Adaptive back-propagation in on-line learning of multilayer networks

An adaptive back-propagation algorithm is studied and compared with gradient descent (standard back-propagation) for on-line learning in two-layer neural networks with an arbitrary number of hidden units. Within a statistical mechanics framework, both numerical studies and a rigorous analysis show that the adaptive back-propagation method results in faster training by breaking the symmetry between hidden units more efficiently and by providing faster convergence to optimal generalization than gradient descent.

Projecto de um regulador linear de tensão: LDO

Os reguladores de tensão LDO são utilizados intensivamente na actual indústria de electrónica, são uma parte essencial de um bloco de gestão de potência para um SoC. O aumento de produtos portáteis alimentados por baterias levou ao crescimento de soluções totalmente integradas, o que degrada o rendimento dos blocos analógicos que o constituem face às perturbações introduzidas na alimentação. Desta forma, surge a necessidade de procurar soluções cada vez mais optimizadas, impondo assim novas soluções, e/ou melhoramentos dos circuitos de gestão de potência, tendo como objectivo final o aumento do desempenho e da autonomia dos dispositivos electrónicos. Normalmente este tipo de reguladores tem a corrente de saída limitada, devido a problemas de estabilidade associados. Numa tentativa de evitar a instabilidade para as correntes de carga definidas e aumentar o PSRR do mesmo, é apresentado um método de implementação que tem como objectivo melhorar estas características, em que se pretende aumentar o rendimento e melhorar a resposta à variação da carga. No entanto, a técnica apresentada utiliza polarização adaptativa do estágio de potência, o que implica um aumento da corrente de consumo. O regulador LDO foi implementado na tecnologia CMOS UMC 0.18μm e ocupa uma área inferior a 0,2mm2. Os resultados da simulação mostram que o mesmo suporta uma transição de corrente 10μA para 100mA, com uma queda de tensão entre a tensão de alimentação e a tensão de saída inferior a 200mV. A estabilidade é assegurada para todas as correntes de carga. O tempo de estabelecimento é inferior a 6μs e as variações da tensão de saída relativamente a seu valor nominal são inferiores a 5mV. A corrente de consumo varia entre os 140μA até 200μA, o que permite atingir as especificações proposta para um PSRR de 40dB@10kHz.

Time standards of nursing in Primary Health Care: an observational study

Abstract OBJECTIVE To determine time standards for interventions and activities conducted by nursing professionals in Family Health Units (FHU) in Brazil to substantiate the calculation of work force. METHOD This was an observational study carried out in 27 FHU, in 12 municipalities in 10 states, in 2013. In each unit, nursing professionals were observed every 10 minutes, for eight work hours, on five consecutive days via the work sampling technique. RESULTS A total of 32,613 observations were made, involving 47 nurses and 93 nursing technicians/assistants. Appointments were the main intervention carried out by nurses, with a mean time of 25.3 minutes, followed by record-keeping, which corresponded to 9.7%. On average, nursing technicians/assistants spent 6.3% of their time keeping records and 30.6 intervention minutes on immunization/vaccination control. CONCLUSION The study resulted in standard times of interventions carried out by the FHU nursing team, which can underpin the determination of nursing staff size and human resource policies. Furthermore, the study showed the panorama of interventions currently employed, allowing for the work process to be reviewed and optimized.

