Neste trabalho a detecção de novidade é tratada como o problema de identificação de conceitos emergentes em dados que podem ser apresentados em um fluxo contínuo. Considerando a relação intrínseca entre tempo e novidade e os desafios impostos por fluxos de dados, uma nova abordagem é proposta. OLINDDA (OnLIne Novelty and Drift Detection Algorithm) vai além da classficação com uma classe e concentra-se no aprendizado contínuo não-supervisionado de novos conceitos. Tendo aprendido uma descrição inicial de um conceito normal, prossegue à análise de novos dados, tratando-os como um fluxo contínuo em que novos conceitos podem aparecer a qualquer momento. Com o uso de técnicas de agrupamento, OLINDDA pode empregar diversos critérios de validação para avaliar grupos em termos de sua coesão e representatividade. Grupos considerados válidos produzem conceitos que podem sofrer fusão, e cujo conhecimento é continuamente incorporado. A técnica é avaliada experimentalmente com dados artificiais e reais. O módulo de classificação com uma classe é comparado a outras técnicas de detecção de novidade, e a abordagem como um todo é analisada sob vários aspectos por meio da evolução temporal de diversas métricas. Os resultados reforçam a importância da detecção contínua de novos conceitos, assim como as dificuldades e desafios do aprendizado não-supervisionado de novos conceitos em fluxos de dados / In this work novelty detection is treated as the problem of identifying emerging concepts in data that may be presented in a continuous ow. Considering the intrinsic relationship between time and novelty and the challenges imposed by data streams, a novel approach is proposed. OLINDDA, an OnLIne Novelty and Drift Detection Algorithm, goes beyond one-class classification and focuses on the unsupervised continuous learning of novel concepts. Having learned an initial description of a normal concept, it proceeds to the analysis of new data, treating them as a continuous ow where novel concepts may appear at any time. By the use of clustering techniques, OLINDDA may employ several validation criteria to evaluate clusters in terms of their cohesiveness and representativeness. Clusters considered valid produce concepts that may be merged, and whose knowledge is continuously incorporated. The technique is experimentally evaluated with artificial and real data. The one-class classification module is compared to other novelty detection techniques, and the whole approach is analyzed from various aspects through the temporal evolution of several metrics. Results reinforce the importance of continuous detection of novel concepts, as well as the dificulties and challenges of the unsupervised learning of novel concepts in data streams

Condition monitoring is very important for critical assets such as gearboxes used in the power and mining industries. Fluctuating operating conditions are inevitable for wind turbines and mining machines such as bucket wheel excavators and draglines due to the continuous uctuating wind speeds and variations in ground properties, respectively. Many of the classical condition monitoring techniques have proven to be ine ective under uctuating operating conditions and therefore more sophisticated techniques have to be developed. However, many of the signal processing tools that are appropriate for uctuating operating conditions can be di cult to interpret, with the presence of incipient damage easily being overlooked. In this study, a cost-e ective diagnostic methodology is developed, using machine learning techniques, to diagnose the condition of the machine in the presence of uctuating operating conditions when only an acceleration signal, generated from a gearbox during normal operation, is available. The measured vibration signal is order tracked to preserve the angle-cyclostationary properties of the data. A robust tacholess order tracking methodology is proposed in this study using probabilistic approaches. The measured vibration signal is order tracked with the tacholess order tracking method (as opposed to computed order tracking), since this reduces the implementation and the running cost of the diagnostic methodology. Machine condition features, which are sensitive to changes in machine condition, are extracted from the order tracked vibration signal and processed. The machine condition features can be sensitive to operating condition changes as well. This makes it difficult to ascertain whether the changes in the machine condition features are due to changes in machine condition (i.e. a developing fault) or changes in operating conditions. This necessitates incorporating operating condition information into the diagnostic methodology to ensure that the inferred condition of the machine is not adversely a ected by the uctuating operating conditions. The operating conditions are not measured and therefore representative features are extracted and modelled with a hidden Markov model. The operating condition machine learning model aims to infer the operating condition state that was present during data acquisition from the operating condition features at each angle increment. The operating condition state information is used to optimise robust machine condition machine learning models, in the form of hidden Markov models. The information from the operating condition and machine condition models are combined using a probabilistic approach to generate a discrepancy signal. This discrepancy signal represents the deviation of the current features from the expected behaviour of the features of a gearbox in a healthy condition. A second synchronous averaging process, an automatic alarm threshold for fault detection, a gear-pinion discrepancy distribution and a healthy-damaged decomposition of the discrepancy signal are proposed to provide an intuitive and robust representation of the condition of the gearbox under uctuating operating conditions. This allows fault detection, localisation as well as trending to be performed on a gearbox during uctuating operation conditions. The proposed tacholess order tracking method is validated on seven datasets and the fault diagnostic methodology is validated on experimental as well as numerical data. Very promising results are obtained by the proposed tacholess order tracking method and by the diagnostic methodology. / Dissertation (MEng)--University of Pretoria, 2016. / Mechanical and Aeronautical Engineering / MEng / Unrestricted

