Motion sensing input devices like Microsoft's Kinect offer an alternative to traditional computer input devices like keyboards and mouses. Daily new applications using this interface appear. Most of them implement their own gesture detection. In our demonstration we show a new approach using the data stream engine Andu IN. The gesture detection is done based on Andu IN's complex event processing functionality. This way we build a system that allows to define new and complex gestures on the basis of a declarative programming interface. On this basis our demonstration data 3 provides a basic natural interaction OLAP interface for a sample star schema database using Microsoft's Kinect.

We present a novel approach in machine learning by combining naı̈ve Bayes classifiers with tree kernels. Tree kernel methods produce promising results in machine learning tasks containing treestructured attribute values. These kernel methods are used to compare two tree-structured attribute values recursively. Up to now tree kernels are only used in kernel machines like Support Vector Machines or Perceptrons. In this paper, we show that tree kernels can be utilized in a naı̈ve Bayes classifier enabling the classifier to handle tree-structured values. We evaluate our approach on three datasets containing tree-structured values. We show that our approach using tree-structures delivers significantly better results in contrast to approaches using non-structured (flat) features extracted from the tree. Additionally, we show that our approach is significantly faster than comparable kernel machines in several settings which makes it more useful in resource-aware settings like mobile devices. Naı̈ve Bayes Classifier; Tree Kernel; Lazy Learning; Tree-structured Values

Eine der größten Herausforderung bei der Anwendungsentwicklung für drahtlose Sensornetzwerke ist der effiziente Umgang mit den oftmals stark beschränkten Ressourcen. Die Entwicklung neuer Systeme gestaltet sich entsprechend aufwendig und teuer. In der vorliegenden Arbeit soll mit AnduIN ein Ansatz gezeigt werden, welcher die Entwicklung entsprechender Lösungen dahingehend vereinfacht, dass lediglich die Zielstellung beschrieben werden muss. Das System analysiert diese und entscheidet selbstständig, welche Funktionen in welcher Form zu realisieren sind.

As a special type of wireless P2P systems, sensor networks are often deployed for detecting events caused by disasters. The peer-to-peer mode of the sensor system itself gets challenged either directly by damages of the disaster or by unreliable wireless links. This work explores possible failure models and compares the performance of several gossip protocols corresponding to the failures models. With further refinement of the gossip protocols, the performance in the failure modes caused by disasters is improved. The evaluation of our simulation results shows that using refined gossip protocols in correspondence to the failure models, the information aggregation and dissemination speed, communication cost and the accuracy of the aggregated data can be improved.

Wireless sensor networks are powerful, distributed, self-organizing systems used for event and environmental monitoring. In-network query processors like TinyDB offer a user friendly SQL-like application development. Due to the sensor nodes’ resource limitations, monolithic approaches often support only a restricted number of operators. For this reason, complex processing is typically outsourced to the base station. Nevertheless, previous work has shown that complete or partial in-network processing can be more efficient than the base station approach. In this paper, we introduce AnduIN , a system for developing, deploying, and running complex in-network processing tasks. In particular, we present the query planning and execution strategies used in AnduIN , a system combining sensor-local in-network processing and a data stream engine. Query planning employs a multi-dimensional cost model taking energy consumption into account and decides autonomously which query parts will be processed within the sensor network and which parts will be processed at the central instance.

In drahtlosen Sensornetzen ist die effiziente und moglichst sensorlokale Analyse der Daten ein wichtiger Ansatz zur Verlangerung der Batterielaufzeit der Sensoren. Neben den klassischen primitiven Analyseverfahren, wie zum Beispiel Filtern oder Aggregation, werden zunehmend auch komplexe Verfahren teilweise oder vollstandig innerhalb der Grenzen der Sensornetze verarbeitet. Die vorliegende Arbeit widmet sich einem dieser Verfahren – der Erkennung von haufig auftretenden Mustern (Frequent Itemsets) uber Attributen mit kontinuierlichen Werten. Schwerpunkt ist die teilweise Auslagerung des vorgestellten Verfahrens auf die Knoten des Sensornetzes.

