Various algorithms for optimization and learning in adaptive systems

This thesis discusses various methods for learning and optimization in adaptive systems. Overall, it emphasizes the relationship between optimization, learning, and adaptive systems; and it illustrates the influence of underlying hardware upon the construction of efficient algorithms for learning and optimization. Chapter 1 provides a summary and an overview.

Chapter 2 discusses a method for using feed-forward neural networks to filter the noise out of noise-corrupted signals. The networks use back-propagation learning, but they use it in a way that qualifies as unsupervised learning. The networks adapt based only on the raw input data-there are no external teachers providing information on correct operation during training. The chapter contains an analysis of the learning and develops a simple expression that, based only on the geometry of the network, predicts performance.

Chapter 3 explains a simple model of the piriform cortex, an area in the brain involved in the processing of olfactory information. The model was used to explore the possible effect of acetylcholine on learning and on odor classification. According to the model, the piriform cortex can classify odors better when acetylcholine is present during learning but not present during recall. This is interesting since it suggests that learning and recall might be separate neurochemical modes (corresponding to whether or not acetylcholine is present). When acetylcholine is turned off at all times, even during learning, the model exhibits behavior somewhat similar to Alzheimer's disease, a disease associated with the degeneration of cells that distribute acetylcholine.

Chapters 4, 5, and 6 discuss algorithms appropriate for adaptive systems implemented entirely in analog hardware. The algorithms inject noise into the systems and correlate the noise with the outputs of the systems. This allows them to estimate gradients and to implement noisy versions of gradient descent, without having to calculate gradients explicitly. The methods require only noise generators, adders, multipliers, integrators, and differentiators; and the number of devices needed scales linearly with the number of adjustable parameters in the adaptive systems. With the exception of one global signal, the algorithms require only local information exchange.

No Coração das Trevas: o paraíso e inferno do outro em Bernardo Carvalho e Joseph Conrad

Esta dissertação estuda o papel do sujeito na literatura e sua relação com a cultura e alteridade através da análise de duas obras: Nove noites, de Bernardo de Carvalho e Coração das Trevas de Joseph Conrad. As obras estudadas mostram a crise que atinge os protagonistas dos dois livros depois do encontro com outras culturas. Em Nove noites o outro é representado pelo índio e em Coração das Trevas pelos africanos. Em Nove noites o antropólogo Buell Quain se suicida depois de uma estada entre os índios Krahô, e em Coração das Trevas vemos a deterioração do homem branco representada pelo personagem de Kurtz. Considerado um homem notável e um altruísta na Europa, Kurtz teria se corrompido no contato com a realidade do Congo e se torna, nas palavras do narrador Marlow, um dos demônios da terra. A dissolução da personalidade e código moral do homem branco, representada pelos dois personagens, será estudada analisando a relação entre personalidade e cultura e como a falta de apoio e controle grupal desarticula valores até então considerados estáveis, assim como o contato com o outro. Esta desarticulação do sujeito causada pelo choque cultural se soma à crise geral do sujeito moderno e ao mal-estar na civilização, como descrito por Freud. A posição paradoxal do antropólogo, que se situa entre duas culturas, faz parte desta análise, do mesmo modo questões pertinentes a posição dos índios e africanos no Congo. No caso específico de Coração das Trevas trabalha-se a interseção entre a análise do sujeito, e suas implicações, e a construção do personagem de Kurtz como símbolo da violência colonial. O trabalho analisa também as semelhanças entre as duas obras, tanto temáticas como em suas técnicas narrativas e a influência da obra de Conrad nos romances de Carvalho

A Reliable Method for Rhythm Analysis during Cardiopulmonary Resuscitation

nterruptions in cardiopulmonary resuscitation (CPR) compromise defibrillation success. However, CPR must be interrupted to analyze the rhythm because although current methods for rhythm analysis during CPR have high sensitivity for shockable rhythms, the specificity for nonshockable rhythms is still too low. This paper introduces a new approach to rhythm analysis during CPR that combines two strategies: a state-of-the-art CPR artifact suppression filter and a shock advice algorithm (SAA) designed to optimally classify the filtered signal. Emphasis is on designing an algorithm with high specificity. The SAA includes a detector for low electrical activity rhythms to increase the specificity, and a shock/no-shock decision algorithm based on a support vector machine classifier using slope and frequency features. For this study, 1185 shockable and 6482 nonshockable 9-s segments corrupted by CPR artifacts were obtained from 247 patients suffering out-of-hospital cardiac arrest. The segments were split into a training and a test set. For the test set, the sensitivity and specificity for rhythm analysis during CPR were 91.0% and 96.6%, respectively. This new approach shows an important increase in specificity without compromising the sensitivity when compared to previous studies.