Network usage is an important quality metric for mobile apps. Slow networks, low monthly traffic quotas and high roaming fees restrict mobile users’ amount of usable Internet traffic. Companies wanting their apps to stay competitive must be aware of their network usage and changes to it. Short feedback loops for the impact of code changes are key in agile software development. To notify stakeholders of changes when they happen without being prohibitively expensive in terms of manpower the change detection must be fully automated. To further decrease the manpower overhead cost of implementing network usage change detection the system need to have low configuration requirements, and keep the false positive rate low while managing to detect larger changes. This thesis proposes an automated change detection method for network activity to quickly notify stakeholders with relevant information to begin a root cause analysis after a change in the network activity is introduced. With measurements of the Spotify’s iOS app we show that the tool achieves a low rate of false positives while detecting relevant changes in the network activity even for apps with dynamic network usage patterns as Spotify. / Nätverksaktivitet är ett viktigt kvalitetsmått för mobilappar. Mobilanvändare begränsas ofta av långsamma nätverk, låg månatlig trafikkvot och höga roamingavgifter. Företag som vill ha konkurrenskraftiga appar behöver vara medveten om deras nätverksaktivitet och förändringar av den. Snabb återkoppling för effekten av kodändringar är vitalt för agil programutveckling. För att underrätta intressenter om ändringar när de händer utan att vara avskräckande dyrt med avseende på arbetskraft måste ändringsdetekteringen vara fullständigt automatiserad. För att ytterligare minska arbetskostnaderna för ändringsdetektering av nätverksaktivitet måste detekteringssystemet vara snabbt att konfigurera, hålla en låg grad av felaktig detektering samtidigt som den lyckas identifiera stora ändringar. Den här uppsatsen föreslår ett automatiserat förändringsdetekteringsverktyg för nätverksaktivitet för att snabbt meddela stakeholders med relevant information för påbörjan av grundorsaksanalys när en ändring som påverkar nätverksaktiviteten introduceras. Med hjälp av mätningar på Spotifys iOS-app visar vi att verktyget når en låg grad av felaktiga detekteringar medan den identifierar ändringar i nätverksaktiviteten även för appar med så dynamisk nätverksanvändning som Spotify.

No description available.

Knowing when one does not know is crucial in decision making. By estimating uncertainties humans can recognize novelty both by intuition and reason, but most AI systems lack this self-reflective ability. In anomaly detection, a common approach is to train a model to learn the distinction between some notion of normal and some notion of anomalies. In contrast, we let the models build their own notion of normal by learning directly from the data in a self-supervised manner, and by introducing estimations of model uncertainty the models can recognize themselves when novel situations are encountered. In our work, the aim is to investigate the relationship between model uncertainty and anomalies in time series data. We develop a method based on a recurrent autoencoder approach, and we design an anomaly score function that aggregates model error with model uncertainty to indicate anomalies. Use the Monte Carlo Dropout as Bayesian approximation to derive model uncertainty. Asa proof of concept we evaluate our method qualitatively on real-world complex time series using stock market data. Results show that our method can identify extreme events in the stock market. We conclude that the relation between model uncertainty and anomalies can be utilized for anomaly detection in time series data.