Das Erfassen der Bedeutung von geschriebener oder gesprochener Sprache ist bis heute eine der größten Herausforderung in der Informatik. Für eine effiziente computergestützte Analyse und Suche ist dies aber unumgänglich. Gegenwärtig ist es nicht möglich Informationen in ausreichender Qualität vollständig automatisch auf ihre Bedeutung hin zu analysieren und entsprechend zu annotieren. Häufig wird daher auf teilautomatische Systeme zurückgegriffen, welche eine Nutzerinteraktion erfordern. In der folgenden Arbeit wollen wir zwei neue Ansätze zur automatischen Annotation von Dokumenten präsentieren.

Das Aufdecken unbekannter Zusammenhange zahlt zu einer der wichtigsten Aufgaben im Data Mining. Fur das Problem des Frequent Pattern Mining uber statischen Daten finden sich daher in der Literatur eine Vielzahl an Losungen. Die Integration von Sensorik in nahezu jeden Lebensbereich fuhrt allerdings zu Datenmengen, welche mittels der klassischen Verfahren zumeist nicht mehr bewaltigt werden konnen. Ein Paradigmenwechsel hin zur DatenstromVerarbeitung ist oftmals unumganglich. Ein interessantes Problem, welches im Zusammenhang mit der Verarbeitung von Sensordaten auftritt ist der prinzipiell stetige Wertebereich von Messungen. Die bekannten Losungen sind fur die Analyse von kontinuierlichen Daten uber stetigen Wertebereichen nur bedingt geeignet. Im folgenden soll mit dem FP -Stream ein entsprechendes Verfahren fur die Analyse quantitativer haufiger Muster uber Datenstromen prasentiert werden.

Sensor networks have evolved to a powerful infrastructure component for event monitoring in many application scenarios. In addition to simple filter and aggregation operations, an important task in processing sensor data is data mining - the identification of relevant information and patterns. Limited capabilities of sensor nodes in terms of storage and processing capacity, battery lifetime, and communication demand a power-efficient, preferably sensor-local processing. In this paper, we present AnduIN, a system for developing, deploying, and running in-network data mining tasks. The system consists of a data stream processing engine, a library of operators for sensor-local processing, a box-and-arrow editor for specifying data mining tasks and deployment, a GUI providing the user with current information about the network and running queries, and an alerter notifying the user if a better query execution plan is available. At the demonstration site, we plan to show our system in action using burst detection as example application.

In the past couple of years, sensor networks have evolved into an important infrastructure component for monitoring and tracking events and phenomena in several, often mission critical application domains. An important task in processing streams of data generated by these networks is the detection of anomalies, e.g., outliers or bursts, and in particular the computation of the location and spatial extent of such anomalies in a sensor network. Such information is then used as an important input to decision making processes.In this paper, we present a novel approach that facilitates the efficient computation of such anomaly regions from individual sensor readings. We propose an algorithm to derive regions with a spatial extent from individual (anomalous) sensor readings, with a particular focus on obstacles present in the sensor network and the influence of such obstacles on anomaly regions. We then improve this approach by describing a distributed in-network processing technique where the region detection is performed at sensor nodes and thus leads to important energy savings. We demonstrate the advantages of this strategy over a traditional, centralized processing strategy by employing a cost model for real sensors and sensor networks.

We consider the problem of efficient data aggregation and processing within wireless sensor networks (WSN). The system we propose allows monitoring and sensor data processing by injecting bytecode into the network. Additionally we present AnduIN, a system whose goal it is to decide which parts of a given query should be processed within the sensor network.