Improving sampling, optimization and feature extraction in Boltzmann machines

L’apprentissage supervisé de réseaux hiérarchiques à grande échelle connaît présentement un succès fulgurant. Malgré cette effervescence, l’apprentissage non-supervisé représente toujours, selon plusieurs chercheurs, un élément clé de l’Intelligence Artificielle, où les agents doivent apprendre à partir d’un nombre potentiellement limité de données. Cette thèse s’inscrit dans cette pensée et aborde divers sujets de recherche liés au problème d’estimation de densité par l’entremise des machines de Boltzmann (BM), modèles graphiques probabilistes au coeur de l’apprentissage profond. Nos contributions touchent les domaines de l’échantillonnage, l’estimation de fonctions de partition, l’optimisation ainsi que l’apprentissage de représentations invariantes. Cette thèse débute par l’exposition d’un nouvel algorithme d'échantillonnage adaptatif, qui ajuste (de fa ̧con automatique) la température des chaînes de Markov sous simulation, afin de maintenir une vitesse de convergence élevée tout au long de l’apprentissage. Lorsqu’utilisé dans le contexte de l’apprentissage par maximum de vraisemblance stochastique (SML), notre algorithme engendre une robustesse accrue face à la sélection du taux d’apprentissage, ainsi qu’une meilleure vitesse de convergence. Nos résultats sont présent ́es dans le domaine des BMs, mais la méthode est générale et applicable à l’apprentissage de tout modèle probabiliste exploitant l’échantillonnage par chaînes de Markov. Tandis que le gradient du maximum de vraisemblance peut-être approximé par échantillonnage, l’évaluation de la log-vraisemblance nécessite un estimé de la fonction de partition. Contrairement aux approches traditionnelles qui considèrent un modèle donné comme une boîte noire, nous proposons plutôt d’exploiter la dynamique de l’apprentissage en estimant les changements successifs de log-partition encourus à chaque mise à jour des paramètres. Le problème d’estimation est reformulé comme un problème d’inférence similaire au filtre de Kalman, mais sur un graphe bi-dimensionnel, où les dimensions correspondent aux axes du temps et au paramètre de température. Sur le thème de l’optimisation, nous présentons également un algorithme permettant d’appliquer, de manière efficace, le gradient naturel à des machines de Boltzmann comportant des milliers d’unités. Jusqu’à présent, son adoption était limitée par son haut coût computationel ainsi que sa demande en mémoire. Notre algorithme, Metric-Free Natural Gradient (MFNG), permet d’éviter le calcul explicite de la matrice d’information de Fisher (et son inverse) en exploitant un solveur linéaire combiné à un produit matrice-vecteur efficace. L’algorithme est prometteur: en terme du nombre d’évaluations de fonctions, MFNG converge plus rapidement que SML. Son implémentation demeure malheureusement inefficace en temps de calcul. Ces travaux explorent également les mécanismes sous-jacents à l’apprentissage de représentations invariantes. À cette fin, nous utilisons la famille de machines de Boltzmann restreintes “spike & slab” (ssRBM), que nous modifions afin de pouvoir modéliser des distributions binaires et parcimonieuses. Les variables latentes binaires de la ssRBM peuvent être rendues invariantes à un sous-espace vectoriel, en associant à chacune d’elles, un vecteur de variables latentes continues (dénommées “slabs”). Ceci se traduit par une invariance accrue au niveau de la représentation et un meilleur taux de classification lorsque peu de données étiquetées sont disponibles. Nous terminons cette thèse sur un sujet ambitieux: l’apprentissage de représentations pouvant séparer les facteurs de variations présents dans le signal d’entrée. Nous proposons une solution à base de ssRBM bilinéaire (avec deux groupes de facteurs latents) et formulons le problème comme l’un de “pooling” dans des sous-espaces vectoriels complémentaires.

Mobile-to-mobile MIMO transmit-receive diversity systems: analysis and performance in three-dimensional double-correlated channels

Mobile-to-mobile (M-to-M) communications are expected to play a crucial role in future wireless systems and networks. In this paper, we consider M-to-M multiple-input multiple-output (MIMO) maximal ratio combining system and assess its performance in spatially correlated channels. The analysis assumes double-correlated Rayleigh-and-Lognormal fading channels and is performed in terms of average symbol error probability, outage probability, and ergodic capacity. To obtain the receive and transmit spatial correlation functions needed for the performance analysis, we used a three-dimensional (3D) M-to-M MIMO channel model, which takes into account the effects of fast fading and shadowing. The expressions for the considered metrics are derived as a function of the average signal-to-noise ratio per receive antenna in closed-form and are further approximated using the recursive adaptive Simpson quadrature method. Numerical results are provided to show the effects of system parameters, such as distance between antenna elements, maximum elevation angle of scatterers, orientation angle of antenna array in the x–y plane, angle between the x–y plane and the antenna array orientation, and degree of scattering in the x–y plane, on the system performance. Copyright © 2011 John Wiley & Sons, Ltd.

