Este trabalho propõe uma nova abordagem conexionista, chamada de IGMN (do inglês Incremental Gaussian Mixture Network), para aproximação incremental de funções e tarefas de tempo real. Ela é inspirada em recentes teorias do cérebro, especialmente o MPF (do inglês Memory-Prediction Framework) e a Inteligência Artificial Construtivista, que fazem com que o modelo proposto possua características especiais que não estão presentes na maioria dos modelos de redes neurais existentes. Além disso, IGMN é baseado em sólidos princípios estatísticos (modelos de mistura gaussianos) e assintoticamente converge para a superfície de regressão ótima a medida que os dados de treinamento chegam. As principais vantagens do IGMN em relação a outros modelos de redes neurais são: (i) IGMN aprende instantaneamente analisando cada padrão de treinamento apenas uma vez (cada dado pode ser imediatamente utilizado e descartado); (ii) o modelo proposto produz estimativas razoáveis baseado em poucos dados de treinamento; (iii) IGMN aprende de forma contínua e perpétua a medida que novos dados de treinamento chegam (não existem fases separadas de treinamento e utilização); (iv) o modelo proposto resolve o dilema da estabilidade-plasticidade e não sofre de interferência catastrófica; (v) a topologia da rede neural é definida automaticamente e de forma incremental (novas unidades são adicionadas sempre que necessário); (vi) IGMN não é sensível às condições de inicialização (de fato IGMN não utiliza nenhuma decisão e/ou inicialização aleatória); (vii) a mesma rede neural IGMN pode ser utilizada em problemas diretos e inversos (o fluxo de informações é bidirecional) mesmo em regiões onde a função alvo tem múltiplas soluções; e (viii) IGMN fornece o nível de confiança de suas estimativas. Outra contribuição relevante desta tese é o uso do IGMN em importantes tarefas nas áreas de robótica e aprendizado de máquina, como por exemplo a identificação de modelos, a formação incremental de conceitos, o aprendizado por reforço, o mapeamento robótico e previsão de séries temporais. De fato, o poder de representação e a eficiência e do modelo proposto permitem expandir o conjunto de tarefas nas quais as redes neurais podem ser utilizadas, abrindo assim novas direções nos quais importantes contribuições do estado da arte podem ser feitas. Através de diversos experimentos, realizados utilizando o modelo proposto, é demonstrado que o IGMN é bastante robusto ao problema de overfitting, não requer um ajuste fino dos parâmetros de configuração e possui uma boa performance computacional que permite o seu uso em aplicações de controle em tempo real. Portanto pode-se afirmar que o IGMN é uma ferramenta de aprendizado de máquina bastante útil em tarefas de aprendizado incremental de funções e predição em tempo real. / This work proposes IGMN (standing for Incremental Gaussian Mixture Network), a new connectionist approach for incremental function approximation and real time tasks. It is inspired on recent theories about the brain, specially the Memory-Prediction Framework and the Constructivist Artificial Intelligence, which endows it with some unique features that are not present in most ANN models such as MLP, RBF and GRNN. Moreover, IGMN is based on strong statistical principles (Gaussian mixture models) and asymptotically converges to the optimal regression surface as more training data arrive. The main advantages of IGMN over other ANN models are: (i) IGMN learns incrementally using a single scan over the training data (each training pattern can be immediately used and discarded); (ii) it can produce reasonable estimates based on few training data; (iii) the learning process can proceed perpetually as new training data arrive (there is no separate phases for leaning and recalling); (iv) IGMN can handle the stability-plasticity dilemma and does not suffer from catastrophic interference; (v) the neural network topology is defined automatically and incrementally (new units added whenever is necessary); (vi) IGMN is not sensible to initialization conditions (in fact there is no random initialization/ decision in IGMN); (vii) the same neural network can be used to solve both forward and inverse problems (the information flow is bidirectional) even in regions where the target data are multi-valued; and (viii) IGMN can provide the confidence levels of its estimates. Another relevant contribution of this thesis is the use of IGMN in some important state-of-the-art machine learning and robotic tasks such as model identification, incremental concept formation, reinforcement learning, robotic mapping and time series prediction. In fact, the efficiency of IGMN and its representational power expand the set of potential tasks in which the neural networks can be applied, thus opening new research directions in which important contributions can be made. Through several experiments using the proposed model it is demonstrated that IGMN is also robust to overfitting, does not require fine-tunning of its configuration parameters and has a very good computational performance, thus allowing its use in real time control applications. Therefore, IGMN is a very useful machine learning tool for incremental function approximation and on-line prediction.

