Model Reference Adaptive Control (MRAC) is a widely studied adaptive control methodology that aims to ensure that a nonlinear plant with significant modeling uncertainty behaves like a chosen reference model. MRAC methods attempt to achieve this by representing the modeling uncertainty as a weighted combination of known nonlinear functions, and using a weight update law that ensures weights take on values such that the effect of the uncertainty is mitigated. If the adaptive weights do arrive at an ideal value that best represent the uncertainty, significant performance and robustness gains can be realized. However, most MRAC adaptive laws use only instantaneous data for adaptation and can only guarantee that the weights arrive at these ideal values if and only if the plant states are Persistently Exciting (PE). The condition on PE reference input is restrictive and often infeasible to implement or monitor online. Consequently, parameter convergence cannot be guaranteed in practice for many adaptive control applications. Hence it is often observed that traditional adaptive controllers do not exhibit long-term-learning and global uncertainty parametrization. That is, they exhibit little performance gain even when the system tracks a repeated command. This thesis presents a novel approach to adaptive control that relies on using current and recorded data concurrently for adaptation. The thesis shows that for a concurrent learning adaptive controller, a verifiable condition on the linear independence of the recorded data is sufficient to guarantee that weights arrive at their ideal values even when the system states are not PE. The thesis also shows that the same condition can guarantee exponential tracking error and weight error convergence to zero, thereby allowing the adaptive controller to recover the desired transient response and robustness properties of the chosen reference models and to exhibit long-term-learning. This condition is found to be less restrictive and easier to verify online than the condition on persistently exciting exogenous input required by traditional adaptive laws that use only instantaneous data for adaptation. The concept is explored for several adaptive control architectures, including neuro-adaptive flight control, where a neural network is used as the adaptive element. The performance gains are justified theoretically using Lyapunov based arguments, and demonstrated experimentally through flight-testing on Unmanned Aerial Systems.

In this thesis we provide a unifying framework for two decades of work in an area of Machine Learning known as cost-sensitive Boosting algorithms. This area is concerned with the fact that most real-world prediction problems are asymmetric, in the sense that different types of errors incur different costs. Adaptive Boosting (AdaBoost) is one of the most well-studied and utilised algorithms in the field of Machine Learning, with a rich theoretical depth as well as practical uptake across numerous industries. However, its inability to handle asymmetric tasks has been the subject of much criticism. As a result, numerous cost-sensitive modifications of the original algorithm have been proposed. Each of these has its own motivations, and its own claims to superiority. With a thorough analysis of the literature 1997-2016, we find 15 distinct cost-sensitive Boosting variants - discounting minor variations. We critique the literature using {\em four} powerful theoretical frameworks: Bayesian decision theory, the functional gradient descent view, margin theory, and probabilistic modelling. From each framework, we derive a set of properties which must be obeyed by boosting algorithms. We find that only 3 of the published Adaboost variants are consistent with the rules of all the frameworks - and even they require their outputs to be calibrated to achieve this. Experiments on 18 datasets, across 21 degrees of cost asymmetry, all support the hypothesis - showing that once calibrated, the three variants perform equivalently, outperforming all others. Our final recommendation - based on theoretical soundness, simplicity, flexibility and performance - is to use the original Adaboost algorithm albeit with a shifted decision threshold and calibrated probability estimates. The conclusion is that novel cost-sensitive boosting algorithms are unnecessary if proper calibration is applied to the original.