Avaliação da reabsorção radicular após a movimentação dentária induzida com forças contínua e contínua interrompida: análise histomorfométrica em ratos

Pós-graduação em Odontologia - FOA

Um método para determinação de pontos de operação com diversidade em linha digital de assinante usando balanceamento de espectro e algoritmo evolucionário

Este trabalho apresenta um método para encontrar um conjunto de pontos de operação, os quais são ótimos de Pareto com diversidade, para linhas digitais de assinante (DSL - digital subscriber line). Em diversos trabalhos encontrados na literatura, têm sido propostos algoritmos para otimização da transmissão de dados em linhas DSL, que fornecem como resultado apenas um ponto de operação para os modems. Esses trabalhos utilizam, em geral, algoritmos de balanceamento de espectro para resolver um problema de alocação de potência, o que difere da abordagem apresentada neste trabalho. O método proposto, chamado de diverseSB , utiliza um processo híbrido composto de um algoritmo evolucionário multiobjetivo (MOEA - multi-objective evolutionary algorithm), mais precisamente, um algoritmo genético com ordenamento por não-dominância (NSGA-II - Non-Dominated Sorting Genetic Algorithm II), e usando ainda, um algoritmo de balanceamento de espectro. Os resultados obtidos por simulações mostram que, para uma dada diversidade, o custo computacional para determinar os pontos de operação com diversidade usando o algoritmo diverseSB proposto é muito menor que métodos de busca de “força bruta”. No método proposto, o NSGA-II executa chamadas ao algoritmo de balanceamento de espectro adotado, por isso, diversos testes envolvendo o mesmo número de chamadas ao algoritmo foram realizadas com o método diverseSB proposto e o método de busca por força bruta, onde os resultados obtidos pelo método diverseSB proposto foram bem superiores do que os resultados do método de busca por força bruta. Por exemplo, o método de força bruta realizando 1600 chamadas ao algoritmo de balanceamento de espectro, obtém um conjunto de pontos de operação com diversidade semelhante ao do método diverseSB proposto com 535 chamadas.

Multigradient field-active contour model for multilayer boundary detection of ultrasound rectal wall image

Extraction and reconstruction of rectal wall structures from an ultrasound image is helpful for surgeons in rectal clinical diagnosis and 3-D reconstruction of rectal structures from ultrasound images. The primary task is to extract the boundary of the muscular layers on the rectal wall. However, due to the low SNR from ultrasound imaging and the thin muscular layer structure of the rectum, this boundary detection task remains a challenge. An active contour model is an effective high-level model, which has been used successfully to aid the tasks of object representation and recognition in many image-processing applications. We present a novel multigradient field active contour algorithm with an extended ability for multiple-object detection, which overcomes some limitations of ordinary active contour models—"snakes." The core part in the algorithm is the proposal of multigradient vector fields, which are used to replace image forces in kinetic function for alternative constraints on the deformation of active contour, thereby partially solving the initialization limitation of active contour for rectal wall boundary detection. An adaptive expanding force is also added to the model to help the active contour go through the homogenous region in the image. The efficacy of the model is explained and tested on the boundary detection of a ring-shaped image, a synthetic image, and an ultrasound image. The experimental results show that the proposed multigradient field-active contour is feasible for multilayer boundary detection of rectal wall

Estimating buffer overflows in three stages using cross-entropy

In this paper we propose a fast adaptive Importance Sampling method for the efficient simulation of buffer overflow probabilities in queueing networks. The method comprises three stages. First we estimate the minimum Cross-Entropy tilting parameter for a small buffer level; next, we use this as a starting value for the estimation of the optimal tilting parameter for the actual (large) buffer level; finally, the tilting parameter just found is used to estimate the overflow probability of interest. We recognize three distinct properties of the method which together explain why the method works well; we conjecture that they hold for quite general queueing networks. Numerical results support this conjecture and demonstrate the high efficiency of the proposed algorithm.