Este trabalho propõe uma nova abordagem conexionista, chamada de IGMN (do inglês Incremental Gaussian Mixture Network), para aproximação incremental de funções e tarefas de tempo real. Ela é inspirada em recentes teorias do cérebro, especialmente o MPF (do inglês Memory-Prediction Framework) e a Inteligência Artificial Construtivista, que fazem com que o modelo proposto possua características especiais que não estão presentes na maioria dos modelos de redes neurais existentes. Além disso, IGMN é baseado em sólidos princípios estatísticos (modelos de mistura gaussianos) e assintoticamente converge para a superfície de regressão ótima a medida que os dados de treinamento chegam. As principais vantagens do IGMN em relação a outros modelos de redes neurais são: (i) IGMN aprende instantaneamente analisando cada padrão de treinamento apenas uma vez (cada dado pode ser imediatamente utilizado e descartado); (ii) o modelo proposto produz estimativas razoáveis baseado em poucos dados de treinamento; (iii) IGMN aprende de forma contínua e perpétua a medida que novos dados de treinamento chegam (não existem fases separadas de treinamento e utilização); (iv) o modelo proposto resolve o dilema da estabilidade-plasticidade e não sofre de interferência catastrófica; (v) a topologia da rede neural é definida automaticamente e de forma incremental (novas unidades são adicionadas sempre que necessário); (vi) IGMN não é sensível às condições de inicialização (de fato IGMN não utiliza nenhuma decisão e/ou inicialização aleatória); (vii) a mesma rede neural IGMN pode ser utilizada em problemas diretos e inversos (o fluxo de informações é bidirecional) mesmo em regiões onde a função alvo tem múltiplas soluções; e (viii) IGMN fornece o nível de confiança de suas estimativas. Outra contribuição relevante desta tese é o uso do IGMN em importantes tarefas nas áreas de robótica e aprendizado de máquina, como por exemplo a identificação de modelos, a formação incremental de conceitos, o aprendizado por reforço, o mapeamento robótico e previsão de séries temporais. De fato, o poder de representação e a eficiência e do modelo proposto permitem expandir o conjunto de tarefas nas quais as redes neurais podem ser utilizadas, abrindo assim novas direções nos quais importantes contribuições do estado da arte podem ser feitas. Através de diversos experimentos, realizados utilizando o modelo proposto, é demonstrado que o IGMN é bastante robusto ao problema de overfitting, não requer um ajuste fino dos parâmetros de configuração e possui uma boa performance computacional que permite o seu uso em aplicações de controle em tempo real. Portanto pode-se afirmar que o IGMN é uma ferramenta de aprendizado de máquina bastante útil em tarefas de aprendizado incremental de funções e predição em tempo real. / This work proposes IGMN (standing for Incremental Gaussian Mixture Network), a new connectionist approach for incremental function approximation and real time tasks. It is inspired on recent theories about the brain, specially the Memory-Prediction Framework and the Constructivist Artificial Intelligence, which endows it with some unique features that are not present in most ANN models such as MLP, RBF and GRNN. Moreover, IGMN is based on strong statistical principles (Gaussian mixture models) and asymptotically converges to the optimal regression surface as more training data arrive. The main advantages of IGMN over other ANN models are: (i) IGMN learns incrementally using a single scan over the training data (each training pattern can be immediately used and discarded); (ii) it can produce reasonable estimates based on few training data; (iii) the learning process can proceed perpetually as new training data arrive (there is no separate phases for leaning and recalling); (iv) IGMN can handle the stability-plasticity dilemma and does not suffer from catastrophic interference; (v) the neural network topology is defined automatically and incrementally (new units added whenever is necessary); (vi) IGMN is not sensible to initialization conditions (in fact there is no random initialization/ decision in IGMN); (vii) the same neural network can be used to solve both forward and inverse problems (the information flow is bidirectional) even in regions where the target data are multi-valued; and (viii) IGMN can provide the confidence levels of its estimates. Another relevant contribution of this thesis is the use of IGMN in some important state-of-the-art machine learning and robotic tasks such as model identification, incremental concept formation, reinforcement learning, robotic mapping and time series prediction. In fact, the efficiency of IGMN and its representational power expand the set of potential tasks in which the neural networks can be applied, thus opening new research directions in which important contributions can be made. Through several experiments using the proposed model it is demonstrated that IGMN is also robust to overfitting, does not require fine-tunning of its configuration parameters and has a very good computational performance, thus allowing its use in real time control applications. Therefore, IGMN is a very useful machine learning tool for incremental function approximation and on-line prediction.