Este trabalho apresenta o desenvolvimento de algoritmos de adaptação do padrão de marcha com a utilização de redes neurais artificiais para uma órtese ativa para membros inferiores. Trajetórias estáveis são geradas durante o processo de otimização, considerando um gerador de trajetórias baseado no critério do ZMP (Zero Moment Point) e no modelo dinâmico do equipamento. Três redes neurais são usadas para diminuir o tempo de cálculo do modelo e da otimização do ZMP, e reproduzir o gerador de trajetórias analítico. A primeira rede aproxima a dinâmica do modelo fornecendo a variação de torque necessária para a realização do processo de otimização dos parâmetros de adaptação da marcha; a segunda rede trabalha no processo de otimização, fornecendo o parâmetro otimizado de acordo com a interação paciente-órtese; a terceira rede reproduz o gerador de trajetórias para um determinado intervalo de tempo do passo que pode ser repetido para qualquer quantidade de passos. Além disso, um controle do tipo torque calculado acrescido de um controle PD é usado para garantir que as trajetórias atuais estejam seguindo as trajetórias desejadas da órtese. O modelo dinâmico da órtese na sua configuração atual, com forças de interação incluídas, é usado para gerar resultados simulados. / This work deals with neural network-based gait-pattern adaptation algorithms for an active lower limbs orthosis. Stable trajectories are generated during the optimization process, considering a trajectory generator based on the Zero Moment Point criterion and on the dynamic model. Additionally, three neural network are used to decrease the time-consuming computation of the model and ZMP optimization and to reproduce the analitical trajectory generator. The first neural network approximates the dynamic model providing the necessary torque variation to gait adaptation parameters process; the second network works in the optimization procedure, giving the adapting parameter according to orthosis-patient interaction; and the third network replaces the trajectory generation for a stablished step time interval which can be reproduced any time during the walking. Also, a computed torque controller plus the PD controller is designed to guarantee the actual trajectories are following the orthosis desired trajectories. The dynamic model of the actual active orthosis, with interaction forces included, is used to generate simulation results.

Submitted by Renata Lopes (renatasil82@gmail.com) on 2017-10-23T17:05:10Z No. of bitstreams: 1 renanmottagoulart.pdf: 506891 bytes, checksum: 01d7b3b82d2bc0af0d295fc75de17b91 (MD5) / Approved for entry into archive by Adriana Oliveira (adriana.oliveira@ufjf.edu.br) on 2017-11-09T13:52:19Z (GMT) No. of bitstreams: 1 renanmottagoulart.pdf: 506891 bytes, checksum: 01d7b3b82d2bc0af0d295fc75de17b91 (MD5) / Made available in DSpace on 2017-11-09T13:52:19Z (GMT). No. of bitstreams: 1 renanmottagoulart.pdf: 506891 bytes, checksum: 01d7b3b82d2bc0af0d295fc75de17b91 (MD5) Previous issue date: 2017-09-01 / CAPES - Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Jogos de repetição é um ramo de Teoria dos Jogos, em que um jogo é jogado repetidas vezes pelos jogadores. Neste cenário, assume-se que os jogadores nem sempre jogam de modo ótimo ou podem estar dispostos, se possível, a colaborar. Neste contexto é possível um jogador analisar o comportamento dos oponentes para encontrar padrões. Estes padrões podem ser usados para aumentar o lucro obtido pelo jogador ou detectar se o oponente está disposto a realizar uma colaboração mutualmente benéfica. Nesta dissertação é proposto um novo algoritmo baseado em kernel de similaridade capaz de prever as ações de jogadores em jogos de repetição. A predição não se limita a ação do próximo round, podendo prever as ações de uma sequência finita de rounds consecutivos. O algoritmo consegue se adaptar rapidamente caso os outros jogadores mudem suas estratégias durante o jogo. É mostrado empiricamente que o algoritmo proposto obtém resultados superiores ao estado da arte atual. / Repeated games is a branch of game theory, where a game can be played several times by the players involved. In this setting, it is assumed that the players do not always play the optimal strategy or that they may be willing to collaborate. In this context it is possible for a player to analyze the opponent’s behaviour to find patters. These patterns can be used to maximize the player’s profit or to detect if the opponent is willing to collaborate. On this dissertation it is proposed a new algorithm based on similarity kernel capable of predicting the opponent’s actions on repeated games. The prediction is not limited to the next round’s action, being able to predict actions on a finite sequence of rounds. It is able to adapt rapidly if the opponents change their strategies during the course of a game. It is shown empirically that the proposed algorithm achieves better results than the current state of the art.