Cette thèse traite de l’apprentissage automatique pour la classification de données. Afin de réduire le coût de l’étiquetage, l’apprentissage actif permet de formuler des requêtes pour demander à un opérateur d’étiqueter seulement quelques données choisies selon un critère d’importance. Nous proposons une nouvelle mesure d’incertitude qui permet de caractériser l’importance des données et qui améliore les performances de l’apprentissage actif par rapport aux mesures existantes. Cette mesure détermine le plus petit poids nécessaire à associer à une nouvelle donnée pour que le classifieur change sa prédiction concernant cette donnée. Nous intégrons ensuite le fait que les données à traiter arrivent en continu dans un flux de longueur infinie. Nous proposons alors un seuil d’incertitude adaptatif qui convient pour un apprentissage actif à partir d’un flux de données et qui réalise un compromis entre le nombre d’erreurs de classification et le nombre d’étiquettes de classes demandées. Les méthodes existantes d’apprentissage actif à partir de flux de données, sont initialisées avec quelques données étiquetées qui couvrent toutes les classes possibles. Cependant, dans de nombreuses applications, la nature évolutive du flux fait que de nouvelles classes peuvent apparaître à tout moment. Nous proposons une méthode efficace de détection active de nouvelles classes dans un flux de données multi-classes. Cette méthode détermine de façon incrémentale une zone couverte par les classes connues, et détecte les données qui sont extérieures à cette zone et proches entre elles, comme étant de nouvelles classes. Enfin, il est souvent difficile d’obtenir un étiquetage totalement fiable car l’opérateur humain est sujet à des erreurs d’étiquetage qui réduisent les performances du classifieur appris. Cette problématique a été résolue par l’introduction d’une mesure qui reflète le degré de désaccord entre la classe donnée manuellement et la classe prédite et une nouvelle mesure d’"informativité" permettant d’exprimer la nécessité pour une donnée mal étiquetée d’être réétiquetée par un opérateur alternatif / This thesis focuses on machine learning for data classification. To reduce the labelling cost, active learning allows to query the class label of only some important instances from a human labeller.We propose a new uncertainty measure that characterizes the importance of data and improves the performance of active learning compared to the existing uncertainty measures. This measure determines the smallest instance weight to associate with new data, so that the classifier changes its prediction concerning this data. We then consider a setting where the data arrives continuously from an infinite length stream. We propose an adaptive uncertainty threshold that is suitable for active learning in the streaming setting and achieves a compromise between the number of classification errors and the number of required labels. The existing stream-based active learning methods are initialized with some labelled instances that cover all possible classes. However, in many applications, the evolving nature of the stream implies that new classes can appear at any time. We propose an effective method of active detection of novel classes in a multi-class data stream. This method incrementally maintains a feature space area which is covered by the known classes, and detects those instances that are self-similar and external to that area as novel classes. Finally, it is often difficult to get a completely reliable labelling because the human labeller is subject to labelling errors that reduce the performance of the learned classifier. This problem was solved by introducing a measure that reflects the degree of disagreement between the manually given class and the predicted class, and a new informativeness measure that expresses the necessity for a mislabelled instance to be re-labeled by an alternative labeller

A novelty detection algorithm inspired by human audio pattern recognition is conceptualized and experimentally tested. This anomaly detection technique can be used to monitor the health of a machine or could also be coupled with a current state of the art system to enhance its fault detection capabilities. Time-domain data obtained from a microphone is processed by applying a short-time FFT, which returns time-frequency patterns. Such patterns are fed to a machine learning algorithm, which is designed to detect novel signals and identify windows in the frequency domain where such novelties occur. The algorithm presented in this paper uses one-dimensional kernel density estimation for different frequency bins. This process eliminates the need for data dimension reduction algorithms. The method of "pseudo-likelihood cross validation" is used to find an independent optimal kernel bandwidth for each frequency bin. Metrics such as the "Individual Node Relative Difference" and "Total Novelty Score" are presented in this work, and used to assess the degree of novelty of a new signal. Experimental datasets containing synthetic and real novelties are used to illustrate and test the novelty detection algorithm. Novelties are successfully detected in all experiments. The presented novelty detection technique could greatly enhance the performance of current state-of-the art condition monitoring systems, or could also be used as a stand-alone system.

Many mobile robot applications require robots to act safely and intelligently in complex unfamiliarenvironments with little structure and limited or unavailable human supervision. As arobot is forced to operate in an environment that it was not engineered or trained for, various aspectsof its performance will inevitably degrade. Roboticists equip robots with powerful sensorsand data sources to deal with uncertainty, only to discover that the robots are able to make onlyminimal use of this data and still find themselves in trouble. Similarly, roboticists develop andtrain their robots in representative areas, only to discover that they encounter new situations thatare not in their experience base. Small problems resulting in mildly sub-optimal performance areoften tolerable, but major failures resulting in vehicle loss or compromised human safety are not.This thesis presents a series of online algorithms to enable a mobile robot to better deal withuncertainty in unfamiliar domains in order to improve its navigational abilities, better utilizeavailable data and resources and reduce risk to the vehicle. We validate these algorithms throughextensive testing onboard large mobile robot systems and argue how such approaches can increasethe reliability and robustness of mobile robots, bringing them closer to the capabilitiesrequired for many real-world applications.

