai dels algoritmes per a la CNN-UM i explorar-lo gràcies a les tècniques genètiques; i segon, s'expliquen els fonaments de l'estudi dels CA per mitjà de la dinàmica no lineal (segons la definició de Chua) i s'il·lustra com aquesta tècnica ha permès trobar resultats innovadors.
Los paradigmas celulares, como las redes neuronales celulares (CNN, en
inglés) y los autómatas celulares (CA, en inglés), son una excelente
herramienta de cálculo, al ser equivalentes a una maquina universal de
Turing. La introducción de la maquina universal CNN (CNN-UM, en
inglés) ha permitido desarrollar hardware cuyo núcleo computacional
funciona según la filosofía celular; dicho hardware ha encontrado
aplicación en varios campos a lo largo de la ultima década. Sin
embargo, hay aun muchas preguntas abiertas sobre como definir los
algoritmos de una CNN-UM y como estudiar la dinámica de los autómatas
celular. En esta tesis se tratan ambos problemas: primero se demuestra
que es posible acotar el espacio de los algoritmos para la CNN-UM y
explorarlo gracias a técnicas genéticas; segundo, se explican los
fundamentos del estudio de los CA por medio de la dinámica no lineal
(según la definición de Chua) y se ilustra como esta técnica ha
permitido encontrar resultados novedosos.
Cellular paradigms, like Cellular Neural Networks (CNNs) and Cellular Automata (CA) are an excellent tool to perform computation, since they are equivalent to a Universal Turing machine. The introduction of the Cellular Neural Network - Universal Machine (CNN-UM) allowed us to develop hardware whose computational core works according to the principles of cellular paradigms; such a hardware has found application in a number of fields throughout the last decade. Nevertheless, there are still many open questions about how to define algorithms for a CNN-UM, and how to study the dynamics of Cellular Automata. In this dissertation both problems are tackled: first, we prove that it is possible to bound the space of all algorithms of CNN-UM and explore it through genetic techniques; second, we explain the fundamentals of the nonlinear perspective of CA (according to Chua's definition), and we illustrate how this technique has allowed us to find novel results.

We attempted a mathematical model for expected prognosis of lung cancer patients based on a multivariate analysis of the values of ER-interacting proteins (ERbeta) and a membrane bound, glycosylated phosphoprotein MUC1), and patients clinical data recorded at the time of initial surgery. We demonstrate that, even with the limited sample size available to use, combination of clinical and biochemical data (in particular, associated with ERbeta and MUC1) allows to predict survival of lung cancer patients with about 80% accuracy while prediction on the basis of clinical data only gives about 70% accuracy. The present work can be viewed as a pilot study on the subject: since results confirm that ER-interacting proteins indeed inuence lung cancer patients' survival, more data is currently being collected.
ID: 030646185; System requirements: World Wide Web browser and PDF reader.; Mode of access: World Wide Web.; Thesis (M.S.)--University of Central Florida, 2011.; Includes bibliographical references (p. 32-33).
M.S.
Masters
Mathematics
Sciences
Mathematical Science

Chaos is a remarkable phenomenon occurring in many nonlinear sys-tems, where the deterministic nature of the system structure conjugates with the irregularity of the behviour. Since the first findings on chaos in mathematical models, the idea of using electronic circuits as experimental testbeds for chaos aroses. The focus of this PhD thesis is indeed on an experimental approach to the study of chaos, to its characteristic features and to the synchronization properties mainly through chaotic circuit design implementation and experiments. Starting from general guidelines on how to impl,ement from a mathe-matical model, an electronic circuit governed by the some equations, a gallery of circuits (Chua, Lorenz, Rössler, Hindmarsh-Rose, Duffing, Langford, Colpitts and a memristive circuit) designed and implemented with off-the-shelf components is presented in Chapter 1. A general methodology for designing a new class of chaotic circuits based on time-delay is then discussed in Chapter 2. Chaos has unique properties even when two or more coupled chaotic systems are consi-dered. The experimental approach to this topic of chaos theory pursued in this thesis led to several important results that otherwise had not been possible to reveal. In fact, in Chapter 3 we discuss findings on the synchronization of chaotic circuits in the presence of either parametric or structural dissimetries, and present a very interesting observation of the circuits is minimized when the two circuits synchronously evolve. Finally, Chapter 4 discusses a new form of synchronization occurring when more than two nonlinear circuits are coupled in networks with particular topologies.