Identificação de danos em estruturas bi-dimensionais via matriz de flexibilidade baseado em um modelo de dano contínuo

No presente trabalho, o modelo de identificação de danos apresentado por Stutz et al. (2005) é utilizado. A contribuição do presente trabalho consiste em avaliar alguns pontos da identificação de danos em vigas e, em seguida, expandir o modelo para identificar danos estruturais em placas. Uma avaliação do comportamento das frequências naturais e da matriz de flexibilidade para uma viga de Euler-Bernoulli simplesmente apoiada, na presença de danos simulados pelo parâmetro de coesão é realizada. Essa análise, permite também o conhecimento das regiões onde há maior sensibilidade ao dano, ajudando a traçar estratégias para melhorar a identificação de danos em regiões que sofrem poucas alterações na presença de falhas estruturais. Comparou-se o comportamento dos dois primeiros modos de vibração da viga simplesmente apoiada na presença de um dano estrutural, com os dois primeiros modos de vibração da estrutura intacta e corrompidos por ruído. Diversos métodos de localização de danos e de otimização são avaliados na tentativa de identificar os danos simulados através do campo de danos proposto por Stutz et al. (2005) na presença de dados ruidosos. Após a apresentação de resultados da identificação de danos obtidos para uma viga de Euler-Bernoulli simplesmente apoiada, uma análise do comportamento das frequências naturais e da matriz de flexibilidade de uma viga de Euler- Bernoulli engastada na presença de danos é apresentada, assim como os resultados de identificação de danos considerando-se diversos cenários e níveis de ruído. Uma importante contribuição do presente trabalho consiste em propor um método de identificação de danos via matriz de flexibilidade onde o campo de defeitos para a placa de Kirchoff é modelado via MEF. Uma análise do comportamento da matriz de flexibilidade devido à presença de danos na placa é apresentada, assim como os resultados numéricos da identificação de danos estruturais com e sem a presença de dados ruidosos. Com a finalidade de reduzir o custo computacional na identificação de danos em estruturas complexas, uma hibridização entre o método de otimização por enxame de particulas (PSO, do inglês, Particle Swarm Optimization) e o método de otimização Levenberg-Marquardt é proposta. Resultados numéricos da hibridização para uma estrutura do tipo placa são apresentados.

Elementos finitos em fluidos dominados pelo fenômeno de advecção: um método semi-Lagrangeano.

Os escoamentos altamente convectivos representam um desafio na simulação pelo método de elementos finitos. Com a solução de elementos finitos de Galerkin para escoamentos incompressíveis, a matriz associada ao termo convectivo é não simétrica, e portanto, a propiedade de aproximação ótima é perdida. Na prática as soluções apresentam oscilações espúrias. Muitos métodos foram desenvolvidos com o fim de resolver esse problema. Neste trabalho apresentamos um método semi- Lagrangeano, o qual é implicitamente um método do tipo upwind, que portanto resolve o problema anterior, e comparamos o desempenho do método na solução das equações de convecção-difusão e Navier-Stokes incompressível com o Streamline Upwind Petrov Galerkin (SUPG), um método estabilizador de reconhecido desempenho. No SUPG, as funções de forma e de teste são tomadas em espaços diferentes, criando um efeito tal que as oscilações espúrias são drasticamente atenuadas. O método semi-Lagrangeano é um método de fator de integração, no qual o fator é um operador de convecção que se desloca para um sistema de coordenadas móveis no fluido, mas restabelece o sistema de coordenadas Lagrangeanas depois de cada passo de tempo. Isto prevê estabilidade e a possibilidade de utilizar passos de tempo maiores.Existem muitos trabalhos na literatura analisando métodos estabilizadores, mas não assim com o método semi-Lagrangeano, o que representa a contribuição principal deste trabalho: reconhecer as virtudes e as fraquezas do método semi-Lagrangeano em escoamentos dominados pelo fenômeno de convecção.