Métodos baseados em grafos consistem em uma poderosa forma de representação e abstração de dados que proporcionam, dentre outras vantagens, representar relações topológicas, visualizar estruturas, representar grupos de dados com formatos distintos, bem como, fornecer medidas alternativas para caracterizar os dados. Esse tipo de abordagem tem sido cada vez mais considerada para solucionar problemas de aprendizado de máquina, principalmente no aprendizado não supervisionado, como agrupamento de dados, e mais recentemente, no aprendizado semissupervisionado. No aprendizado supervisionado, por outro lado, o uso de algoritmos baseados em grafos ainda tem sido pouco explorado na literatura. Este trabalho apresenta um algoritmo não paramétrico baseado em grafos para problemas de classificação com distribuição estacionária, bem como sua extensão para problemas que apresentam distribuição não estacionária. O algoritmo desenvolvido baseia-se em dois conceitos, a saber, 1) em uma estrutura chamada grafo K-associado ótimo, que representa o conjunto de treinamento como um grafo esparso e dividido em componentes; e 2) na medida de pureza de cada componente, que utiliza a estrutura do grafo para determinar o nível de mistura local dos dados em relação às suas classes. O trabalho também considera problemas de classificação que apresentam alteração na distribuição de novos dados. Este problema caracteriza a mudança de conceito e degrada o desempenho do classificador. De modo que, para manter bom desempenho, é necessário que o classificador continue aprendendo durante a fase de aplicação, por exemplo, por meio de aprendizado incremental. Resultados experimentais sugerem que ambas as abordagens apresentam vantagens na classificação de dados em relação aos algoritmos testados / Graph-based methods consist in a powerful form for data representation and abstraction which provides, among others advantages, representing topological relations, visualizing structures, representing groups of data with distinct formats, as well as, supplying alternative measures to characterize data. Such approach has been each time more considered to solve machine learning related problems, mainly concerning unsupervised learning, like clustering, and recently, semi-supervised learning. However, graph-based solutions for supervised learning tasks still remain underexplored in literature. This work presents a non-parametric graph-based algorithm suitable for classification problems with stationary distribution, as well as its extension to cope with problems of non-stationary distributed data. The developed algorithm relies on the following concepts, 1) a graph structure called optimal K-associated graph, which represents the training set as a sparse graph separated into components; and 2) the purity measure for each component, which uses the graph structure to determine local data mixture level in relation to their classes. This work also considers classification problems that exhibit modification on distribution of data flow. This problem qualifies concept drift and worsens any static classifier performance. Hence, in order to maintain accuracy performance, it is necessary for the classifier to keep learning during application phase, for example, by implementing incremental learning. Experimental results, concerning both algorithms, suggest that they had presented advantages over the tested algorithms on data classification tasks

Nous introduisons la configuration de la "Classification Incrémentale Implicitement Raffinée / Incremental Implicitly-Refined Classification (IIRC)", une extension de la configuration de l'apprentissage incrémental des classes où les lots de classes entrants possèdent deux niveaux de granularité, c'est-à-dire que chaque échantillon peut avoir une étiquette (label) de haut niveau (brute), comme "ours”, et une étiquette de bas niveau (plus fine), comme "ours polaire". Une seule étiquette (label) est fournie à la fois, et le modèle doit trouver l’autre étiquette s’il l’a déjà apprise. Cette configuration est plus conforme aux scénarios de la vie réelle, où un apprenant aura tendance à interagir avec la même famille d’entités plusieurs fois, découvrant ainsi encore plus de granularité à leur sujet, tout en essayant de ne pas oublier les connaissances acquises précédemment. De plus, cette configuration permet d’évaluer les modèles pour certains défis importants liés à l’apprentissage tout au long de la vie (lifelong learning) qui ne peuvent pas être facilement abordés dans les configurations existantes. Ces défis peuvent être motivés par l’exemple suivant: “si un modèle a été entraîné sur la classe ours dans une tâche et sur ours polaire dans une autre tâche; oubliera-t-il le concept d’ours, déduira-t-il à juste titre qu’un ours polaire est également un ours ? et associera-t-il à tort l’étiquette d’ours polaire à d’autres races d’ours ?” Nous développons un benchmark qui permet d’évaluer les modèles sur la configuration de l’IIRC. Nous évaluons plusieurs algorithmes d’apprentissage ”tout au long de la vie” (lifelong learning) de l’état de l’art. Par exemple, les méthodes basées sur la distillation sont relativement performantes mais ont tendance à prédire de manière incorrecte un trop grand nombre d’étiquettes par image. Nous espérons que la configuration proposée, ainsi que le benchmark, fourniront un cadre de problème significatif aux praticiens. / We introduce the "Incremental Implicitly-Refined Classification (IIRC)" setup, an extension to the class incremental learning setup where the incoming batches of classes have two granularity levels. i.e., each sample could have a high-level (coarse) label like "bear" and a low-level (fine) label like "polar bear". Only one label is provided at a time, and the model has to figure out the other label if it has already learned it. This setup is more aligned with real-life scenarios, where a learner usually interacts with the same family of entities multiple times, discovers more granularity about them, while still trying not to forget previous knowledge. Moreover, this setup enables evaluating models for some important lifelong learning challenges that cannot be easily addressed under the existing setups. These challenges can be motivated by the example "if a model was trained on the class bear in one task and on polar bear in another task, will it forget the concept of bear, will it rightfully infer that a polar bear is still a bear? and will it wrongfully associate the label of polar bear to other breeds of bear?". We develop a standardized benchmark that enables evaluating models on the IIRC setup. We evaluate several state-of-the-art lifelong learning algorithms and highlight their strengths and limitations. For example, distillation-based methods perform relatively well but are prone to incorrectly predicting too many labels per image. We hope that the proposed setup, along with the benchmark, would provide a meaningful problem setting to the practitioners.