Modeling of flow in porous and fractured media is a very important problem in reservoir engineering. As for numerical simulations conventional Navier-Stokes codes are applied to flow in both porous and fractured media. But they have long computation times, poor convergence and problems of numerical instabilities. Therefore, it is desired to develop another computational method that is more efficient and use simple rules to represent the flow in fractured media rather than partial differential equations. In this thesis Lattice Boltzmann Automaton Model will be used to represent the single phase fluid flow in two dimensional synthetic fractures and the simulation results obtained from this model are used to train Artificial Neural Networks. It has been found that as the mean aperture-fractal dimension ratio increases permeability increases. Moreover as the anisotropy factor increases permeability decreases with a second order polynomial relationship.

Made available in DSpace on 2017-07-21T19:25:54Z (GMT). No. of bitstreams: 1 Fernando da Silva Borges.pdf: 3003505 bytes, checksum: c77a390868c21644a0396314c4bf4e0e (MD5) Previous issue date: 2016-11-22
Fundação Araucária de Apoio ao Desenvolvimento Científico e Tecnológico do Paraná
In this thesis, we use mathematical models to study the dynamic range of neural networks. The dynamic range is the difference between maximum and minimum levels of sensation produced by known stimuli. Using cellular automata to model neuronal dynamics and different network topologies with different types of synapses, we investigate for which conditions the dynamic range is enhanced. In a network where local connections represent the electrical synapses and nonlocal connections the chemical synapses, we analyze the dynamic range in function of the number of nonlocal connections and time delay between these connections. We find that the dynamic range is enhanced for neural networks with low time delay when the number of nonlocal connections increases. Furthermore, we propose a neural network model separated into two layers, where one layer corresponds to inhibitory and the other to excitatory neurons. We randomly distribute electrical and chemical synapses in the network in order to analyse the effects on the dynamic range. In our proposed model, the chemical synapses, that are directed, can be excitatory or inhibitory, while the electrical synapses are bidirectional. Through the mean-field approximation, we analytically calculate the dynamic range as a function of the model parameters. The values that we find are very close to the results obtained from simulations. We verify that electrical synapses have a complementary effect on the enhancement of the dynamic range. Finally, we found that electrical synapses on excitatory layer are responsible for this complementary effect, while the electrical synapses in inhibitory layer promote a small increase in the dynamic range value.
Nesta tese usamos modelos matemáticos para estudar a faixa dinâmica de redes neuronais. A faixa dinâmica é a diferença entre a resposta máxima e mínima produzida por um determinado estímulo. Utilizando autômatos celulares para modelar a dinamica neuronal e diversas topologias de redes com diferentes tipos de sinapses, investigamos para quais configurações a faixa dinamica ´e maximizada. Em uma rede onde conexões locais representam sinapses elétricas e conexões não locais as sinapses químicas, analisamos o que ocorre com a faixa dinamica quando varia-se a quantidade de conexões não locais ou um tempo de atraso entre essas conexões é considerado. Neste caso, verificamos que a faixa dinamica é maior para redes neuronais com valores baixos de atraso e aumenta com o acréscimo de conexões não locais. Além disso, propomos um modelo de rede de neurônios dispostos em duas camadas, uma excitatória e outra inibitótira, com sinapses química e elétricas distribuidas aleatoriamente. Neste modelo, as sinapses quimicas são direcionadas e podem ser excitatorias ou inibitórias, enquanto as sinapses elétricas são bidirecionais e apresentam apenas carater excitatorio. Fazendo aproximações de campo médio, calculamos analiticamente a faixa dinamica em função dos parametros do modelo. Os valores encontrados estão muito próximos dos obtidos por simulações e mostram que a faixa dinãmica é maximizada em pontos que dependem complementarmente das sinapses quimicas e elétricas. Finalmente, verificamos que as sinapses eletricas na camada excitatória sao responsaveis por esse efeito complementar, enquanto as sinapses elétricas na camada inibitoria promovem um pequeno acrescimo no valor da faixa dinamica.

