[pt] Esta tese investiga a estimação de curvas de juros sob o ponto de vista de métodos não-paramétricos. O texto está dividido em dois blocos. O primeiro investiga a questão do critério utilizado para selecionar o método de melhor desempenho na tarefa de interpolar a curva de juros brasileira em uma dada amostra. Foi proposto um critério de seleção de método baseado em estratégias de re-amostragem do tipo leave-k-out cross validation, onde K k £ £ 1 e K é função do número de contratos observados a cada curva da amostra. Especificidades do problema reduzem o esforço computacional requerido, tornando o critério factível. A amostra tem freqüência diária: janeiro de 1997 a fevereiro de 2001. O critério proposto apontou o spline cúbico natural -utilizado com método de ajuste perfeito aos dados - como o método de melhor desempenho. Considerando a precisão de negociação, este spline mostrou-se não viesado. A análise quantitativa de seu desempenho identificou, contudo, heterocedasticidades nos erros simulados. A partir da especificação da variância condicional destes erros e de algumas hipóteses, foi proposto um esquema de intervalo de segurança para a estimação de taxas de juros pelo spline cúbico natural, empregado como método de ajuste perfeito aos dados. O backtest sugere que o esquema proposto é consistente, acomodando bem as hipóteses e aproximações envolvidas. O segundo bloco investiga a estimação da curva de juros norte-americana construída a partir dos contratos de swaps de taxas de juros dólar-Libor pela Máquina de Vetores Suporte (MVS), parte do corpo da Teoria do Aprendizado Estatístico. A pesquisa em MVS tem obtido importantes avanços teóricos, embora ainda sejam escassas as implementações em problemas reais de regressão. A MVS possui características atrativas para a modelagem de curva de juros: é capaz de introduzir já na estimação informações a priori sobre o formato da curva e sobre aspectos da formação das taxas e liquidez de cada um dos contratos a partir dos quais ela é construída. Estas últimas são quantificadas pelo bid-ask spread (BAS) de cada contrato. A formulação básica da MVS é alterada para assimilar diferentes valores do BAS sem que as propriedades dela sejam perdidas. É dada especial atenção ao levantamento de informação a priori para seleção dos parâmetros da MVS a partir do formato típico da curva. A amostra tem freqüência diária: março de 1997 a abril de 2001. Os desempenhos fora da amostra de diversas especificações da MVS foram confrontados com aqueles de outros métodos de estimação. A MVS foi o método que melhor controlou o trade- off entre viés e variância dos erros. / [en] This thesis investigates interest rates curve estimation under non-parametric approach. The text is divided into two parts. The first one focus on which criterion to use to select the best performance method in the task of interpolating Brazilian interest rate curve. A selection criterion is proposed to measure out-of-sample performance by combining resample strategies leave-k-out cross validation applied upon the whole sample curves, where K k £ £ 1 and K is function of observed contract number in each curve. Some particularities reduce substantially the required computational effort, making the proposed criterion feasible. The data sample range is daily, from January 1997 to February 2001. The proposed criterion selected natural cubic spline, used as data perfect-fitting estimation method. Considering the trade rate precision, the spline is non-biased. However, quantitative analysis of performance determinant factors showed the existence of out-of-sample error heteroskedasticities. From a conditional variance specification of these errors, a security interval scheme is proposed for interest rate generated by perfect-fitting natural cubic spline. A backtest showed that the proposed security interval is consistent, accommodating the evolved assumptions and approximations. The second part estimate US free-for-floating interest rate swap contract curve by using Support Vector Machine (SVM), a method derived from Statistical Learning Theory. The SVM research has got important theoretical results, however the number of implementation on real regression problems is low. SVM has some attractive characteristics for interest rates curves modeling: it has the ability to introduce already in its estimation process a priori information about curve shape and about liquidity and price formation aspects of the contracts that generate the curve. The last information set is quantified by the bid-ask spread. The basic SVM formulation is changed in order to be able to incorporate the different values for bid-ask spreads, without losing its properties. Great attention is given to the question of how to extract a priori information from swap curve typical shape to be used in MVS parameter selection. The data sample range is daily, from March 1997 to April 2001. The out-of-sample performances of different SVM specifications are faced with others method performances. SVM got the better control of trade- off between bias and variance of out-of-sample errors.