Le travail présenté dans cette étude se place dans le contexte de l’analyse automatique des vidéos. A cause du nombre croissant des données vidéo, il est souvent difficile, voire impossible qu’un ou plusieurs opérateurs puissent les regarder toutes. Une demande récurrente est d’identifier les moments dans la vidéo quand il y a quelque chose d’inhabituel qui se passe, c’est-à-dire la détection des événements anormaux.Nous proposons donc plusieurs algorithmes permettant d’identifier des événements inhabituels, en faisant l’hypothèse que ces événements ont une faible probabilité. Nous abordons plusieurs types d’événements, de l’analyse des zones en mouvement à l’analyse des trajectoires des objets suivis.Après avoir dédié une partie de la thèse à la construction d’un système de suivi,nous proposons plusieurs mesures de similarité entre des trajectoires. Ces mesures, basées sur DTW (Dynamic Time Warping), estiment la similarité des trajectoires prenant en compte différents aspects : spatial, mais aussi temporel, pour pouvoir - par exemple - faire la différence entre des trajectoires qui ne sont pas parcourues de la même façon (en termes de vitesse de déplacement). Ensuite, nous construisons des modèles de trajectoires, permettant de représenter les comportements habituels des objets pour pouvoir ensuite détecter ceux qui s’éloignent de la normale.Pour pallier les défauts de suivi qui apparaissent dans la pratique, nous analysons les vecteurs de flot optique et nous construisons une carte de mouvement. Cette carte modélise sous la forme d’un codebook les directions privilégiées qui apparaissent pour chaque pixel, permettant ainsi d’identifier tout déplacement anormal, sans avoir pour autant la notion d’objet suivi. En utilisant la cohérence temporelle, nous pouvons améliorer encore plus le taux de détection, affecté par les erreurs d’estimation de flot optique. Dans un deuxième temps, nous changeons la méthode de constructions de cette carte de mouvements, pour pouvoir extraire des caractéristiques de plus haut niveau — l’équivalent des trajectoires, mais toujours sans nécessiter le suivi des objets. Nous pouvons ainsi réutiliser partiellement l’analyse des trajectoires pour détecter des événements rares.Tous les aspects présentés dans cette thèse ont été implémentés et nous avons construit certaines applications, comme la prédiction des déplacements des objets ou la mémorisation et la recherche des objets suivis. / The growing number of video data makes often difficult, even impossible, any attemptof watching them entirely. In the context of automatic analysis of videos, a recurring request is to identify moments in the video when something unusual happens.We propose several algorithms to identify unusual events, making the hypothesis that these events have a low probability. We address several types of events, from those generates by moving areas to the trajectories of objects tracked. In the first part of the study, we build a simple tracking system. We propose several measures of similarity between trajectories. These measures give an estimate of the similarity of trajectories by taking into account both spatial and/or temporal aspects. It is possible to differentiate between objects moving on the same path, but with different speeds. Based on these measures, we build models of trajectories representing the common behavior of objects, so that we can identify those that are abnormal.We noticed that the tracking yields bad results, especially in crowd situations. Therefore, we use the optical flow vectors to build a movement model based on a codebook. This model stores the preferred movement directions for each pixel. It is possible to identify abnormal movement at pixel-level, without having to use a tracker. By using temporal coherence, we can further improve the detection rate, affected by errors of estimation of optic flow. In a second step, we change the method of construction of this model. With the new approach, we can extract higher-level features — the equivalent trajectories, but still without the notion of object tracking. In this situation, we can reuse partial trajectory analysis to detect rare events.All aspects presented in this study have been implemented. In addition, we have design some applications, like predicting the trajectories of visible objects or storing and retrieving tracked objects in a database.