Extended VTS for noise-robust speech recognition

Model compensation is a standard way of improving the robustness of speech recognition systems to noise. A number of popular schemes are based on vector Taylor series (VTS) compensation, which uses a linear approximation to represent the influence of noise on the clean speech. To compensate the dynamic parameters, the continuous time approximation is often used. This approximation uses a point estimate of the gradient, which fails to take into account that dynamic coefficients are a function of a number of consecutive static coefficients. In this paper, the accuracy of dynamic parameter compensation is improved by representing the dynamic features as a linear transformation of a window of static features. A modified version of VTS compensation is applied to the distribution of the window of static features and, importantly, their correlations. These compensated distributions are then transformed to distributions over standard static and dynamic features. With this improved approximation, it is also possible to obtain full-covariance corrupted speech distributions. This addresses the correlation changes that occur in noise. The proposed scheme outperformed the standard VTS scheme by 10% to 20% relative on a range of tasks. © 2006 IEEE.

Structured discriminative models for noise robust continuous speech recognition

Recently there has been interest in structured discriminative models for speech recognition. In these models sentence posteriors are directly modelled, given a set of features extracted from the observation sequence, and hypothesised word sequence. In previous work these discriminative models have been combined with features derived from generative models for noise-robust speech recognition for continuous digits. This paper extends this work to medium to large vocabulary tasks. The form of the score-space extracted using the generative models, and parameter tying of the discriminative model, are both discussed. Update formulae for both conditional maximum likelihood and minimum Bayes' risk training are described. Experimental results are presented on small and medium to large vocabulary noise-corrupted speech recognition tasks: AURORA 2 and 4. © 2011 IEEE.

Accurate real-time feedback of surface EMG during fMRI.

Real-time acquisition of EMG during functional MRI (fMRI) provides a novel method of controlling motor experiments in the scanner using feedback of EMG. Because of the redundancy in the human muscle system, this is not possible from recordings of joint torque and kinematics alone, because these provide no information about individual muscle activation. This is particularly critical during brain imaging because brain activations are not only related to joint torques and kinematics but are also related to individual muscle activation. However, EMG collected during imaging is corrupted by large artifacts induced by the varying magnetic fields and radio frequency (RF) pulses in the scanner. Methods proposed in literature for artifact removal are complex, computationally expensive, and difficult to implement for real-time noise removal. We describe an acquisition system and algorithm that enables real-time acquisition for the first time. The algorithm removes particular frequencies from the EMG spectrum in which the noise is concentrated. Although this decreases the power content of the EMG, this method provides excellent estimates of EMG with good resolution. Comparisons show that the cleaned EMG obtained with the algorithm is, like actual EMG, very well correlated with joint torque and can thus be used for real-time visual feedback during functional studies.

Derivative Kernels for Noise Robust ASR

Recently there has been interest in combined gen- erative/discriminative classifiers. In these classifiers features for the discriminative models are derived from generative kernels. One advantage of using generative kernels is that systematic approaches exist how to introduce complex dependencies beyond conditional independence assumptions. Furthermore, by using generative kernels model-based compensation/adaptation tech- niques can be applied to make discriminative models robust to noise/speaker conditions. This paper extends previous work with combined generative/discriminative classifiers in several directions. First, it introduces derivative kernels based on context- dependent generative models. Second, it describes how derivative kernels can be incorporated in continuous discriminative models. Third, it addresses the issues associated with large number of classes and parameters when context-dependent models and high- dimensional features of derivative kernels are used. The approach is evaluated on two noise-corrupted tasks: small vocabulary AURORA 2 and medium-to-large vocabulary AURORA 4 task.

Improving reverberant VTS for hands-free robust speech recognition

Model-based approaches to handling additive background noise and channel distortion, such as Vector Taylor Series (VTS), have been intensively studied and extended in a number of ways. In previous work, VTS has been extended to handle both reverberant and background noise, yielding the Reverberant VTS (RVTS) scheme. In this work, rather than assuming the observation vector is generated by the reverberation of a sequence of background noise corrupted speech vectors, as in RVTS, the observation vector is modelled as a superposition of the background noise and the reverberation of clean speech. This yields a new compensation scheme RVTS Joint (RVTSJ), which allows an easy formulation for joint estimation of both additive and reverberation noise parameters. These two compensation schemes were evaluated and compared on a simulated reverberant noise corrupted AURORA4 task. Both yielded large gains over VTS baseline system, with RVTSJ outperforming the previous RVTS scheme. © 2011 IEEE.