This thesis deals with a comparison of classification methods. At first, these classification methods based on machine learning are described, then a classifier comparison system is designed and implemented. This thesis also describes some classification tasks and datasets on which the designed system will be tested. The evaluation of classification tasks is done according to standard metrics. In this thesis is presented design and implementation of a classifier that is based on the principle of evolutionary algorithms.

As all genres of popular music, classical music consists of many different subgenres. The aim of this work is to recognize those subgenres from orchestral recordings. It is focused on the time period from the very end of 16th century to the beginning of 20th century, which means that Baroque era, Classical era and Romantic era are researched. The Music Information Retrieval (MIR) method was used to classify chosen subgenres. In the first phase of MIR method, parameters were extracted from musical recordings and were evaluated. Only the best parameters were used as input data for machine learning classifiers, to be specific: kNN (K-Nearest Neighbor), LDA (Linear Discriminant Analysis), GMM (Gaussian Mixture Models) and SVM (Support Vector Machines). In the final chapter, all the best results are summarized. According to the results, there is significant difference between the Baroque era and the other researched eras. This significant difference led to better identification of the Baroque era recordings. On the contrary, Classical era ended up to be relatively similar to Romantic era and therefore all classifiers had less success in identification of recordings from this era. The results are in line with music theory and characteristics of chosen musical eras.

This work deals with the diagnosis of Parkinson's disease by analyzing the speech signal. At the beginning of this work there is described speech signal production. The following is a description of the speech signal analysis, its preparation and subsequent feature extraction. Next there is described Parkinson's disease and change of the speech signal by this disability. The following describes the symptoms, which are used for the diagnosis of Parkinson's disease (FCR, VSA, VOT, etc.). Another part of the work deals with the selection and reduction symptoms using the learning algorithms (SVM, ANN, k-NN) and their subsequent evaluation. In the last part of the thesis is described a program to count symptoms. Further is described selection and the end evaluated all the result.

Training a Support Vector Machine (SVM) requires the solution of a very large quadratic programming (QP) optimization problem. Sequential Minimal Optimization (SMO) is a decomposition-based algorithm which breaks this large QP problem into a series of smallest possible QP problems. However, it still costs O(n2) computation time. In our SVM implementation, we can do training with huge data sets in a distributed manner (by breaking the dataset into chunks, then using Message Passing Interface (MPI) to distribute each chunk to a different machine and processing SVM training within each chunk). In addition, we moved the kernel calculation part in SVM classification to a graphics processing unit (GPU) which has zero scheduling overhead to create concurrent threads. In this thesis, we will take advantage of this GPU architecture to improve the classification performance of SVM.

<p>Synthesized speech may be used for nefarious purposes, such as fraud, spoofing, and misinformation campaigns. We present several speech forensics methods based on deep learning to protect against such attacks. First, we use a convolutional neural network (CNN) and transformers to detect synthesized speech. Then, we investigate closed set and open set speech synthesizer attribution. We use a transformer to attribute a speech signal to its source (i.e., to identify the speech synthesizer that created it). Additionally, we show that our approach separates different known and unknown speech synthesizers in its latent space, even though it has not seen any of the unknown speech synthesizers during training. Next, we explore machine learning for an objective in the aerospace domain.</p> <p><br></p> <p>Compared to conventional ballistic vehicles and cruise vehicles, hypersonic glide vehicles (HGVs) exhibit unprecedented abilities. They travel faster than Mach 5 and maneuver to evade defense systems and hinder prediction of their final destinations. We investigate machine learning for identifying different HGVs and a conic reentry vehicle (CRV) based on their aerodynamic state estimates. We also propose a HGV flight phase prediction method. Inspired by natural language processing (NLP), we model flight phases as “words” and HGV trajectories as “sentences.” Next, we learn a “grammar” from the HGV trajectories that describes their flight phase transition patterns. Given “words” from the initial part of a HGV trajectory and the “grammar”, we predict future “words” in the “sentence” (i.e., future HGV flight phases in the trajectory). We demonstrate that this approach successfully predicts future flight phases for HGV trajectories, especially in scenarios with limited training data. We also show that it can be used in a transfer learning scenario to predict flight phases of HGV trajectories that exhibit new maneuvers and behaviors never seen before during training.</p>

No description available.