O presente trabalho investiga um método de detecção de anomalias baseado em sistemas imunológicos artificiais, especificamente em uma técnica de reconhecimento próprio/não-próprio chamada algoritmo de seleção negativa (NSA). Foi utilizado um esquema de representação baseado em hiperesferas com centros e raios variáveis e um modelo capaz de gerar detectores, com esta representação, de forma eficiente. Tal modelo utiliza algoritmos genéticos onde cada gene do cromossomo contém um índice para um ponto de uma distribuição quasi-aleatória que servirá como centro do detector e uma função decodificadora responsável por determinar os raios apropriados. A aptidão do cromossomo é dada por uma estimativa do volume coberto através uma integral de Monte Carlo. Este algoritmo teve seu desempenho verificado em diferentes dimensões e suas limitações levantadas. Com isso, pode-se focar as melhorias no algoritmo, feitas através da implementação de operadores genéticos mais adequados para a representação utilizada, de técnicas de redução do número de pontos do conjunto próprio e de um método de pré-processamento baseado em bitmaps de séries temporais. Avaliações com dados sintéticos e experimentos com dados reais demonstram o bom desempenho do algoritmo proposto e a diminuição do tempo de execução. / This work investigates a novel detection method based on Artificial Immune Systems, specifically on a self/non-self recognition technique called negative selection algorithm (NSA). A representation scheme based on hyperspheres with variable center and radius and a model that is able to generate detectors, based on that representation scheme, have been used. This model employs Genetic Algorithms where each chromosome gene represents an index to a point in a quasi-random distribution, which serves as a detector center, and a decoder function that determines the appropriate radius. The chromosome fitness is given by an estimation of the covered volume, which is calculated through a Monte Carlo integral. This algorithm had its performance evaluated for different dimensions, and more suitable genetic operators for the used representation, techniques of reducing self-points number and a preprocessing method based on bitmap time series have been therefore implemented. Evaluations with synthetic data and experiments with real data demonstrate the performance of the proposed algorithm and the decrease in execution time.

Trend prediction has become an extremely popular practice in many industrial sectors and academia. It is beneficial for strategic planning and decision making, and facilitates exploring new research directions that are not yet matured. To anticipate future trends in academic environment, a researcher needs to analyze an extensive amount of literature and scientific publications, and gain expertise in the particular research domain. This approach is time-consuming and extremely complicated due to abundance of data and its diversity. Modern machine learning tools, on the other hand, are capable of processing tremendous volumes of data, reaching the real-time human-level performance for various applications. Achieving high performance in unsupervised prediction of emerging trends in text can indicate promising directions for future research and potentially lead to breakthrough discoveries in any field of science. This thesis addresses the problem of emerging trend prediction in text in two main steps: it utilizes HDP topic model to represent latent topic space of a given temporal collection of documents, DBSCAN clustering algorithm to detect groups with high-density regions in the document space potentially leading to emerging trends, and applies KLdivergence in order to capture deviating text which might indicate birth of a new not-yet-seen phenomenon. In order to empirically evaluate the effectiveness of the proposed framework and estimate its predictive capability, both synthetically generated corpora and real-world text collections from arXiv.org, an open-access electronic archive of scientific publications (category: Computer Science), and NIPS publications are used. For synthetic data, a text generator is designed which provides ground truth to evaluate the performance of anomaly detection algorithms. This work contributes to the body of knowledge in the area of emerging trend prediction in several ways. First of all, the method of incorporating topic modeling and anomaly detection algorithms for emerging trend prediction is a novel approach and highlights new perspectives in the subject area. Secondly, the three-level word-document-topic topology of anomalies is formalized in order to detect anomalies in temporal text collections which might lead to emerging trends. Finally, a framework for unsupervised detection of early signals of emerging trends in text is designed. The framework captures new vocabulary, documents with deviating word/topic distribution, and drifts in latent topic space as three main indicators of a novel phenomenon to occur, in accordance with the three-level topology of anomalies. The framework is not limited by particular sources of data and can be applied to any temporal text collections in combination with any online methods for soft clustering.