EFICÁCIA ADAPTATIVA DE ADOLESCENTES UNIVERSITÁRIOS TRABALHADORES

Simon (1989) define adaptação como o conjunto de respostas de um organismo vivo, em vários momentos, a situações que o modificam, permitindo manutenção de sua organização compatível com a vida. O presente trabalho teve como objetivo avaliar a eficácia adaptativa de adolescentes universitários; verificar possíveis diferenças na qualidade da adaptação entre adolescentes universitários trabalhadores e não trabalhadores; e identificar, por meio da EDAO, os fatores presentes e determinantes da qualidade da adaptação de um adolescente trabalhador e um não trabalhador. Para tanto, foi aplicado primeiramente o Questionário Diagnóstico Adaptativo Operacionalizado em 144 alunos ingressantes de uma universidade particular da região do grande ABC, de ambos os gêneros e períodos, sendo 115 do gênero feminino e 29 do masculino e 85 do período matutino e 59 do noturno, de 17 a 20 anos de idade, com 46 que trabalham e estudam e 98 que apenas estudam. Após a avaliação do questionário, foram escolhidas duas alunas para a aplicação da Escala Diagnóstica Adaptativa Operacionalizada. A primeira, com 19 anos, estudante do período matutino e não trabalhava, foi escolhida por apresentar as médias mais baixas do grupo, apresentando indícios de adaptação ineficaz, e a segunda por ser uma universitária trabalhadora, de 18 anos, com indicativo de adaptação eficaz. Os resultados mostraram que não existe diferença estatisticamente significante na qualidade da adaptação ao considerar-se as variáveis setor da adaptação, trabalho, idade, gênero e curso, com exceção da variável gênero no setor Sócio-Cultural, no qual foi evidenciado que as mulheres estão significativamente mais adequadas que os rapazes. Os dados revelaram que 100 participantes (69,4%) obtiveram 3 pontos, e tiveram como classificação diagnóstica Adaptação Ineficaz Moderada (Grupo 3), 27 (18,8%) obtiveram 4 pontos e foram classificados com Adaptação Ineficaz Leve (Grupo 2), 17 (11,8%) obtiveram 5 pontos e foram classificados com Adaptação Eficaz (Grupo 1) e nenhum participante foi classificado no Grupo 4 ou no Grupo 5. As duas aplicações da EDAO mostraram as mesmas adequações de respostas nos setores Afetivo-Relacional, Produtividade e Sócio-Cultural, assim como mas apresentou diferenças no setor Orgânico. Foi verificado que a rotina de trabalhar e estudar não é um fator que contribui para a ineficácia adaptativa, mas sim os sentimentos que envolvem estudos e atividade laboral. Ao revelar correlações significativas e positivas, esta pesquisa comprovou a afirmação de Simon (2005) de que os setores adaptativos interagem. Com base nos trabalhos de Simon (1989, 2005) e Gandini (1996), a presente pesquisa contribuiu com a aplicação da avaliação da eficácia adaptativa ao elaborar um método para avaliar a eficácia adaptativa de uma população maior em um único momento do tempo.

Mitigation of pattern sensitivity in full-field electronic dispersion compensation

We investigate the pattern-dependent decoding failures in full-field electronic dispersion compensation (EDC) by offline processing of experimental signals, and find that the performance of such an EDC receiver may be degraded by an isolated "1" bit surrounded by long strings of consecutive "0s". By reducing the probability of occurrence of this kind of isolated "1" and using a novel adaptive threshold decoding method, we greatly improve the compensation performance to achieve 10-Gb/s on-off keyed signal transmission over 496-km field-installed single-mode fiber without optical dispersion compensation.