<p dir="ltr">Unintentional electromagnetic emission (called emanation) from electronic devices and cables contains a significant correlation with the source signal and can be used to recover otherwise confidential data. In our work, EM emanation has been exploited to recover keystrokes from USB keyboards. Also, such emission has been utilized to form a covert channel for data exfiltration from air-gapped devices without being detected by IDS. To protect sensitive information, an automated emanation detection system has been proposed by developing two emanation detection algorithms (CNN-based and harmonic-based) through the characterization of emanation signals from a wide range of devices. Apart from emanation, data theft can happen due to the failure of access control methods. Traditional wireless devices are susceptible to various spoofing attacks as they only use digital signature-based authentication systems, ignoring the physical signatures completely. To circumvent that, RF-PUF was proposed to use device-specific signatures to be used for trust augmentation in traditional methods. By forming an extensive experimental dataset, we established RF-PUF as a strong PUF with a low-power overhead that outperformed the state-of-the-art methods and is robust against typical attacks. For real-time authentication, we proposed DIRAC, which forms dynamic device clusters and incrementally learns as more device data becomes available. Since our root of trust is in the physical signature of the ICs, they also need to be secured. However, counterfeited ICs may jeopardize that goal. We have proposed RF-PSF, which uses device-specific physical properties to authenticate its process technology which is a big part of the cloned IC detection.</p>

Self-adaptive systems have limited time to adjust their configurations whenever their adaptation goals, i.e., quality requirements, are violated due to some runtime uncertainties. Within the available time, they need to analyze their adaptation space, i.e., a set of configurations, to find the best adaptation option, i.e., configuration, that can achieve their adaptation goals. Existing formal analysis approaches find the best adaptation option by analyzing the entire adaptation space. However, exhaustive analysis requires time and resources and is therefore only efficient when the adaptation space is small. The size of the adaptation space is often in hundreds or thousands, which makes formal analysis approaches inefficient in large-scale self-adaptive systems. In this thesis, we tackle this problem by presenting an online learning approach that enables formal analysis approaches to analyze large adaptation spaces efficiently. The approach integrates with the standard feedback loop and reduces the adaptation space to a subset of adaptation options that are relevant to the current runtime uncertainties. The subset is then analyzed by the formal analysis approaches, which allows them to complete the analysis faster and efficiently within the available time. We evaluate our approach on two different instances of an Internet of Things application. The evaluation shows that our approach dramatically reduces the adaptation space and analysis time without compromising the adaptation goals.