The goal of this project was to create a web-based, crowd-sourced, correlational database, that easily allowed users to submit objects and receive correlated objects as results. The webservice was created in the web development languages of HTML, CSS, PHP and Javscript, with MySQL to handle the database. Simultaneous development was kept in check with the aid of the source code management system GIT. Upon completion, the service contained several HTML-views, the ability to add and rate objects, a per-object dedicated page with information parsed from Wikipedia.org, and a view with objects ranked in accordance to the preferences specific to the current user. Roughly a month after the beginning of development, the website was publicly launched and promoted in order to collect data, and improvements were added to the website as needed. Two weeks after the public launch, the collected data was measured and analyzed. The algorithm proved effective and scalable, especially with the introduction of tags and simultaneous computation of object features.

This thesis deals with convolutional neural networks. It is a kind of deep neural networks that are presently widely used mainly for image recognition and natural language processing. The thesis describes specifics of convolutional neural networks in comparison with traditional neural networks and is focused on inner computations in the process of learning. Convolutional neural networks typically consist of a different types of layers of neurons and the core part of this thesis is to demonstrate computations of individual types of layers. Learning demonstrating program of a simple convolutional network was designed and implemented using own implementation of neural network. Validity of the implementation was tested by training models for solving a classification task. Experiments with different types of architectures were conducted and their performance was compared.

The Convolutional Neural Network (CNN) is a mathematical model designed to distill input information into a more useful representation. This distillation process removes information over time through a series of dimensionality reductions, which ultimately, grant the model the ability to resist noise, and generalize effectively. However, CNNs often contain elements that are ineffective at contributing towards useful representations. This Thesis aims at providing a remedy for this problem by introducing Absolute Cosine Value Regularization (ACVR). This is a regularization technique hypothesized to increase the representational power of CNNs by using a Gradient Descent Orthogonalization algorithm to force the vectors that constitute their filters at any given convolutional layer to occupy unique positions in Rⁿ. This method should in theory, lead to a more effective balance between information loss and representational power, ultimately, increasing network performance. The following Thesis proposes and examines the mathematics and intuition behind ACVR, and goes on to propose Dynamic-ACVR (D-ACVR). This Thesis also proposes and examines the effects of ACVR on the filters of a low-dimensional CNN, as well as the effects of ACVR and D-ACVR on traditional Convolutional filters in VGG-19. Finally, this Thesis proposes and examines regularization of the Pointwise filters in MobileNetv1.