Detecting bursts in data streams is an important and challenging task. Due to the complexity of this task, usually burst detection cannot be formulated using standard query operators. Therefore, we show how to integrate burst detection for stationary as well as non-stationary data into query formulation and processing, from the language level to the operator level. Afterwards, we present fundamentals of threshold-based burst detection. We focus on the applicability of time series forecasting techniques in order to dynamically identify suitable thresholds for stream data containing arbitrary trends and periods. The proposed approach is evaluated with respect to quality and performance on synthetic and real-world sensor data using a full-fledged DSMS.

Wireless sensor networks have become important architectures for many application scenarios, e.g., traffic monitoring or environmental monitoring in general. As these sensors are battery-powered, query processing strategies aim at minimizing energy consumption. Because sending all sensor readings to a central stream data management system consumes too much energy, parts of the query can already be processed within the network (in-network query processing). An important optimization criterion in this context is where to process which intermediate results and how to route them efficiently. To overcome these problems, we propose AnduIN, a system addressing these problems and offering an optimizer that decides which parts of the query should be processed within the sensor network. It also considers optimization with respect to complex data analysis tasks, such as burst detection. Furthermore, AnduIN offers a Web-based frontend for declarative query formulation and deployment. In this paper, we present our research prototype and focus on AnduIN's components alleviating deployment and usability.

Materializing results from complex aggregation queries helps to significantly improve response times in OLAP servers. This problem is known as the view selection problem: choosing the optimal set of aggregation tables (called configuration) for a given workload. In this paper we present an online approach for adjusting the configuration dynamically to the current workload. This approach is implemented as part of an open source OLAP server and acts on the level of multidimensional MDX queries. The work presents the details of cost estimation and optimization of the system demonstrated in [10] and extends it by an online tuning strategy.

OLAP servers based on relational backends typically exploit materialized aggregate tables to improve response times of complex analytical queries. One of the key problems in this context is the view selection problem: choosing the optimal set of aggregation tables (called configuration) for a given workload. In this paper, we present a system that continuously monitors the workload and raises a quantified alert, when a better configuration is available. We address the tasks of query monitoring and view selection at the OLAP level instead of the SQL level, which simplifies the containment checks as well as rewriting and in this way helps to reduce the complexity of the backend system. At the demo we plan to show how our system works, i.e., how the system reacts upon arbitrary (interactive) workloads and how the user is alerted that a better configuration is available.

Data replication is a central technique to increase availability and performance of distributed systems. While offering many advantages it also requires more effort for ensuring data consistency in case of updates. In the research literature various approaches for replication management in distributed databases have been presented, but they are mostly limited either in scalability or in the consistency guarantees they provide. On the other hand, P2P systems usually provide replication support but ignore the update problem. In this paper we present a new approach for managing replicated data in wide area distributed networks. Our solution is orthogonal to the underlying infrastructure and managed in a decentralized manner. It guarantees single-master consistency and allows updates at any node of the system by combining traditional replication techniques with ideas known from P2P systems.

Detecting bursts in data streams is an important and challenging task, especially in stock market, traffic control or sensor network streams. Burst detection means the identification of non regular behavior within data streams. A specifically crucial challenge on burst detection is to identify bursts in the case of non-stationary data. One approach is to apply thresholds to discover such bursts. In this paper, we propose a new approach to dynamically identify suitable thresholds using techniques known from time series forecasting. We present fundamentals and discuss requirements for threshold-based burst detection on stream data containing arbitrary trends and periods.

Modern buildings are equipped with high-tech systems that take care of several fundamental aspects, e.g., air-conditioning, heating and water supply. The requirements posed on facility management by such buildings are challenging. Modern techniques implement adaptive control systems to achieve this, in which decisions are preferably based on the results of (multiple correlated) mining tasks on recently gathered sensor data. In this work, we discuss the general relationship between such control systems and the underlying mining tasks. We exemplary choose change detection in the context of pattern analysis as a representative, because this mining task involves general requirements known from stream processing like the need for incremental algorithms, but also poses specific challenges like in-time detection. We present three concrete approaches for this and an according evaluation.