Il paradigma delle Reti Neurali Artificiali (ANN) consiste nell’applicazione del modello neurale “biologico” per la risoluzione di problemi che spesso sono troppo complessi per un’architettura di Von Neumann. La letteratura offre differenti approcci per l’implementazione di ANN. Qualche implementazione è di tipo software, altre sono soluzioni circuitali come circuiti digitali full-custom o FPGA (Field Programmable Gate Array), come pure circuiti analogici, e il tipo di implementazione di certo dipende dal tempo di esecuzione adeguato al tipo di applicazione. Questa tesi riguarda la progettazione di nuovi circuiti analogici adattati per le Reti Neurali. In particolare, saranno utilizzate le Reti Neurali Cellulari (CNN) proposte nel 1981 dal Prof. L.O.Chua (University of California – Berkeley). Il “Laboratorio di Circuiti” dell’Università di Roma “Tor Vergata” ha progettato e realizzato alcuni chip analogici dedicati a questo tipo di Reti Neurali. Questi chip appartengono alla famiglia “Digital Programmable CNN” (DPCNN) e presentano principalmente due caratteristiche: la programmabilità digitale dei pesi sinaptici come una particolare architettura orientata ad una struttura interconnessa (cioè connettendo tra loro più di questi chip è possibile realizzare reti di grande dimensione). In questa tesi viene data una visione di insieme sulle ANN, sulle CNN e sulle Star-CNN: cosa sono, come funzionano ed a cosa servono. In perticolare verrà descritta la famiglia DP-CNN. Questa tesi propone una nuova architettura chiamata TD-CNN (Time Division CNN), che sfrutta una particolare strategia mirata a ridurra l’area di occupazione su silicio di una cella elementare, per aumentare l’integrabilità della rete. Oltretutto la stessa strategia a divisione di tempo verrà applicata alle TD-Star CNN. In particolare questi circuiti sono le non-linerità digitalmente programmabili (cioè DPTA – Digital Programmable Transconductance Amplifier e DPTA – Digital Programmable Transconductance Comparator) e circuiti particolari per la multiplazione (DM-SH – Dynamic Mirror Sample and Hold e DM-MUX – Dynamic Mirror Multiplexer). Sono mostrate alcune simulazioni dei circuiti per permettere lo studio di queste nuove architetture, e la modifica delle dinamiche introdotte dalla strategia a divisione di tempo.
The Artificial Neural Network (ANN) paradigm consists of the application of biological “neural” models to the solution of particular problems that often are very hard to solve for the classical “Von Neumann” architectures. Different are the approaches proposed in literature for the implementation of an ANN. Some of them are software implementations only while, others are circuital solutions as full custom digital circuits or programmed FPGAs (Field Programmable Gate Array) as well as analogue circuits and the typology of the implementation certainly depends on the length of the processing time that you believe adequate for the particular application. This thesis is focused on the design of new analogue circuits well suited for Neural Network applications. In particular, the class of the Cellular Neural Networks (CNN), proposed in 1981 by Prof. L.O.Chua (University of California - Berkeley), will be exploited. In this area, the “Laboratorio Circuiti” at University of Rome “Tor Vergata” designed and manufactured several analogue chips devoted to this class of Neural Networks. These chips belong to the Digital Programmable CNN (DPCNN) chip family and present two main features: the digital programmability of the synaptic weights as well as a special architecture oriented to an interconnection structure (i.e. it is possible to carry out large network by connecting together more of these chips). In this thesis work you will find an overview about the Artificial Neural Network, the Cellular Neural Network and the Star Cellular Neural Network: what they are, how they work and why they are useful. In particular, the DP-CNN chip family will be deeply described. This thesis proposes the TD-CNN (Time Division CNN), a particular design strategy, devoted to reduce the silicon area occupation of the a elementary cell in order to improve the VLSI integrability of the network. Moreover, the same time-division strategy will be applied to TD-Star CNN. In particular, these circuits consist of the digitally programmable non-linearity circuits (i.e. the Digital Programmable Transconductance Amplifier - DPTA and Digital Programmable Transconductance Comparator – DTPC) and special circuit for to carry out the multiplexing feature (i.e. the Dynamic Mirror Sample and Hold – DM-SH and the Multiplexer – DM-MUX). Several circuital simulations will be shown in order to study the behavior of this modified architecture and the modifications on the dynamics introduced by the time division strategy.