Remote Electrical Tilt optimization is an effective method to obtain the optimal Key Performance Indicators (KPIs) by remotely controlling the base station antenna’s vertical tilt. To improve the KPIs aims to improve antennas’ cooperation effect since KPIs measure the quality of cooperation between the antenna to be optimized and its neighbor antennas. Reinforcement Learning (RL) is an appropriate method to learn an antenna tilt control policy since the agent in RL can generate the optimal epsilon greedy tilt optimization policy by observing the environment and learning from the state- action pairs. However, existing models only produced tilt modification strategies by interpreting the to- be- optimized antenna’s features, which cannot fully characterize the mobile cellular network formed by the to- be- optimized antenna and its neighbors. Therefore, incorporating the features of the neighboring antennas into the model is an important measure to improve the optimization strategy. This work will introduce the Graph Attention Network to model the neighborhood antenna’s impact on the antenna to be optimized through the attention mechanism. Furthermore, it will generate a low- dimensional embedding vector with more expressive power to represent the to- be- optimized antenna’s state in the RL framework through dealing with graph- structural data. This new model, namely Graph Attention Q- Network (GAQ), is a model based on DQN and aims to acquire a higher performance than the Deep Q- Network (DQN) model, which is the baseline, evaluated by the same metric — KPI Improvement. Since GAQ has a richer perception of the environment than the vanilla DQN model, it thereby outperforms the DQN model, obtaining fourteen percent performance improvement compared to the baseline. Besides, GAQ also performs 14 per cent better than DQN in terms of convergence efficiency. / Optimering av fjärrlutning är en effektiv metod för att nå optimala nyckeltal genom fjärrstyrning av den vertikala lutningen av en antenn i en basstation. Att förbättra nyckeltalen innebär att förbättra sammarbetseffekten mellan antenner eftersom nyckeltalen är mått på kvalitén av sammarbetet mellan den antenn som optimeras och dess angränsande antenner. Förstärkande Inlärning (FI) är en lämplig metod för att lära sig en optimal strategi för reglering av antennlutningen eftersom agenten inom FI kan generera den optimala epsilongiriga optimeringsstrategin genom att observera miljön och lära sig från par av tillstånd och aktioner. Nuvarande modeller genererar dock endast lutningsstrategier genom att tolka egenskaperna hos den antenn som ska optimeras, vilket inte är tillräckligt för att karatärisera mobilnätverket bestående av antennen som ska optimeras samt dess angränsande antenner. Därav är inkluderingen av de angränsande antennernas egenskaper i modellen viktig för att förbättra optimeringsstrategin. Detta arbete introducerar Graf- Uppmärksammat Nätverk för att modellera de angränsande antennernas påverkan på den antenn som ska optimeras genom uppmärksamhetsmekanismen. Metoden genererar en lågdimensionell vektor med större förmåga att representera den optimerade antennens tillstånd i FI modellen genom att hantera data i struktur av en graf. Den nya modellen, Graf- Uppmärksammat Q- Nätverk (GUQ), är en modell baserad på DQN med mål att nå bättre prestanda än en standard DQN- modell, utvärderat efter samma mätvärde –– förbättring av nyckeltalen. Eftersom GUQ har en större upfattning av miljön så överträffar metoden DQN- modellen genom en fjorton procent bättre prestandaökning. Dessutom, så överträffar GUQ även DQN i form av snabbare konvergens.

In this thesis, an optimal investment problem is studied for an investor who can only invest in a financial market modelled by an Itô-Lévy process; with one risk free (bond) and one risky (stock) investment possibility. We present the dynamic programming method and the associated Hamilton-Jacobi-Bellman (HJB) equation to explicitly solve this problem. It is shown that with purification and simplification to the standard jump diffusion process, closed form solutions for the optimal investment strategy and for the value function are attainable. It is also shown that, an explicit solution can be obtained via a finite training of a neural network using Stochastic gradient descent (SGD) for a specific case.

Lithium-ion (Li-ion) batteries have become a crucial factor in the recent electro-mobility trend. People's increased interest in electric vehicles (EVs) has motivated several automotive manufacturers and research organizations to develop suitable drivetrain designs involving batteries. Especially the development of the 48V Li-ion battery has been of great importance to reduce CO2 emissions and meet emission standards. However, accurately modeling Li-ion batteries is a difficult task since multiple factors have to be considered. Conservative Methods are using pyhsico-chemical models or electrical circuits in order to mimic the battery behavior. This thesis deals with developing a Li-ion battery model using artificial neural network (ANN) algorithms to predict the state of charge (SOC) as one of the key battery management system states. Due to the rising power of GPUs and the amount of available data, ANNs became popular in recent years. ANNs are also applicable to different areas of battery technology. Using battery data like the battery voltage, temperature, and current as input features, a neural network is trained that predicts battery SOC. A novel approach based on ANNs and one of the most commonly used SOC estimation methods are presented in this thesis to model the battery behavior. Furthermore, an approach for dealing with the highly unbalanced data by creating multidimensional bins and compare different neural network architectures for time series forecasting is introduced. By creating the model, our main priority is to reduce the model's errors in extreme operating areas of the battery. According to our results, long short-term memory (LSTM) architectures appear to be the best fit for this task. Finally, the developed ANN model can successfully learn battery behavior, however the model's accuracy under harsh operating circumstances is highly dependent on the data quality gathered.