Structural optimization

The non-linear programming algorithms for the minimum weight design of structural frames are presented in this thesis. The first, which is applied to rigidly jointed and pin jointed plane frames subject to deflexion constraints, consists of a search in a feasible design space. Successive trial designs are developed so that the feasibility and the optimality of the designs are improved simultaneously. It is found that this method is restricted lo the design of structures with few unknown variables. The second non-linear programming algorithm is presented .in a general form. This consists of two types of search, one improving feasibility and the other optimality. The method speeds up the 'feasible direction' approach by obtaining a constant weight direction vector that is influenced by dominating constraints. For pin jointed plane and space frames this method is used to obtain a 'minimum weight' design which satisfies restrictions on stresses and deflexions. The matrix force method enables the design requirements to be expressed in a general form and the design problem is automatically formulated within the computer. Examples are given to explain the method and the design criteria are extended to include member buckling. Fundamental theorems are proposed and proved to confirm that structures are inter-related. These theorems are applicable to linear elastic structures and facilitate the prediction of the behaviour of one structure from the results of analysing another, more general, or related structure. It becomes possible to evaluate the significance of each member in the behaviour of a structure and the problem of minimum weight design is extended to include shape. A method is proposed to design structures of optimum shape with stress and deflexion limitations. Finally a detailed investigation is carried out into the design of structures to study the factors that influence their shape.

Mitigation of pattern sensitivity in full-field electronic dispersion compensation

We investigate the pattern-dependent decoding failures in full-field electronic dispersion compensation (EDC) by offline processing of experimental signals, and find that the performance of such an EDC receiver may be degraded by an isolated "1" bit surrounded by long strings of consecutive "0s". By reducing the probability of occurrence of this kind of isolated "1" and using a novel adaptive threshold decoding method, we greatly improve the compensation performance to achieve 10-Gb/s on-off keyed signal transmission over 496-km field-installed single-mode fiber without optical dispersion compensation.

LES of flame acceleration and DDT in hydrogen-air mixture using artificially thickened flame approach and detailed chemical kinetics

A large eddy simulation is performed to study the deflagration to detonation transition phenomenon in an obstructed channel containing premixed stoichiometric hydrogen–air mixture. Two-dimensional filtered reactive Navier–Stokes equations are solved utilizing the artificially thickened flame approach (ATF) for modeling sub-grid scale combustion. To include the effect of induction time, a 27-step detailed mechanism is utilized along with an in situ adaptive tabulation (ISAT) method to reduce the computational cost due to the detailed chemistry. The results show that in the slow flame propagation regime, the flame–vortex interaction and the resulting flame folding and wrinkling are the main mechanisms for the increase of the flame surface and consequently acceleration of the flame. Furthermore, at high speed, the major mechanisms responsible for flame propagation are repeated reflected shock–flame interactions and the resulting baroclinic vorticity. These interactions intensify the rate of heat release and maintain the turbulence and flame speed at high level. During the flame acceleration, it is seen that the turbulent flame enters the ‘thickened reaction zones’ regime. Therefore, it is necessary to utilize the chemistry based combustion model with detailed chemical kinetics to properly capture the salient features of the fast deflagration propagation.

Sparse time-frequency data analysis : a multi-scale approach

In this work, we further extend the recently developed adaptive data analysis method, the Sparse Time-Frequency Representation (STFR) method. This method is based on the assumption that many physical signals inherently contain AM-FM representations. We propose a sparse optimization method to extract the AM-FM representations of such signals. We prove the convergence of the method for periodic signals under certain assumptions and provide practical algorithms specifically for the non-periodic STFR, which extends the method to tackle problems that former STFR methods could not handle, including stability to noise and non-periodic data analysis. This is a significant improvement since many adaptive and non-adaptive signal processing methods are not fully capable of handling non-periodic signals. Moreover, we propose a new STFR algorithm to study intrawave signals with strong frequency modulation and analyze the convergence of this new algorithm for periodic signals. Such signals have previously remained a bottleneck for all signal processing methods. Furthermore, we propose a modified version of STFR that facilitates the extraction of intrawaves that have overlaping frequency content. We show that the STFR methods can be applied to the realm of dynamical systems and cardiovascular signals. In particular, we present a simplified and modified version of the STFR algorithm that is potentially useful for the diagnosis of some cardiovascular diseases. We further explain some preliminary work on the nature of Intrinsic Mode Functions (IMFs) and how they can have different representations in different phase coordinates. This analysis shows that the uncertainty principle is fundamental to all oscillating signals.