Derivative kernels for noise robust ASR

Recently there has been interest in combining generative and discriminative classifiers. In these classifiers features for the discriminative models are derived from the generative kernels. One advantage of using generative kernels is that systematic approaches exist to introduce complex dependencies into the feature-space. Furthermore, as the features are based on generative models standard model-based compensation and adaptation techniques can be applied to make discriminative models robust to noise and speaker conditions. This paper extends previous work in this framework in several directions. First, it introduces derivative kernels based on context-dependent generative models. Second, it describes how derivative kernels can be incorporated in structured discriminative models. Third, it addresses the issues associated with large number of classes and parameters when context-dependent models and high-dimensional feature-spaces of derivative kernels are used. The approach is evaluated on two noise-corrupted tasks: small vocabulary AURORA 2 and medium-to-large vocabulary AURORA 4 task. © 2011 IEEE.

Extending noise robust structured support vector machines to larger vocabulary tasks

This paper describes a structured SVM framework suitable for noise-robust medium/large vocabulary speech recognition. Several theoretical and practical extensions to previous work on small vocabulary tasks are detailed. The joint feature space based on word models is extended to allow context-dependent triphone models to be used. By interpreting the structured SVM as a large margin log-linear model, illustrates that there is an implicit assumption that the prior of the discriminative parameter is a zero mean Gaussian. However, depending on the definition of likelihood feature space, a non-zero prior may be more appropriate. A general Gaussian prior is incorporated into the large margin training criterion in a form that allows the cutting plan algorithm to be directly applied. To further speed up the training process, 1-slack algorithm, caching competing hypothesis and parallelization strategies are also proposed. The performance of structured SVMs is evaluated on noise corrupted medium vocabulary speech recognition task: AURORA 4. © 2011 IEEE.

Gaussian Process training with input noise

In standard Gaussian Process regression input locations are assumed to be noise free. We present a simple yet effective GP model for training on input points corrupted by i.i.d. Gaussian noise. To make computations tractable we use a local linear expansion about each input point. This allows the input noise to be recast as output noise proportional to the squared gradient of the GP posterior mean. The input noise variances are inferred from the data as extra hyperparameters. They are trained alongside other hyperparameters by the usual method of maximisation of the marginal likelihood. Training uses an iterative scheme, which alternates between optimising the hyperparameters and calculating the posterior gradient. Analytic predictive moments can then be found for Gaussian distributed test points. We compare our model to others over a range of different regression problems and show that it improves over current methods.

Factor analysis based VTS discriminative adaptive training

Vector Taylor Series (VTS) model based compensation is a powerful approach for noise robust speech recognition. An important extension to this approach is VTS adaptive training (VAT), which allows canonical models to be estimated on diverse noise-degraded training data. These canonical model can be estimated using EM-based approaches, allowing simple extensions to discriminative VAT (DVAT). However to ensure a diagonal corrupted speech covariance matrix the Jacobian (loading matrix) relating the noise and clean speech is diagonalised. In this work an approach for yielding optimal diagonal loading matrices based on minimising the expected KL-divergence between the diagonal loading matrix and "correct" distributions is proposed. The performance of DVAT using the standard and optimal diagonalisation was evaluated on both in-car collected data and the Aurora4 task. © 2012 IEEE.

State estimation schemes for independent component coupled hidden Markov models

Conventional Hidden Markov models generally consist of a Markov chain observed through a linear map corrupted by additive noise. This general class of model has enjoyed a huge and diverse range of applications, for example, speech processing, biomedical signal processing and more recently quantitative finance. However, a lesser known extension of this general class of model is the so-called Factorial Hidden Markov Model (FHMM). FHMMs also have diverse applications, notably in machine learning, artificial intelligence and speech recognition [13, 17]. FHMMs extend the usual class of HMMs, by supposing the partially observed state process is a finite collection of distinct Markov chains, either statistically independent or dependent. There is also considerable current activity in applying collections of partially observed Markov chains to complex action recognition problems, see, for example, [6]. In this article we consider the Maximum Likelihood (ML) parameter estimation problem for FHMMs. Much of the extant literature concerning this problem presents parameter estimation schemes based on full data log-likelihood EM algorithms. This approach can be slow to converge and often imposes heavy demands on computer memory. The latter point is particularly relevant for the class of FHMMs where state space dimensions are relatively large. The contribution in this article is to develop new recursive formulae for a filter-based EM algorithm that can be implemented online. Our new formulae are equivalent ML estimators, however, these formulae are purely recursive and so, significantly reduce numerical complexity and memory requirements. A computer simulation is included to demonstrate the performance of our results. © Taylor & Francis Group, LLC.