博士
國立中正大學
電機工程所
97
The objective of this thesis is to explore three of the critical topics in implementing multi-functional cellular neural network (CNN). First of all, we developed the theoretical background underlies the methods to embed multiple functions into a single cellular neural network. The properties and the efficacies of the embedding methods were discussed. Secondly, we propose a novel paradigm for cellular neural network (CNN) to handle discrete wavelet transform (DWT) of two-dimensional (2D) images. The paradigm has two distinct features, namely, it can simultaneously calculate the four subband images in DWT and it can embed the DWT and a subsequent image processor into a single CNN. Thirdly, we proposed a configurable digital cellular neural network (DCNN) with decomposable templates to process grey-scale images. CNN is a highly parallel analog circuit suitable for real-time image processing. However, the needs of preprocessors in real-world applications usually destroy the parallel processing nature of CNN. In Chapter 3, we propose methods to embed the function of a preprocessor into a single CNN universal machine (CNN-UM) such that the novel CNN contains the functions of both the preprocessor and the original CNN. The embedding processes are divided into three categories according to the structure of the B template. Case 1 is the general case. Since the unbiased method for embedding requires special CNN-UMs that support twice the neighborhood size, we proposed three approximation methods (AMs) that require only regular CNN-UM. Among the three approximation methods, AM I and AM III are comparable, while AM III shows a slightly better performance (lower ER) than AM I in the general case. Both of them outperform AM II. In Cheater 4, a novel CNN paradigm is proposed to cope with the above problems raised when one intends to include wavelet transform in the processing of images using CNN. Two sets of experiments were designated to test the performance of the WECNN, namely feature extraction in (1) the LL subband image and (2) the combination of different subband images. In the first set of experiments, the combination of DB53 wavelet filter and the r=2 halftoning processor was determined to be most appropriate for low-resolution applications such as in a tactile/vision substitution system (TVSS) for the blind. The second set of experiments demonstrated the capacity of the paradigm in the extraction of features from multi-subband images. The proposed edge detector is capable of delineating more subtle details than using typical CNN edge detector alone, and is more robust in dealing with low-contrast images than traditional edge detectors. In Cheater 5, we propose a digital CNN emulator which provides a powerful and integrated platform for the development and testing of novel CNN applications. The configurable DCNN with decomposable templates is presented to process grey-scale images with a lower complexity of O( ). When the proposed structure is implemented on an embedded system with Virtex II XC2V2000 chip, the DCNN is realized with 48014 gates at 18-bit data format and running at an optimal working frequency of 90 MHz. The result outperforms the other approach for DCNN realization with much reduced gate counts and higher working frequency.

This project proposes an hardware implementation of a CNN (Cellular Neural Network), a type of neural network that emphasizes locality by allowing each cell to be directly connected only to other cells in its vicinity. This reduces the complexity related to the interconnections between cells, and allows the CNN to be a very suitable architecture for hardware implementation.
The concept of Cellular Neural Networks was proposed by Leon O. Chua and Lin Yang in 1988, and it has ever since seen only relatively modest but promising improvements. The goal of this work is to delve further into this subject, which we consider to be largely unexplored, and to try to make a positive contribution to it. We also consider that these advancements fill an important gap in the context of the recent breakthrough we have been witnessing in the field of artificial intelligence and data fusion.

This project proposes an hardware implementation of a CNN (Cellular Neural Network), a type of neural network that emphasizes locality by allowing each cell to be directly connected only to other cells in its vicinity. This reduces the complexity related to the interconnections between cells, and allows the CNN to be a very suitable architecture for hardware implementation.
The concept of Cellular Neural Networks was proposed by Leon O. Chua and Lin Yang in 1988, and it has ever since seen only relatively modest but promising improvements. The goal of this work is to delve further into this subject, which we consider to be largely unexplored, and to try to make a positive contribution to it. We also consider that these advancements fill an important gap in the context of the recent breakthrough we have been witnessing in the field of artificial intelligence and data fusion.

碩士
國立交通大學
應用數學系
90
In this paper, we study a cellular neural network(CNN) with a quadratic term x^2 in the integer lattice Z^2 on the plane R^2. We impose a symmetric coupling between nearest neighbors and use two parameters to describe the weight between such interacting cells. The existence and stability of mosaic patterns are been proved. We also investigate the relationship between mosaic patterns and the parameter space.

碩士
義守大學
資訊工程學系碩士班
93
Active contour model (ACM) is an important contour segmentation method for image processing especially for low contrast and high noised image such as medical and remote sensing images. Traditional methods adopt the combination of internal and external energies, which is combination of optimized so as to determine the contour. Such methods require sequential computations composed of a vest amount of convolution-wise operations with if-then optimization branchings. Therefore, it is time consuming. This project proposes a cellular neural network method to implement ACM, which provides parallel computation in the CNN universal machine (CNN-UM) architecture. The proposed CNN method can follow the local rules of the optimization in searching the contour. Moreover, it takes into account the global properties of images to introduce noise-proof mechanism, which is intrinsic done by propagation. As a consequence faster computation can be achieved and robust segmentation can be obtained against to low contrasts and high noises.