Es presenta un esquema únic per a la síntesi i anàlisi audiovisual personalitzable realista de seqüències audiovisuals de cares parlants i seqüències visuals de llengua de signes en àmbit domèstic. En el primer cas, amb animació totalment sincronitzada a través d'una font de text o veu; en el segon, utilitzant la tècnica de lletrejar paraules mitjançant la ma. Les seves possibilitats de personalització faciliten la creació de seqüències audiovisuals per part d'usuaris no experts. Les aplicacions possibles d'aquest esquema de síntesis comprenen des de la creació de personatges virtuals realistes per interacció natural o vídeo jocs fins vídeo conferència des de molt baix ample de banda i telefonia visual per a les persones amb problemes d'oïda, passant per oferir ajuda a la pronunciació i la comunicació a aquest mateix col·lectiu. El sistema permet processar seqüències llargues amb un consum de recursos molt reduït, sobre tot, en el referent a l'emmagatzematge, gràcies al desenvolupament d'un nou procediment de càlcul incremental per a la descomposició en valors singulars amb actualització de la informació mitja. Aquest procediment es complementa amb altres tres: el decremental, el de partició i el de composició. / Se presenta un esquema único para la síntesis y análisis audiovisual personalizable realista de secuencias audiovisuales de caras parlantes y secuencias visuales de lengua de signos en entorno doméstico. En el primer caso, con animación totalmente sincronizada a través de una fuente de texto o voz; en el segundo, utilizando la técnica de deletreo de palabras mediante la mano. Sus posibilidades de personalización facilitan la creación de secuencias audiovisuales por parte de usuarios no expertos. Las aplicaciones posibles de este esquema de síntesis comprenden desde la creación de personajes virtuales realistas para interacción natural o vídeo juegos hasta vídeo conferencia de muy bajo ancho de banda y telefonía visual para las personas con problemas de oído, pasando por ofrecer ayuda en la pronunciación y la comunicación a este mismo colectivo. El sistema permite procesar secuencias largas con un consumo de recursos muy reducido gracias al desarrollo de un nuevo procedimiento de cálculo incremental para la descomposición en valores singulares con actualización de la información media. / A shared framework for realistic and personalizable audiovisual synthesis and analysis of audiovisual sequences of talking heads and visual sequences of sign language is presented in a domestic environment. The former has full synchronized animation using a text or auditory source of information; the latter consists in finger spelling. Their personalization capabilities ease the creation of audiovisual sequences by non expert users. The applications range from realistic virtual avatars for natural interaction or videogames to low bandwidth videoconference and visual telephony for the hard of hearing, including help to speech therapists. Long sequences can be processed with reduced resources, specially storing ones. This is allowed thanks to the proposed scheme for the incremental singular value decomposition with mean preservation. This scheme is complemented with another three: the decremental, the split and the composed ones.